fix R-devel errors (#4251)

* Initial commit to support multi-node multi-gpu xgboost using dask

* Fixed NCCL initialization by not ignoring the opg parameter.

- it now crashes on NCCL initialization, but at least we're attempting it properly

* At the root node, perform a rabit::Allreduce to get initial sum_gradient across workers

* Synchronizing in a couple of more places.

- now the workers don't go down, but just hang
- no more "wild" values of gradients
- probably needs syncing in more places

* Added another missing max-allreduce operation inside BuildHistLeftRight

* Removed unnecessary collective operations.

* Simplified rabit::Allreduce() sync of gradient sums.

* Removed unnecessary rabit syncs around ncclAllReduce.

- this improves performance _significantly_ (7x faster for overall training,
  20x faster for xgboost proper)

* pulling in latest xgboost

* removing changes to updater_quantile_hist.cc

* changing use_nccl_opg initialization, removing unnecessary if statements

* added definition for opaque ncclUniqueId struct to properly encapsulate GetUniqueId

* placing struct defintion in guard to avoid duplicate code errors

* addressing linting errors

* removing

* removing additional arguments to AllReduer initialization

* removing distributed flag

* making comm init symmetric

* removing distributed flag

* changing ncclCommInit to support multiple modalities

* fix indenting

* updating ncclCommInitRank block with necessary group calls

* fix indenting

* adding print statement, and updating accessor in vector

* improving print statement to end-line

* generalizing nccl_rank construction using rabit

* assume device_ordinals is the same for every node

* test, assume device_ordinals is identical for all nodes

* test, assume device_ordinals is unique for all nodes

* changing names of offset variable to be more descriptive, editing indenting

* wrapping ncclUniqueId GetUniqueId() and aesthetic changes

* adding synchronization, and tests for distributed

* adding  to tests

* fixing broken #endif

* fixing initialization of gpu histograms, correcting errors in tests

* adding to contributors list

* adding distributed tests to jenkins

* fixing bad path in distributed test

* debugging

* adding kubernetes for distributed tests

* adding proper import for OrderedDict

* adding urllib3==1.22 to address ordered_dict import error

* added sleep to allow workers to save their models for comparison

* adding name to GPU contributors under docs

.clang-tidy

@@ -1,4 +1,4 @@
-Checks: 'modernize-*,-modernize-make-*,-modernize-raw-string-literal,google-*,-google-default-arguments,-clang-diagnostic-#pragma-messages,readability-identifier-naming'
+Checks: 'modernize-*,-modernize-make-*,-modernize-use-auto,-modernize-raw-string-literal,google-*,-google-default-arguments,-clang-diagnostic-#pragma-messages,readability-identifier-naming'
 CheckOptions:
   - { key: readability-identifier-naming.ClassCase,           value: CamelCase  }
   - { key: readability-identifier-naming.StructCase,          value: CamelCase  }

.github/lock.yml

@@ -0,0 +1,32 @@
+# Configuration for lock-threads - https://github.com/dessant/lock-threads
+
+# Number of days of inactivity before a closed issue or pull request is locked
+daysUntilLock: 90
+
+# Issues and pull requests with these labels will not be locked. Set to `[]` to disable
+exemptLabels:
+  - feature-request
+
+# Label to add before locking, such as `outdated`. Set to `false` to disable
+lockLabel: false
+
+# Comment to post before locking. Set to `false` to disable
+lockComment: false
+
+# Assign `resolved` as the reason for locking. Set to `false` to disable
+setLockReason: true
+
+# Limit to only `issues` or `pulls`
+# only: issues
+
+# Optionally, specify configuration settings just for `issues` or `pulls`
+# issues:
+#   exemptLabels:
+#     - help-wanted
+#   lockLabel: outdated
+
+# pulls:
+#   daysUntilLock: 30
+
+# Repository to extend settings from
+# _extends: repo

.gitignore

@@ -91,3 +91,4 @@ lib/
 metastore_db
 
 plugin/updater_gpu/test/cpp/data
+/include/xgboost/build_config.h

.travis.yml

@@ -6,9 +6,7 @@ os:
   - linux
   - osx
 
-osx_image: xcode8
-
-group: deprecated-2017Q4
+osx_image: xcode9.3
 
 # Use Build Matrix to do lint and build seperately
 env:
@@ -28,6 +26,8 @@ env:
     - TASK=cpp_test
     # distributed test
     - TASK=distributed_test
+    # address sanitizer test
+    - TASK=sanitizer_test
 
 matrix:
   exclude:
@@ -43,6 +43,8 @@ matrix:
       env: TASK=cpp_test
     - os: osx
       env: TASK=distributed_test
+    - os: osx
+      env: TASK=sanitizer_test
 
 # dependent apt packages
 addons:
@@ -62,6 +64,13 @@ addons:
       - graphviz
       - gcc-4.8
       - g++-4.8
+      - gcc-7
+      - g++-7
+  homebrew:
+    packages:
+      - gcc@7
+      - graphviz
+    update: true
 
 before_install:
   - source dmlc-core/scripts/travis/travis_setup_env.sh

CMakeLists.txt

@@ -8,23 +8,31 @@ set_default_configuration_release()
 msvc_use_static_runtime()
 
 # Options
-option(USE_CUDA  "Build with GPU acceleration")
-option(USE_AVX  "Build with AVX instructions. May not produce identical results due to approximate math." OFF)
-option(USE_NCCL "Build using NCCL for multi-GPU. Also requires USE_CUDA")
-option(JVM_BINDINGS "Build JVM bindings" OFF)
-option(GOOGLE_TEST "Build google tests" OFF)
-option(R_LIB "Build shared library for R package" OFF)
-option(USE_SANITIZER "Use santizer flags" OFF)
+## GPUs
+option(USE_CUDA  "Build with GPU acceleration" OFF)
+option(USE_NCCL  "Build with multiple GPUs support" OFF)
 set(GPU_COMPUTE_VER "" CACHE STRING
   "Space separated list of compute versions to be built against, e.g. '35 61'")
+
+## Bindings
+option(JVM_BINDINGS "Build JVM bindings" OFF)
+option(R_LIB "Build shared library for R package" OFF)
+
+## Devs
+option(USE_SANITIZER "Use santizer flags" OFF)
+option(SANITIZER_PATH "Path to sanitizes.")
 set(ENABLED_SANITIZERS "address" "leak" CACHE STRING
   "Semicolon separated list of sanitizer names. E.g 'address;leak'. Supported sanitizers are
 address, leak and thread.")
+option(GOOGLE_TEST "Build google tests" OFF)
+
+# Plugins
+option(PLUGIN_LZ4 "Build lz4 plugin" OFF)
+option(PLUGIN_DENSE_PARSER "Build dense parser plugin" OFF)
 
 # Deprecation warning
-if(PLUGIN_UPDATER_GPU)
-  set(USE_CUDA ON)
-  message(WARNING "The option 'PLUGIN_UPDATER_GPU' is deprecated. Set 'USE_CUDA' instead.")
+if(USE_AVX)
+  message(WARNING "The option 'USE_AVX' is deprecated as experimental AVX features have been removed from xgboost.")
 endif()
 
 # Compiler flags
@@ -47,22 +55,32 @@ if(WIN32 AND MINGW)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -static-libstdc++")
 endif()
 
+# Check existence of software pre-fetching
+include(CheckCXXSourceCompiles)
+check_cxx_source_compiles("
+#include <xmmintrin.h>
+int main() {
+  char data = 0;
+  const char* address = &data;
+  _mm_prefetch(address, _MM_HINT_NTA);
+  return 0;
+}
+" XGBOOST_MM_PREFETCH_PRESENT)
+check_cxx_source_compiles("
+int main() {
+  char data = 0;
+  const char* address = &data;
+  __builtin_prefetch(address, 0, 0);
+  return 0;
+}
+" XGBOOST_BUILTIN_PREFETCH_PRESENT)
+
 # Sanitizer
 if(USE_SANITIZER)
   include(cmake/Sanitizer.cmake)
   enable_sanitizers("${ENABLED_SANITIZERS}")
 endif(USE_SANITIZER)
 
-# AVX
-if(USE_AVX)
-  if(MSVC)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX")
-  else()
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx")
-  endif()
-  add_definitions(-DXGBOOST_USE_AVX)
-endif()
-
 # dmlc-core
 add_subdirectory(dmlc-core)
 set(LINK_LIBRARIES dmlc rabit)
@@ -83,12 +101,19 @@ if(R_LIB)
   )
 endif()
 
+# Gather source files
 include_directories (
     ${PROJECT_SOURCE_DIR}/include
     ${PROJECT_SOURCE_DIR}/dmlc-core/include
     ${PROJECT_SOURCE_DIR}/rabit/include
 )
 
+# Generate configurable header
+set(CMAKE_LOCAL "${PROJECT_SOURCE_DIR}/cmake")
+set(INCLUDE_ROOT "${PROJECT_SOURCE_DIR}/include")
+message(STATUS "${CMAKE_LOCAL}/build_config.h.in -> ${INCLUDE_ROOT}/xgboost/build_config.h")
+configure_file("${CMAKE_LOCAL}/build_config.h.in" "${INCLUDE_ROOT}/xgboost/build_config.h")
+
 file(GLOB_RECURSE SOURCES
     src/*.cc
     src/*.h
@@ -103,8 +128,17 @@ file(GLOB_RECURSE CUDA_SOURCES
     src/*.cuh
 )
 
+# Add plugins to source files
+if(PLUGIN_LZ4)
+  list(APPEND SOURCES plugin/lz4/sparse_page_lz4_format.cc)
+  link_libraries(lz4)
+endif()
+if(PLUGIN_DENSE_PARSER)
+  list(APPEND SOURCES plugin/dense_parser/dense_libsvm.cc)
+endif()
+
 # rabit
-# TODO: Create rabit cmakelists.txt
+# TODO: Use CMakeLists.txt from rabit.
 set(RABIT_SOURCES
     rabit/src/allreduce_base.cc
     rabit/src/allreduce_robust.cc
@@ -115,6 +149,7 @@ set(RABIT_EMPTY_SOURCES
     rabit/src/engine_empty.cc
     rabit/src/c_api.cc
 )
+
 if(MINGW OR R_LIB)
   # build a dummy rabit library
   add_library(rabit STATIC ${RABIT_EMPTY_SOURCES})
@@ -122,7 +157,11 @@ else()
   add_library(rabit STATIC ${RABIT_SOURCES})
 endif()
 
-if(USE_CUDA)
+if (GENERATE_COMPILATION_DATABASE)
+  set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+endif (GENERATE_COMPILATION_DATABASE)
+
+if(USE_CUDA AND (NOT GENERATE_COMPILATION_DATABASE))
   find_package(CUDA 8.0 REQUIRED)
   cmake_minimum_required(VERSION 3.5)
 
@@ -132,7 +171,7 @@ if(USE_CUDA)
 
   if(USE_NCCL)
     find_package(Nccl REQUIRED)
-    include_directories(${NCCL_INCLUDE_DIR})
+    cuda_include_directories(${NCCL_INCLUDE_DIR})
     add_definitions(-DXGBOOST_USE_NCCL)
   endif()
 
@@ -152,6 +191,39 @@ if(USE_CUDA)
     target_link_libraries(gpuxgboost ${NCCL_LIB_NAME})
   endif()
   list(APPEND LINK_LIBRARIES gpuxgboost)
+
+elseif (USE_CUDA AND GENERATE_COMPILATION_DATABASE)
+  # Enable CUDA language to generate a compilation database.
+  cmake_minimum_required(VERSION 3.8)
+
+  find_package(CUDA 8.0 REQUIRED)
+  enable_language(CUDA)
+  set(CMAKE_CUDA_COMPILER clang++)
+  set(CUDA_SEPARABLE_COMPILATION ON)
+  if (NOT CLANG_CUDA_GENCODE)
+    set(CLANG_CUDA_GENCODE "--cuda-gpu-arch=sm_35")
+  endif (NOT CLANG_CUDA_GENCODE)
+  set(CMAKE_CUDA_FLAGS " -Wno-deprecated  ${CLANG_CUDA_GENCODE} -fPIC ${GENCODE} -std=c++11 -x cuda")
+  message(STATUS "CMAKE_CUDA_FLAGS: ${CMAKE_CUDA_FLAGS}")
+
+  add_library(gpuxgboost STATIC ${CUDA_SOURCES})
+
+  if(USE_NCCL)
+    find_package(Nccl REQUIRED)
+    target_include_directories(gpuxgboost PUBLIC ${NCCL_INCLUDE_DIR})
+    target_compile_definitions(gpuxgboost PUBLIC -DXGBOOST_USE_NCCL)
+    target_link_libraries(gpuxgboost PUBLIC ${NCCL_LIB_NAME})
+  endif()
+
+  target_compile_definitions(gpuxgboost PUBLIC -DXGBOOST_USE_CUDA)
+  # A hack for CMake to make arguments valid for clang++
+  string(REPLACE "-x cu" "-x cuda" CMAKE_CUDA_COMPILE_PTX_COMPILATION
+    ${CMAKE_CUDA_COMPILE_PTX_COMPILATION})
+  string(REPLACE "-x cu" "-x cuda" CMAKE_CUDA_COMPILE_WHOLE_COMPILATION
+    ${CMAKE_CUDA_COMPILE_WHOLE_COMPILATION})
+  string(REPLACE "-x cu" "-x cuda" CMAKE_CUDA_COMPILE_SEPARABLE_COMPILATION
+    ${CMAKE_CUDA_COMPILE_SEPARABLE_COMPILATION})
+  target_include_directories(gpuxgboost PUBLIC cub)
 endif()
 
 
@@ -167,7 +239,6 @@ endif()
 
 add_library(objxgboost OBJECT ${SOURCES})
 
-
 # building shared library for R package
 if(R_LIB)
   find_package(LibR REQUIRED)
@@ -175,22 +246,25 @@ if(R_LIB)
   list(APPEND LINK_LIBRARIES "${LIBR_CORE_LIBRARY}")
   MESSAGE(STATUS "LIBR_CORE_LIBRARY " ${LIBR_CORE_LIBRARY})
 
-  include_directories(
+  # Shared library target for the R package
+  add_library(xgboost SHARED $<TARGET_OBJECTS:objxgboost>)
+  include_directories(xgboost
     "${LIBR_INCLUDE_DIRS}"
     "${PROJECT_SOURCE_DIR}"
   )
 
-  # Shared library target for the R package
-  add_library(xgboost SHARED $<TARGET_OBJECTS:objxgboost>)
   target_link_libraries(xgboost ${LINK_LIBRARIES})
   # R uses no lib prefix in shared library names of its packages
   set_target_properties(xgboost PROPERTIES PREFIX "")
+  if(APPLE)
+    set_target_properties(xgboost PROPERTIES SUFFIX ".so")
+  endif()
 
   setup_rpackage_install_target(xgboost ${CMAKE_CURRENT_BINARY_DIR})
   # use a dummy location for any other remaining installs
   set(CMAKE_INSTALL_PREFIX "${CMAKE_CURRENT_BINARY_DIR}/dummy_inst")
 
-# main targets: shared library & exe
+  # main targets: shared library & exe
 else()
   # Executable
   add_executable(runxgboost $<TARGET_OBJECTS:objxgboost> src/cli_main.cc)
@@ -213,20 +287,20 @@ else()
   add_dependencies(xgboost runxgboost)
 endif()
 
-
 # JVM
 if(JVM_BINDINGS)
     find_package(JNI QUIET REQUIRED)
 
-    include_directories(${JNI_INCLUDE_DIRS} jvm-packages/xgboost4j/src/native)
-
     add_library(xgboost4j SHARED
-        $<TARGET_OBJECTS:objxgboost>
-        jvm-packages/xgboost4j/src/native/xgboost4j.cpp)
-    set_output_directory(xgboost4j ${PROJECT_SOURCE_DIR}/lib)
+      $<TARGET_OBJECTS:objxgboost>
+      jvm-packages/xgboost4j/src/native/xgboost4j.cpp)
+    target_include_directories(xgboost4j
+      PRIVATE ${JNI_INCLUDE_DIRS}
+      PRIVATE jvm-packages/xgboost4j/src/native)
     target_link_libraries(xgboost4j
-        ${LINK_LIBRARIES}
-        ${JAVA_JVM_LIBRARY})
+      ${LINK_LIBRARIES}
+      ${JAVA_JVM_LIBRARY})
+    set_output_directory(xgboost4j ${PROJECT_SOURCE_DIR}/lib)
 endif()
 
 
@@ -237,17 +311,29 @@ if(GOOGLE_TEST)
 
   file(GLOB_RECURSE TEST_SOURCES "tests/cpp/*.cc")
   auto_source_group("${TEST_SOURCES}")
-  include_directories(${GTEST_INCLUDE_DIRS})
 
-  if(USE_CUDA)
+  if(USE_CUDA AND (NOT GENERATE_COMPILATION_DATABASE))
     file(GLOB_RECURSE CUDA_TEST_SOURCES "tests/cpp/*.cu")
+    cuda_include_directories(${GTEST_INCLUDE_DIRS})
     cuda_compile(CUDA_TEST_OBJS ${CUDA_TEST_SOURCES})
+  elseif (USE_CUDA AND GENERATE_COMPILATION_DATABASE)
+    file(GLOB_RECURSE CUDA_TEST_SOURCES "tests/cpp/*.cu")
   else()
     set(CUDA_TEST_OBJS "")
   endif()
 
-  add_executable(testxgboost ${TEST_SOURCES} ${CUDA_TEST_OBJS} $<TARGET_OBJECTS:objxgboost>)
+  if (USE_CUDA AND GENERATE_COMPILATION_DATABASE)
+    add_executable(testxgboost ${TEST_SOURCES} ${CUDA_TEST_SOURCES}
+      $<TARGET_OBJECTS:objxgboost>)
+    target_include_directories(testxgboost PRIVATE cub)
+  else ()
+    add_executable(testxgboost ${TEST_SOURCES} ${CUDA_TEST_OBJS}
+      $<TARGET_OBJECTS:objxgboost>)
+  endif ()
+
   set_output_directory(testxgboost ${PROJECT_SOURCE_DIR})
+  target_include_directories(testxgboost
+    PRIVATE ${GTEST_INCLUDE_DIRS})
   target_link_libraries(testxgboost ${GTEST_LIBRARIES} ${LINK_LIBRARIES})
 
   add_test(TestXGBoost testxgboost)

CONTRIBUTORS.md

@@ -6,21 +6,30 @@ Committers
 ----------
 Committers are people who have made substantial contribution to the project and granted write access to the project.
 * [Tianqi Chen](https://github.com/tqchen), University of Washington
-  - Tianqi is a PhD working on large-scale machine learning, he is the creator of the project.
+  - Tianqi is a Ph.D. student working on large-scale machine learning. He is the creator of the project.
 * [Tong He](https://github.com/hetong007), Amazon AI
-  - Tong is an applied scientist in Amazon AI, he is the maintainer of xgboost R package.
+  - Tong is an applied scientist in Amazon AI. He is the maintainer of XGBoost R package.
 * [Vadim Khotilovich](https://github.com/khotilov)
   - Vadim contributes many improvements in R and core packages.
 * [Bing Xu](https://github.com/antinucleon)
-  - Bing is the original creator of xgboost python package and currently the maintainer of [XGBoost.jl](https://github.com/antinucleon/XGBoost.jl).
+  - Bing is the original creator of XGBoost Python package and currently the maintainer of [XGBoost.jl](https://github.com/antinucleon/XGBoost.jl).
 * [Michael Benesty](https://github.com/pommedeterresautee)
-  - Micheal is a lawyer, data scientist in France, he is the creator of xgboost interactive analysis module in R.
-* [Yuan Tang](https://github.com/terrytangyuan)
-  - Yuan is a data scientist in Chicago, US. He contributed mostly in R and Python packages.
-* [Nan Zhu](https://github.com/CodingCat)
-  - Nan is a software engineer in Microsoft. He contributed mostly in JVM packages.
-* [Sergei Lebedev](https://github.com/superbobry)
-  - Serget is a software engineer in Criteo. He contributed mostly in JVM packages.
+  - Michael is a lawyer and data scientist in France. He is the creator of XGBoost interactive analysis module in R.
+* [Yuan Tang](https://github.com/terrytangyuan), Ant Financial
+  - Yuan is a software engineer in Ant Financial. He contributed mostly in R and Python packages.
+* [Nan Zhu](https://github.com/CodingCat), Uber
+  - Nan is a software engineer in Uber. He contributed mostly in JVM packages.
+* [Sergei Lebedev](https://github.com/superbobry), Criteo
+  - Sergei is a software engineer in Criteo. He contributed mostly in JVM packages.
+* [Hongliang Liu](https://github.com/phunterlau)
+* [Scott Lundberg](http://scottlundberg.com/), University of Washington
+  - Scott is a Ph.D. student at University of Washington. He is the creator of SHAP, a unified approach to explain the output of machine learning models such as decision tree ensembles. He also helps maintain the XGBoost Julia package.
+* [Rory Mitchell](https://github.com/RAMitchell), University of Waikato
+  - Rory is a Ph.D. student at University of Waikato. He is the original creator of the GPU training algorithms. He improved the CMake build system and continuous integration. 
+* [Hyunsu Cho](http://hyunsu-cho.io/), Amazon AI
+  - Hyunsu is an applied scientist in Amazon AI. He is the maintainer of the XGBoost Python package. He also manages the Jenkins continuous integration system (https://xgboost-ci.net/). He is the initial author of the CPU 'hist' updater. 
+* [Jiaming](https://github.com/trivialfis)
+  - Jiaming contributed to the GPU algorithms. He has also introduced new abstractions to improve the quality of the C++ codebase.
 
 Become a Committer
 ------------------
@@ -36,28 +45,25 @@ List of Contributors
 * [Full List of Contributors](https://github.com/dmlc/xgboost/graphs/contributors)
   - To contributors: please add your name to the list when you submit a patch to the project:)
 * [Kailong Chen](https://github.com/kalenhaha)
-  - Kailong is an early contributor of xgboost, he is creator of ranking objectives in xgboost.
+  - Kailong is an early contributor of XGBoost, he is creator of ranking objectives in XGBoost.
 * [Skipper Seabold](https://github.com/jseabold)
-  - Skipper is the major contributor to the scikit-learn module of xgboost.
+  - Skipper is the major contributor to the scikit-learn module of XGBoost.
 * [Zygmunt Zając](https://github.com/zygmuntz)
   - Zygmunt is the master behind the early stopping feature frequently used by kagglers.
 * [Ajinkya Kale](https://github.com/ajkl)
 * [Boliang Chen](https://github.com/cblsjtu)
 * [Yangqing Men](https://github.com/yanqingmen)
-  - Yangqing is the creator of xgboost java package.
+  - Yangqing is the creator of XGBoost java package.
 * [Engpeng Yao](https://github.com/yepyao)
 * [Giulio](https://github.com/giuliohome)
-  - Giulio is the creator of windows project of xgboost
+  - Giulio is the creator of Windows project of XGBoost
 * [Jamie Hall](https://github.com/nerdcha)
-  - Jamie is the initial creator of xgboost sklearn module.
+  - Jamie is the initial creator of XGBoost scikit-learn module.
 * [Yen-Ying Lee](https://github.com/white1033)
 * [Masaaki Horikoshi](https://github.com/sinhrks)
-  - Masaaki is the initial creator of xgboost python plotting module.
-* [Hongliang Liu](https://github.com/phunterlau)
-* [Hyunsu Cho](http://hyunsu-cho.io/)
-  - Hyunsu is the maintainer of the XGBoost Python package. He is in charge of submitting the Python package to Python Package Index (PyPI). He is also the initial author of the CPU 'hist' updater.
+  - Masaaki is the initial creator of XGBoost Python plotting module.
 * [daiyl0320](https://github.com/daiyl0320)
-  - daiyl0320 contributed patch to xgboost distributed version more robust, and scales stably on TB scale datasets.
+  - daiyl0320 contributed patch to XGBoost distributed version more robust, and scales stably on TB scale datasets.
 * [Huayi Zhang](https://github.com/irachex)
 * [Johan Manders](https://github.com/johanmanders)
 * [yoori](https://github.com/yoori)
@@ -68,8 +74,6 @@ List of Contributors
 * [Alex Bain](https://github.com/convexquad)
 * [Baltazar Bieniek](https://github.com/bbieniek)
 * [Adam Pocock](https://github.com/Craigacp)
-* [Rory Mitchell](https://github.com/RAMitchell)
-  - Rory is the author of the GPU plugin and also contributed the cmake build system and windows continuous integration
 * [Gideon Whitehead](https://github.com/gaw89)
 * [Yi-Lin Juang](https://github.com/frankyjuang)
 * [Andrew Hannigan](https://github.com/andrewhannigan)
@@ -78,3 +82,9 @@ List of Contributors
 * [Pierre de Sahb](https://github.com/pdesahb)
 * [liuliang01](https://github.com/liuliang01)
   - liuliang01 added support for the qid column for LibSVM input format. This makes ranking task easier in distributed setting.
+* [Andrew Thia](https://github.com/BlueTea88)
+  - Andrew Thia implemented feature interaction constraints
+* [Wei Tian](https://github.com/weitian)
+* [Chen Qin](https://github.com/chenqin)
+* [Sam Wilkinson](https://samwilkinson.io)
+* [Matthew Jones](https://github.com/mt-jones)

Jenkinsfile

@@ -3,10 +3,18 @@
 // Jenkins pipeline
 // See documents at https://jenkins.io/doc/book/pipeline/jenkinsfile/
 
+import groovy.transform.Field
+
+/* Unrestricted tasks: tasks that do NOT generate artifacts */
+
 // Command to run command inside a docker container
-dockerRun = 'tests/ci_build/ci_build.sh'
+def dockerRun = 'tests/ci_build/ci_build.sh'
+// Utility functions
+@Field
+def utils
 
 def buildMatrix = [
+    [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "9.2", "multiGpu": true],
     [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "9.2" ],
     [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "8.0" ],
     [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": false, "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "8.0" ],
@@ -26,126 +34,94 @@ pipeline {
 
     // Build stages
     stages {
-        stage('Get sources') {
-            agent any
+        stage('Jenkins: Get sources') {
+            agent {
+                label 'unrestricted'
+            }
             steps {
-                checkoutSrcs()
+                script {
+                    utils = load('tests/ci_build/jenkins_tools.Groovy')
+                    utils.checkoutSrcs()
+                }
                 stash name: 'srcs', excludes: '.git/'
                 milestone label: 'Sources ready', ordinal: 1
             }
         }
-        stage('Build doc') {
-            agent any
-            steps {
-                script {
-                    if (env.CHANGE_ID == null) {  // This is a branch
-                        def commit_id = "${GIT_COMMIT}"
-                        def branch_name = "${GIT_LOCAL_BRANCH}"
-                        echo 'Building doc...'
-                        dir ('jvm-packages') {
-                            sh "bash ./build_doc.sh ${commit_id}"
-                            archiveArtifacts artifacts: "${commit_id}.tar.bz2", allowEmptyArchive: true
-                            echo 'Deploying doc...'
-                            withAWS(credentials:'xgboost-doc-bucket') {
-                                s3Upload file: "${commit_id}.tar.bz2", bucket: 'xgboost-docs', acl: 'PublicRead', path: "${branch_name}.tar.bz2"
-                            }
-                        }
-                    } else {                      // This is a pull request
-                        echo 'Skipping doc build step for pull request'
-                    }
-                }
-            }
-        }
-        stage('Build & Test') {
+        stage('Jenkins: Build & Test') {
             steps {
                 script {
                     parallel (buildMatrix.findAll{it['enabled']}.collectEntries{ c ->
-                        def buildName = getBuildName(c)
-                        buildFactory(buildName, c)
-                    })
+                        def buildName = utils.getBuildName(c)
+                        utils.buildFactory(buildName, c, false, this.&buildPlatformCmake)
+                    } + [ "clang-tidy" : { buildClangTidyJob() } ])
                 }
             }
         }
     }
 }
 
-// initialize source codes
-def checkoutSrcs() {
-  retry(5) {
-    try {
-      timeout(time: 2, unit: 'MINUTES') {
-        checkout scm
-        sh 'git submodule update --init'
-      }
-    } catch (exc) {
-      deleteDir()
-      error "Failed to fetch source codes"
-    }
-  }
-}
-
-/**
- * Creates cmake and make builds
- */
-def buildFactory(buildName, conf) {
-    def os = conf["os"]
-    def nodeReq = conf["withGpu"] ? "${os} && gpu" : "${os}"
-    def dockerTarget = conf["withGpu"] ? "gpu" : "cpu"
-    [ ("${buildName}") : { buildPlatformCmake("${buildName}", conf, nodeReq, dockerTarget) }
-    ]
-}
-
 /**
  * Build platform and test it via cmake.
  */
 def buildPlatformCmake(buildName, conf, nodeReq, dockerTarget) {
-    def opts = cmakeOptions(conf)
+    def opts = utils.cmakeOptions(conf)
     // Destination dir for artifacts
     def distDir = "dist/${buildName}"
     def dockerArgs = ""
-    if(conf["withGpu"]){
+    if (conf["withGpu"]) {
         dockerArgs = "--build-arg CUDA_VERSION=" + conf["cudaVersion"]
     }
+    def test_suite = conf["withGpu"] ? (conf["multiGpu"] ? "mgpu" : "gpu") : "cpu"
     // Build node - this is returned result
-    node(nodeReq) {
-        unstash name: 'srcs'
-        echo """
-        |===== XGBoost CMake build =====
-        |  dockerTarget: ${dockerTarget}
-        |  cmakeOpts   : ${opts}
-        |=========================
-        """.stripMargin('|')
-        // Invoke command inside docker
-        sh """
-        ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/build_via_cmake.sh ${opts}
-        ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/test_${dockerTarget}.sh
-        ${dockerRun} ${dockerTarget} ${dockerArgs} bash -c "cd python-package; rm -f dist/*; python setup.py bdist_wheel --universal"
-        rm -rf "${distDir}"; mkdir -p "${distDir}/py"
-        cp xgboost "${distDir}"
-        cp -r lib "${distDir}"
-        cp -r python-package/dist "${distDir}/py"
-        # Test the wheel for compatibility on a barebones CPU container
-        ${dockerRun} release ${dockerArgs} bash -c " \
-            auditwheel show xgboost-*-py2-none-any.whl
-            pip install --user python-package/dist/xgboost-*-none-any.whl && \
-            python -m nose tests/python"
-        """
-        archiveArtifacts artifacts: "${distDir}/**/*.*", allowEmptyArchive: true
+    retry(1) {
+        node(nodeReq) {
+            unstash name: 'srcs'
+            echo """
+            |===== XGBoost CMake build =====
+            |  dockerTarget: ${dockerTarget}
+            |  cmakeOpts   : ${opts}
+            |=========================
+            """.stripMargin('|')
+            // Invoke command inside docker
+            sh """
+            ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/build_via_cmake.sh ${opts}
+            ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/test_${test_suite}.sh
+            """
+            if (!conf["multiGpu"]) {
+                sh """
+                ${dockerRun} ${dockerTarget} ${dockerArgs} bash -c "cd python-package; rm -f dist/*; python setup.py bdist_wheel --universal"
+                rm -rf "${distDir}"; mkdir -p "${distDir}/py"
+                cp xgboost "${distDir}"
+                cp -r python-package/dist "${distDir}/py"
+                # Test the wheel for compatibility on a barebones CPU container
+                ${dockerRun} release ${dockerArgs} bash -c " \
+                    pip install --user python-package/dist/xgboost-*-none-any.whl && \
+		    pytest -v --fulltrace -s tests/python"
+                # Test the wheel for compatibility on CUDA 10.0 container
+                ${dockerRun} gpu --build-arg CUDA_VERSION=10.0 bash -c " \
+                    pip install --user python-package/dist/xgboost-*-none-any.whl && \
+		    pytest -v -s --fulltrace -m '(not mgpu) and (not slow)' tests/python-gpu"
+                """
+            }
+        }
     }
 }
 
-def cmakeOptions(conf) {
-    return ([
-        conf["withGpu"] ? '-DUSE_CUDA=ON' : '-DUSE_CUDA=OFF',
-        conf["withNccl"] ? '-DUSE_NCCL=ON' : '-DUSE_NCCL=OFF',
-        conf["withOmp"] ? '-DOPEN_MP:BOOL=ON' : '']
-        ).join(" ")
-}
-
-def getBuildName(conf) {
-    def gpuLabel = conf['withGpu'] ? ("_cuda" + conf['cudaVersion'] + (conf['withNccl'] ? "_nccl" : "_nonccl")) : "_cpu"
-    def ompLabel = conf['withOmp'] ? "_omp" : ""
-    def pyLabel = "_py${conf['pythonVersion']}"
-    return "${conf['os']}${gpuLabel}${ompLabel}${pyLabel}"
-}
+/**
+ * Run a clang-tidy job on a GPU machine
+ */
+def buildClangTidyJob() {
+    def nodeReq = "linux && gpu && unrestricted"
+    node(nodeReq) {
+        unstash name: 'srcs'
+        echo "Running clang-tidy job..."
+        // Invoke command inside docker
+        // Install Google Test and Python yaml
+        dockerTarget = "clang_tidy"
+        dockerArgs = "--build-arg CUDA_VERSION=9.2"
+        sh """
+        ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/clang_tidy.sh
+        """
+      }
+  }
 

Jenkinsfile-restricted

@@ -0,0 +1,123 @@
+#!/usr/bin/groovy
+// -*- mode: groovy -*-
+// Jenkins pipeline
+// See documents at https://jenkins.io/doc/book/pipeline/jenkinsfile/
+
+import groovy.transform.Field
+
+/* Restricted tasks: tasks generating artifacts, such as binary wheels and
+                     documentation */
+
+// Command to run command inside a docker container
+def dockerRun = 'tests/ci_build/ci_build.sh'
+// Utility functions
+@Field
+def utils
+@Field
+def commit_id
+@Field
+def branch_name
+
+def buildMatrix = [
+    [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "9.2" ],
+    [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "8.0" ],
+    [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": false, "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "8.0" ],
+]
+
+pipeline {
+    // Each stage specify its own agent
+    agent none
+
+    // Setup common job properties
+    options {
+        ansiColor('xterm')
+        timestamps()
+        timeout(time: 120, unit: 'MINUTES')
+        buildDiscarder(logRotator(numToKeepStr: '10'))
+    }
+
+    // Build stages
+    stages {
+        stage('Jenkins: Get sources') {
+            agent {
+                label 'restricted'
+            }
+            steps {
+                script {
+                    utils = load('tests/ci_build/jenkins_tools.Groovy')
+                    utils.checkoutSrcs()
+                    commit_id = "${GIT_COMMIT}"
+                    branch_name = "${GIT_LOCAL_BRANCH}"
+                }
+                stash name: 'srcs', excludes: '.git/'
+                milestone label: 'Sources ready', ordinal: 1
+            }
+        }
+        stage('Jenkins: Build doc') {
+            steps {
+                script {
+                    retry(1) {
+                        node('linux && cpu && restricted') {
+                            unstash name: 'srcs'
+                            echo 'Building doc...'
+                            dir ('jvm-packages') {
+                                sh "bash ./build_doc.sh ${commit_id}"
+                                archiveArtifacts artifacts: "${commit_id}.tar.bz2", allowEmptyArchive: true
+                                echo 'Deploying doc...'
+                                withAWS(credentials:'xgboost-doc-bucket') {
+                                    s3Upload file: "${commit_id}.tar.bz2", bucket: 'xgboost-docs', acl: 'PublicRead', path: "${branch_name}.tar.bz2"
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        stage('Jenkins: Build artifacts') {
+            steps {
+                script {
+                    parallel (buildMatrix.findAll{it['enabled']}.collectEntries{ c ->
+                        def buildName = utils.getBuildName(c)
+                        utils.buildFactory(buildName, c, true, this.&buildPlatformCmake)
+                    })
+                }
+            }
+        }
+    }
+}
+
+/**
+ * Build platform and test it via cmake.
+ */
+def buildPlatformCmake(buildName, conf, nodeReq, dockerTarget) {
+    def opts = utils.cmakeOptions(conf)
+    // Destination dir for artifacts
+    def distDir = "dist/${buildName}"
+    def dockerArgs = ""
+    if(conf["withGpu"]){
+        dockerArgs = "--build-arg CUDA_VERSION=" + conf["cudaVersion"]
+    }
+    // Build node - this is returned result
+    retry(1) {
+        node(nodeReq) {
+            unstash name: 'srcs'
+            echo """
+            |===== XGBoost CMake build =====
+            |  dockerTarget: ${dockerTarget}
+            |  cmakeOpts   : ${opts}
+            |=========================
+            """.stripMargin('|')
+            // Invoke command inside docker
+            sh """
+            ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/build_via_cmake.sh ${opts}
+            ${dockerRun} ${dockerTarget} ${dockerArgs} bash -c "cd python-package; rm -f dist/*; python setup.py bdist_wheel --universal"
+            rm -rf "${distDir}"; mkdir -p "${distDir}/py"
+            cp xgboost "${distDir}"
+            cp -r lib "${distDir}"
+            cp -r python-package/dist "${distDir}/py"
+            """
+            archiveArtifacts artifacts: "${distDir}/**/*.*", allowEmptyArchive: true
+        }
+    }
+}

LICENSE

@@ -1,13 +1,201 @@
-Copyright (c) 2016 by Contributors
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
 
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
 
-   http://www.apache.org/licenses/LICENSE-2.0
+   1. Definitions.
 
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "{}"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright (c) 2018 by Contributors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

Makefile

@@ -260,7 +260,8 @@ Rpack: clean_all
 	cp ./LICENSE xgboost
 	cat R-package/src/Makevars.in|sed '2s/.*/PKGROOT=./' | sed '3s/.*/ENABLE_STD_THREAD=0/' > xgboost/src/Makevars.in
 	cp xgboost/src/Makevars.in xgboost/src/Makevars.win
-	sed -i -e 's/@OPENMP_CXXFLAGS@/$$\(SHLIB_OPENMP_CFLAGS\)/g' xgboost/src/Makevars.win
+	sed -i -e 's/@OPENMP_CXXFLAGS@/$$\(SHLIB_OPENMP_CXXFLAGS\)/g' xgboost/src/Makevars.win
+	sed -i -e 's/-pthread/$$\(SHLIB_PTHREAD_FLAGS\)/g' xgboost/src/Makevars.win
 	bash R-package/remove_warning_suppression_pragma.sh
 	rm xgboost/remove_warning_suppression_pragma.sh
 

NEWS.md

@@ -3,6 +3,331 @@ XGBoost Change Log
 
 This file records the changes in xgboost library in reverse chronological order.
 
+## v0.82 (2019.03.03)
+This release is packed with many new features and bug fixes.
+
+### Roadmap: better performance scaling for multi-core CPUs (#3957)
+* Poor performance scaling of the `hist` algorithm for multi-core CPUs has been under investigation (#3810). #3957 marks an important step toward better performance scaling, by using software pre-fetching and replacing STL vectors with C-style arrays. Special thanks to @Laurae2 and @SmirnovEgorRu.
+* See #3810 for latest progress on this roadmap.
+
+### New feature: Distributed Fast Histogram Algorithm (`hist`) (#4011, #4102, #4140, #4128)
+* It is now possible to run the `hist` algorithm in distributed setting. Special thanks to @CodingCat. The benefits include:
+  1. Faster local computation via feature binning
+  2. Support for monotonic constraints and feature interaction constraints
+  3. Simpler codebase than `approx`, allowing for future improvement
+* Depth-wise tree growing is now performed in a separate code path, so that cross-node syncronization is performed only once per level.
+
+### New feature: Multi-Node, Multi-GPU training (#4095)
+* Distributed training is now able to utilize clusters equipped with NVIDIA GPUs. In particular, the rabit AllReduce layer will communicate GPU device information. Special thanks to @mt-jones, @RAMitchell, @rongou, @trivialfis, @canonizer, and @jeffdk.
+* Resource management systems will be able to assign a rank for each GPU in the cluster.
+* In Dask, users will be able to construct a collection of XGBoost processes over an inhomogeneous device cluster (i.e. workers with different number and/or kinds of GPUs).
+
+### New feature: Multiple validation datasets in XGBoost4J-Spark (#3904, #3910)
+* You can now track the performance of the model during training with multiple evaluation datasets. By specifying `eval_sets` or call `setEvalSets` over a `XGBoostClassifier` or `XGBoostRegressor`, you can pass in multiple evaluation datasets typed as a `Map` from `String` to `DataFrame`. Special thanks to @CodingCat.
+* See the usage of multiple validation datasets [here](https://github.com/dmlc/xgboost/blob/0c1d5f1120c0a159f2567b267f0ec4ffadee00d0/jvm-packages/xgboost4j-example/src/main/scala/ml/dmlc/xgboost4j/scala/example/spark/SparkTraining.scala#L66-L78)
+
+### New feature: Additional metric functions for GPUs (#3952)
+* Element-wise metrics have been ported to GPU: `rmse`, `mae`, `logloss`, `poisson-nloglik`, `gamma-deviance`, `gamma-nloglik`, `error`, `tweedie-nloglik`. Special thanks to @trivialfis and @RAMitchell.
+* With supported metrics, XGBoost will select the correct devices based on your system and `n_gpus` parameter.
+
+### New feature: Column sampling at individual nodes (splits) (#3971)
+* Columns (features) can now be sampled at individual tree nodes, in addition to per-tree and per-level sampling. To enable per-node sampling, set `colsample_bynode` parameter, which represents the fraction of columns sampled at each node. This parameter is set to 1.0 by default (i.e. no sampling per node). Special thanks to @canonizer.
+* The `colsample_bynode` parameter works cumulatively with other `colsample_by*` parameters: for example, `{'colsample_bynode':0.5, 'colsample_bytree':0.5}` with 100 columns will give 25 features to choose from at each split.
+
+### Major API change: consistent logging level via `verbosity` (#3982, #4002, #4138)
+* XGBoost now allows fine-grained control over logging. You can set `verbosity` to 0 (silent), 1 (warning), 2 (info), and 3 (debug). This is useful for controlling the amount of logging outputs. Special thanks to @trivialfis.
+* Parameters `silent` and `debug_verbose` are now deprecated.
+* Note: Sometimes XGBoost tries to change configurations based on heuristics, which is displayed as warning message.  If there's unexpected behaviour, please try to increase value of verbosity.
+
+### Major bug fix: external memory (#4040, #4193)
+* Clarify object ownership in multi-threaded prefetcher, to avoid memory error.
+* Correctly merge two column batches (which uses [CSC layout](https://en.wikipedia.org/wiki/Sparse_matrix#Compressed_sparse_column_(CSC_or_CCS))).
+* Add unit tests for external memory.
+* Special thanks to @trivialfis and @hcho3.
+
+### Major bug fix: early stopping fixed in XGBoost4J and XGBoost4J-Spark (#3928, #4176)
+* Early stopping in XGBoost4J and XGBoost4J-Spark is now consistent with its counterpart in the Python package. Training stops if the current iteration is `earlyStoppingSteps` away from the best iteration. If there are multiple evaluation sets, only the last one is used to determinate early stop.
+* See the updated documentation [here](https://xgboost.readthedocs.io/en/release_0.82/jvm/xgboost4j_spark_tutorial.html#early-stopping)
+* Special thanks to @CodingCat, @yanboliang, and @mingyang.
+
+### Major bug fix: infrequent features should not crash distributed training (#4045)
+* For infrequently occuring features, some partitions may not get any instance. This scenario used to crash distributed training due to mal-formed ranges. The problem has now been fixed.
+* In practice, one-hot-encoded categorical variables tend to produce rare features, particularly when the cardinality is high.
+* Special thanks to @CodingCat.
+
+### Performance improvements
+* Faster, more space-efficient radix sorting in `gpu_hist` (#3895)
+* Subtraction trick in histogram calculation in `gpu_hist` (#3945)
+* More performant re-partition in XGBoost4J-Spark (#4049)
+
+### Bug-fixes
+* Fix semantics of `gpu_id` when running multiple XGBoost processes on a multi-GPU machine (#3851)
+* Fix page storage path for external memory on Windows (#3869)
+* Fix configuration setup so that DART utilizes GPU (#4024)
+* Eliminate NAN values from SHAP prediction (#3943)
+* Prevent empty quantile sketches in `hist` (#4155)
+* Enable running objectives with 0 GPU (#3878)
+* Parameters are no longer dependent on system locale (#3891, #3907)
+* Use consistent data type in the GPU coordinate descent code (#3917)
+* Remove undefined behavior in the CLI config parser on the ARM platform (#3976)
+* Initialize counters in GPU AllReduce (#3987)
+* Prevent deadlocks in GPU AllReduce (#4113)
+* Load correct values from sliced NumPy arrays (#4147, #4165)
+* Fix incorrect GPU device selection (#4161)
+* Make feature binning logic in `hist` aware of query groups when running a ranking task (#4115). For ranking task, query groups are weighted, not individual instances.
+* Generate correct C++ exception type for `LOG(FATAL)` macro (#4159)
+* Python package
+  - Python package should run on system without `PATH` environment variable (#3845)
+  - Fix `coef_` and `intercept_` signature to be compatible with `sklearn.RFECV` (#3873)
+  - Use UTF-8 encoding in Python package README, to support non-English locale (#3867)
+  - Add AUC-PR to list of metrics to maximize for early stopping (#3936)
+  - Allow loading pickles without `self.booster` attribute, for backward compatibility (#3938, #3944)
+  - White-list DART for feature importances (#4073)
+  - Update usage of [h2oai/datatable](https://github.com/h2oai/datatable) (#4123)
+* XGBoost4J-Spark
+  - Address scalability issue in prediction (#4033)
+  - Enforce the use of per-group weights for ranking task (#4118)
+  - Fix vector size of `rawPredictionCol` in `XGBoostClassificationModel` (#3932)
+  - More robust error handling in Spark tracker (#4046, #4108)
+  - Fix return type of `setEvalSets` (#4105)
+  - Return correct value of `getMaxLeaves` (#4114)
+
+### API changes
+* Add experimental parameter `single_precision_histogram` to use single-precision histograms for the `gpu_hist` algorithm (#3965)
+* Python package
+  - Add option to select type of feature importances in the scikit-learn inferface (#3876)
+  - Add `trees_to_df()` method to dump decision trees as Pandas data frame (#4153)
+  - Add options to control node shapes in the GraphViz plotting function (#3859)
+  - Add `xgb_model` option to `XGBClassifier`, to load previously saved model (#4092)
+  - Passing lists into `DMatrix` is now deprecated (#3970)
+* XGBoost4J
+  - Support multiple feature importance features (#3801)
+
+### Maintenance: Refactor C++ code for legibility and maintainability
+* Refactor `hist` algorithm code and add unit tests (#3836)
+* Minor refactoring of split evaluator in `gpu_hist` (#3889)
+* Removed unused leaf vector field in the tree model (#3989)
+* Simplify the tree representation by combining `TreeModel` and `RegTree` classes (#3995)
+* Simplify and harden tree expansion code (#4008, #4015)
+* De-duplicate parameter classes in the linear model algorithms (#4013)
+* Robust handling of ranges with C++20 span in `gpu_exact` and `gpu_coord_descent` (#4020, #4029)
+* Simplify tree training code (#3825). Also use Span class for robust handling of ranges.
+
+### Maintenance: testing, continuous integration, build system
+* Disallow `std::regex` since it's not supported by GCC 4.8.x (#3870)
+* Add multi-GPU tests for coordinate descent algorithm for linear models (#3893, #3974)
+* Enforce naming style in Python lint (#3896)
+* Refactor Python tests (#3897, #3901): Use pytest exclusively, display full trace upon failure
+* Address `DeprecationWarning` when using Python collections (#3909)
+* Use correct group for maven site plugin (#3937)
+* Jenkins CI is now using on-demand EC2 instances exclusively, due to unreliability of Spot instances (#3948)
+* Better GPU performance logging (#3945)
+* Fix GPU tests on machines with only 1 GPU (#4053)
+* Eliminate CRAN check warnings and notes (#3988)
+* Add unit tests for tree serialization (#3989)
+* Add unit tests for tree fitting functions in `hist` (#4155)
+* Add a unit test for `gpu_exact` algorithm (#4020)
+* Correct JVM CMake GPU flag (#4071)
+* Fix failing Travis CI on Mac (#4086)
+* Speed up Jenkins by not compiling CMake (#4099)
+* Analyze C++ and CUDA code using clang-tidy, as part of Jenkins CI pipeline (#4034)
+* Fix broken R test: Install Homebrew GCC (#4142)
+* Check for empty datasets in GPU unit tests (#4151)
+* Fix Windows compilation (#4139)
+* Comply with latest convention of cpplint (#4157)
+* Fix a unit test in `gpu_hist` (#4158)
+* Speed up data generation in Python tests (#4164)
+
+### Usability Improvements
+* Add link to [InfoWorld 2019 Technology of the Year Award](https://www.infoworld.com/article/3336072/application-development/infoworlds-2019-technology-of-the-year-award-winners.html) (#4116)
+* Remove outdated AWS YARN tutorial (#3885)
+* Document current limitation in number of features (#3886)
+* Remove unnecessary warning when `gblinear` is selected (#3888)
+* Document limitation of CSV parser: header not supported (#3934)
+* Log training parameters in XGBoost4J-Spark (#4091)
+* Clarify early stopping behavior in the scikit-learn interface (#3967)
+* Clarify behavior of `max_depth` parameter (#4078)
+* Revise Python docstrings for ranking task (#4121). In particular, weights must be per-group in learning-to-rank setting.
+* Document parameter `num_parallel_tree` (#4022)
+* Add Jenkins status badge (#4090)
+* Warn users against using internal functions of `Booster` object (#4066)
+* Reformat `benchmark_tree.py` to comply with Python style convention (#4126)
+* Clarify a comment in `objectiveTrait` (#4174)
+* Fix typos and broken links in documentation (#3890, #3872, #3902, #3919, #3975, #4027, #4156, #4167)
+
+### Acknowledgement
+**Contributors** (in no particular order): Jiaming Yuan (@trivialfis), Hyunsu Cho (@hcho3), Nan Zhu (@CodingCat), Rory Mitchell (@RAMitchell), Yanbo Liang (@yanboliang), Andy Adinets (@canonizer), Tong He (@hetong007), Yuan Tang (@terrytangyuan)
+
+**First-time Contributors** (in no particular order): Jelle Zijlstra (@JelleZijlstra), Jiacheng Xu (@jiachengxu), @ajing, Kashif Rasul (@kashif), @theycallhimavi, Joey Gao (@pjgao), Prabakaran Kumaresshan (@nixphix), Huafeng Wang (@huafengw), @lyxthe, Sam Wilkinson (@scwilkinson), Tatsuhito Kato (@stabacov), Shayak Banerjee (@shayakbanerjee), Kodi Arfer (@Kodiologist), @KyleLi1985, Egor Smirnov (@SmirnovEgorRu), @tmitanitky, Pasha Stetsenko (@st-pasha), Kenichi Nagahara (@keni-chi), Abhai Kollara Dilip (@abhaikollara), Patrick Ford (@pford221), @hshujuan, Matthew Jones (@mt-jones), Thejaswi Rao (@teju85), Adam November (@anovember)
+
+**First-time Reviewers** (in no particular order): Mingyang Hu (@mingyang), Theodore Vasiloudis (@thvasilo), Jakub Troszok (@troszok), Rong Ou (@rongou), @Denisevi4, Matthew Jones (@mt-jones), Jeff Kaplan (@jeffdk)
+
+## v0.81 (2018.11.04)
+### New feature: feature interaction constraints
+* Users are now able to control which features (independent variables) are allowed to interact by specifying feature interaction constraints (#3466).
+* [Tutorial](https://xgboost.readthedocs.io/en/release_0.81/tutorials/feature_interaction_constraint.html) is available, as well as [R](https://github.com/dmlc/xgboost/blob/9254c58e4dfff6a59dc0829a2ceb02e45ed17cd0/R-package/demo/interaction_constraints.R) and [Python](https://github.com/dmlc/xgboost/blob/9254c58e4dfff6a59dc0829a2ceb02e45ed17cd0/tests/python/test_interaction_constraints.py) examples.
+
+### New feature: learning to rank using scikit-learn interface
+* Learning to rank task is now available for the scikit-learn interface of the Python package (#3560, #3848). It is now possible to integrate the XGBoost ranking model into the scikit-learn learning pipeline.
+* Examples of using `XGBRanker` class is found at [demo/rank/rank_sklearn.py](https://github.com/dmlc/xgboost/blob/24a268a2e3cb17302db3d72da8f04016b7d352d9/demo/rank/rank_sklearn.py).
+
+### New feature: R interface for SHAP interactions
+* SHAP (SHapley Additive exPlanations) is a unified approach to explain the output of any machine learning model. Previously, this feature was only available from the Python package; now it is available from the R package as well (#3636).
+
+### New feature: GPU predictor now use multiple GPUs to predict
+* GPU predictor is now able to utilize multiple GPUs at once to accelerate prediction (#3738)
+
+### New feature: Scale distributed XGBoost to large-scale clusters
+* Fix OS file descriptor limit assertion error on large cluster (#3835, dmlc/rabit#73) by replacing `select()` based AllReduce/Broadcast with `poll()` based implementation.
+* Mitigate tracker "thundering herd" issue on large cluster. Add exponential backoff retry when workers connect to tracker.
+* With this change, we were able to scale to 1.5k executors on a 12 billion row dataset after some tweaks here and there.
+
+### New feature: Additional objective functions for GPUs
+* New objective functions ported to GPU: `hinge`, `multi:softmax`, `multi:softprob`, `count:poisson`, `reg:gamma`, `"reg:tweedie`.
+* With supported objectives, XGBoost will select the correct devices based on your system and `n_gpus` parameter.
+
+### Major bug fix: learning to rank with XGBoost4J-Spark
+* Previously, `repartitionForData` would shuffle data and lose ordering necessary for ranking task.
+* To fix this issue, data points within each RDD partition is explicitly group by their group (query session) IDs (#3654). Also handle empty RDD partition carefully (#3750).
+
+### Major bug fix: early stopping fixed in XGBoost4J-Spark
+* Earlier implementation of early stopping had incorrect semantics and didn't let users to specify direction for optimizing (maximize / minimize)
+* A parameter `maximize_evaluation_metrics` is defined so as to tell whether a metric should be maximized or minimized as part of early stopping criteria (#3808). Also early stopping now has correct semantics.
+
+### API changes
+* Column sampling by level (`colsample_bylevel`) is now functional for `hist` algorithm (#3635, #3862)
+* GPU tag `gpu:` for regression objectives are now deprecated. XGBoost will select the correct devices automatically (#3643)
+* Add `disable_default_eval_metric` parameter to disable default metric (#3606)
+* Experimental AVX support for gradient computation is removed (#3752)
+* XGBoost4J-Spark
+  - Add `rank:ndcg` and `rank:map` to supported objectives (#3697)
+* Python package
+  - Add `callbacks` argument to `fit()` function of sciki-learn API (#3682)
+  - Add `XGBRanker` to scikit-learn interface (#3560, #3848)
+  - Add `validate_features` argument to `predict()` function of scikit-learn API (#3653)
+  - Allow scikit-learn grid search over parameters specified as keyword arguments (#3791)
+  - Add `coef_` and `intercept_` as properties of scikit-learn wrapper (#3855). Some scikit-learn functions expect these properties.
+
+### Performance improvements
+* Address very high GPU memory usage for large data (#3635)
+* Fix performance regression within `EvaluateSplits()` of `gpu_hist` algorithm. (#3680)
+
+### Bug-fixes
+* Fix a problem in GPU quantile sketch with tiny instance weights. (#3628)
+* Fix copy constructor for `HostDeviceVectorImpl` to prevent dangling pointers (#3657)
+* Fix a bug in partitioned file loading (#3673)
+* Fixed an uninitialized pointer in `gpu_hist` (#3703)
+* Reshared data among GPUs when number of GPUs is changed (#3721)
+* Add back `max_delta_step` to split evaluation (#3668)
+* Do not round up integer thresholds for integer features in JSON dump (#3717)
+* Use `dmlc::TemporaryDirectory` to handle temporaries in cross-platform way (#3783)
+* Fix accuracy problem with `gpu_hist` when `min_child_weight` and `lambda` are set to 0 (#3793)
+* Make sure that `tree_method` parameter is recognized and not silently ignored (#3849)
+* XGBoost4J-Spark
+  - Make sure `thresholds` are considered when executing `predict()` method (#3577)
+  - Avoid losing precision when computing probabilities by converting to `Double` early (#3576)
+  - `getTreeLimit()` should return `Int` (#3602)
+  - Fix checkpoint serialization on HDFS (#3614)
+  - Throw `ControlThrowable` instead of `InterruptedException` so that it is properly re-thrown (#3632)
+  - Remove extraneous output to stdout (#3665)
+  - Allow specification of task type for custom objectives and evaluations (#3646)
+  - Fix distributed updater check (#3739)
+  - Fix issue when spark job execution thread cannot return before we execute `first()` (#3758)
+* Python package
+  - Fix accessing `DMatrix.handle` before it is set (#3599)
+  - `XGBClassifier.predict()` should return margin scores when `output_margin` is set to true (#3651)
+  - Early stopping callback should maximize metric of form `NDCG@n-` (#3685)
+  - Preserve feature names when slicing `DMatrix` (#3766)
+* R package
+  - Replace `nround` with `nrounds` to match actual parameter (#3592)
+  - Amend `xgb.createFolds` to handle classes of a single element (#3630)
+  - Fix buggy random generator and make `colsample_bytree` functional (#3781)
+
+### Maintenance: testing, continuous integration, build system
+* Add sanitizers tests to Travis CI (#3557)
+* Add NumPy, Matplotlib, Graphviz as requirements for doc build (#3669)
+* Comply with CRAN submission policy (#3660, #3728)
+* Remove copy-paste error in JVM test suite (#3692)
+* Disable flaky tests in `R-package/tests/testthat/test_update.R` (#3723)
+* Make Python tests compatible with scikit-learn 0.20 release (#3731)
+* Separate out restricted and unrestricted tasks, so that pull requests don't build downloadable artifacts (#3736)
+* Add multi-GPU unit test environment (#3741)
+* Allow plug-ins to be built by CMake (#3752)
+* Test wheel compatibility on CPU containers for pull requests (#3762)
+* Fix broken doc build due to Matplotlib 3.0 release (#3764)
+* Produce `xgboost.so` for XGBoost-R on Mac OSX, so that `make install` works (#3767)
+* Retry Jenkins CI tests up to 3 times to improve reliability (#3769, #3769, #3775, #3776, #3777)
+* Add basic unit tests for `gpu_hist` algorithm (#3785)
+* Fix Python environment for distributed unit tests (#3806)
+* Test wheels on CUDA 10.0 container for compatibility (#3838)
+* Fix JVM doc build (#3853)
+
+### Maintenance: Refactor C++ code for legibility and maintainability
+* Merge generic device helper functions into `GPUSet` class (#3626)
+* Re-factor column sampling logic into `ColumnSampler` class (#3635, #3637)
+* Replace `std::vector` with `HostDeviceVector` in `MetaInfo` and `SparsePage` (#3446)
+* Simplify `DMatrix` class (#3395)
+* De-duplicate CPU/GPU code using `Transform` class (#3643, #3751)
+* Remove obsoleted `QuantileHistMaker` class (#3761)
+* Remove obsoleted `NoConstraint` class (#3792)
+
+### Other Features
+* C++20-compliant Span class for safe pointer indexing (#3548, #3588)
+* Add helper functions to manipulate multiple GPU devices (#3693)
+* XGBoost4J-Spark
+  - Allow specifying host ip from the `xgboost-tracker.properties file` (#3833). This comes in handy when `hosts` files doesn't correctly define localhost.
+
+### Usability Improvements
+* Add reference to GitHub repository in `pom.xml` of JVM packages (#3589)
+* Add R demo of multi-class classification (#3695)
+* Document JSON dump functionality (#3600, #3603)
+* Document CUDA requirement and lack of external memory for GPU algorithms (#3624)
+* Document LambdaMART objectives, both pairwise and listwise (#3672)
+* Document `aucpr` evaluation metric (#3687)
+* Document gblinear parameters: `feature_selector` and `top_k` (#3780)
+* Add instructions for using MinGW-built XGBoost with Python. (#3774)
+* Removed nonexistent parameter `use_buffer` from documentation (#3610)
+* Update Python API doc to include all classes and members (#3619, #3682)
+* Fix typos and broken links in documentation (#3618, #3640, #3676, #3713, #3759, #3784, #3843, #3852)
+* Binary classification demo should produce LIBSVM with 0-based indexing (#3652)
+* Process data once for Python and CLI examples of learning to rank (#3666)
+* Include full text of Apache 2.0 license in the repository (#3698)
+* Save predictor parameters in model file (#3856)
+* JVM packages
+  - Let users specify feature names when calling `getModelDump` and `getFeatureScore` (#3733)
+  - Warn the user about the lack of over-the-wire encryption (#3667)
+  - Fix errors in examples (#3719)
+  - Document choice of trackers (#3831)
+  - Document that vanilla Apache Spark is required (#3854)
+* Python package
+  - Document that custom objective can't contain colon (:) (#3601)
+  - Show a better error message for failed library loading (#3690)
+  - Document that feature importance is unavailable for non-tree learners (#3765)
+  - Document behavior of `get_fscore()` for zero-importance features (#3763)
+  - Recommend pickling as the way to save `XGBClassifier` / `XGBRegressor` / `XGBRanker` (#3829)
+* R package
+  - Enlarge variable importance plot to make it more visible (#3820)
+
+### BREAKING CHANGES
+* External memory page files have changed, breaking backwards compatibility for temporary storage used during external memory training. This only affects external memory users upgrading their xgboost version - we recommend clearing all `*.page` files before resuming training. Model serialization is unaffected.
+
+### Known issues
+* Quantile sketcher fails to produce any quantile for some edge cases (#2943)
+* The `hist` algorithm leaks memory when used with learning rate decay callback (#3579)
+* Using custom evaluation funciton together with early stopping causes assertion failure in XGBoost4J-Spark (#3595)
+* Early stopping doesn't work with `gblinear` learner (#3789)
+* Label and weight vectors are not reshared upon the change in number of GPUs (#3794). To get around this issue, delete the `DMatrix` object and re-load.
+* The `DMatrix` Python objects are initialized with incorrect values when given array slices (#3841)
+* The `gpu_id` parameter is broken and not yet properly supported (#3850)
+
+### Acknowledgement
+**Contributors** (in no particular order): Hyunsu Cho (@hcho3), Jiaming Yuan (@trivialfis), Nan Zhu (@CodingCat), Rory Mitchell (@RAMitchell), Andy Adinets (@canonizer), Vadim Khotilovich (@khotilov), Sergei Lebedev (@superbobry)
+
+**First-time Contributors** (in no particular order): Matthew Tovbin (@tovbinm), Jakob Richter (@jakob-r), Grace Lam (@grace-lam), Grant W Schneider (@grantschneider), Andrew Thia (@BlueTea88), Sergei Chipiga (@schipiga), Joseph Bradley (@jkbradley), Chen Qin (@chenqin), Jerry Lin (@linjer), Dmitriy Rybalko (@rdtft), Michael Mui (@mmui), Takahiro Kojima (@515hikaru), Bruce Zhao (@BruceZhaoR), Wei Tian (@weitian), Saumya Bhatnagar (@Sam1301), Juzer Shakir (@JuzerShakir), Zhao Hang (@cleghom), Jonathan Friedman (@jontonsoup), Bruno Tremblay (@meztez), Boris Filippov (@frenzykryger), @Shiki-H, @mrgutkun, @gorogm, @htgeis, @jakehoare, @zengxy, @KOLANICH
+
+**First-time Reviewers** (in no particular order): Nikita Titov (@StrikerRUS), Xiangrui Meng (@mengxr), Nirmal Borah (@Nirmal-Neel)
+
+
 ## v0.80 (2018.08.13)
 * **JVM packages received a major upgrade**: To consolidate the APIs and improve the user experience, we refactored the design of XGBoost4J-Spark in a significant manner. (#3387)
   - Consolidated APIs: It is now much easier to integrate XGBoost models into a Spark ML pipeline. Users can control behaviors like output leaf prediction results by setting corresponding column names. Training is now more consistent with other Estimators in Spark MLLIB: there is now one single method `fit()` to train decision trees.
@@ -13,7 +338,7 @@ This file records the changes in xgboost library in reverse chronological order.
   - Latest master: https://xgboost.readthedocs.io/en/latest
   - 0.80 stable: https://xgboost.readthedocs.io/en/release_0.80
   - 0.72 stable: https://xgboost.readthedocs.io/en/release_0.72
-* Ranking task now uses instance weights (#3379)
+* Support for per-group weights in ranking objective (#3379)
 * Fix inaccurate decimal parsing (#3546)
 * New functionality
   - Query ID column support in LIBSVM data files (#2749). This is convenient for performing ranking task in distributed setting.
@@ -173,7 +498,7 @@ This version is only applicable for the Python package. The content is identical
   - Compatibility fix for Python 2.6
   - Call `print_evaluation` callback at last iteration
   - Use appropriate integer types when calling native code, to prevent truncation and memory error
-  - Fix shared library loading on Mac OS X 
+  - Fix shared library loading on Mac OS X
 * R package:
   - New parameters:
     - `silent` in `xgb.DMatrix()`
@@ -214,7 +539,7 @@ This version is only applicable for the Python package. The content is identical
   - Support instance weights
   - Use `SparkParallelismTracker` to prevent jobs from hanging forever
   - Expose train-time evaluation metrics via `XGBoostModel.summary`
-  - Option to specify `host-ip` explicitly in the Rabit tracker 
+  - Option to specify `host-ip` explicitly in the Rabit tracker
 * Documentation
   - Better math notation for gradient boosting
   - Updated build instructions for Mac OS X

R-package/DESCRIPTION

@@ -1,8 +1,8 @@
 Package: xgboost
 Type: Package
 Title: Extreme Gradient Boosting
-Version: 0.80.1
-Date: 2018-08-13
+Version: 0.82.0.1
+Date: 2019-03-11
 Authors@R: c(
   person("Tianqi", "Chen", role = c("aut"),
          email = "tianqi.tchen@gmail.com"),
@@ -61,5 +61,5 @@ Imports:
     data.table (>= 1.9.6),
     magrittr (>= 1.5),
     stringi (>= 0.5.2)
-RoxygenNote: 6.0.1
+RoxygenNote: 6.1.0
 SystemRequirements: GNU make, C++11

R-package/R/callbacks.R

@@ -168,7 +168,7 @@ cb.evaluation.log <- function() {
 #' at the beginning of each iteration.
 #' 
 #' Note that when training is resumed from some previous model, and a function is used to 
-#' reset a parameter value, the \code{nround} argument in this function would be the 
+#' reset a parameter value, the \code{nrounds} argument in this function would be the 
 #' the number of boosting rounds in the current training.
 #'
 #' Callback function expects the following values to be set in its calling frame:

R-package/R/utils.R

@@ -74,6 +74,19 @@ check.booster.params <- function(params, ...) {
         params[['monotone_constraints']] = vec2str
   }
   
+  # interaction constraints parser (convert from list of column indices to string)
+  if (!is.null(params[['interaction_constraints']]) && 
+      typeof(params[['interaction_constraints']]) != "character"){
+    # check input class
+    if (class(params[['interaction_constraints']]) != 'list') stop('interaction_constraints should be class list')
+    if (!all(unique(sapply(params[['interaction_constraints']], class)) %in% c('numeric','integer'))) {
+      stop('interaction_constraints should be a list of numeric/integer vectors')
+    }
+
+    # recast parameter as string
+    interaction_constraints <- sapply(params[['interaction_constraints']], function(x) paste0('[', paste(x, collapse=','), ']'))
+    params[['interaction_constraints']] <- paste0('[', paste(interaction_constraints, collapse=','), ']')
+  }
   return(params)
 }
 
@@ -262,7 +275,8 @@ xgb.createFolds <- function(y, k = 10)
       ## add enough random integers to get  length(seqVector) == numInClass[i]
       if (numInClass[i] %% k > 0) seqVector <- c(seqVector, sample.int(k, numInClass[i] %% k))
       ## shuffle the integers for fold assignment and assign to this classes's data
-      foldVector[y == dimnames(numInClass)$y[i]] <- sample(seqVector)
+      ## seqVector[sample.int(length(seqVector))] is used to handle length(seqVector) == 1
+      foldVector[y == dimnames(numInClass)$y[i]] <- seqVector[sample.int(length(seqVector))]
     }
   } else {
     foldVector <- seq(along = y)

R-package/R/xgb.Booster.R

@@ -129,11 +129,13 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #'        logistic regression would result in predictions for log-odds instead of probabilities.
 #' @param ntreelimit limit the number of model's trees or boosting iterations used in prediction (see Details).
 #'        It will use all the trees by default (\code{NULL} value).
-#' @param predleaf whether predict leaf index instead.
-#' @param predcontrib whether to return feature contributions to individual predictions instead (see Details).
+#' @param predleaf whether predict leaf index.
+#' @param predcontrib whether to return feature contributions to individual predictions (see Details).
 #' @param approxcontrib whether to use a fast approximation for feature contributions (see Details).
+#' @param predinteraction whether to return contributions of feature interactions to individual predictions (see Details).
 #' @param reshape whether to reshape the vector of predictions to a matrix form when there are several
-#'        prediction outputs per case. This option has no effect when \code{predleaf = TRUE}.
+#'        prediction outputs per case. This option has no effect when either of predleaf, predcontrib,
+#'        or predinteraction flags is TRUE.
 #' @param ... Parameters passed to \code{predict.xgb.Booster}
 #'
 #' @details
@@ -158,6 +160,11 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' Setting \code{approxcontrib = TRUE} approximates these values following the idea explained
 #' in \url{http://blog.datadive.net/interpreting-random-forests/}.
 #'
+#' With \code{predinteraction = TRUE}, SHAP values of contributions of interaction of each pair of features
+#' are computed. Note that this operation might be rather expensive in terms of compute and memory.
+#' Since it quadratically depends on the number of features, it is recommended to perfom selection
+#' of the most important features first. See below about the format of the returned results.
+#'
 #' @return
 #' For regression or binary classification, it returns a vector of length \code{nrows(newdata)}.
 #' For multiclass classification, either a \code{num_class * nrows(newdata)} vector or
@@ -173,6 +180,14 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' such a matrix. The contribution values are on the scale of untransformed margin
 #' (e.g., for binary classification would mean that the contributions are log-odds deviations from bias).
 #'
+#' When \code{predinteraction = TRUE} and it is not a multiclass setting, the output is a 3d array with
+#' dimensions \code{c(nrow, num_features + 1, num_features + 1)}. The off-diagonal (in the last two dimensions)
+#' elements represent different features interaction contributions. The array is symmetric WRT the last
+#' two dimensions. The "+ 1" columns corresponds to bias. Summing this array along the last dimension should
+#' produce practically the same result as predict with \code{predcontrib = TRUE}.
+#' For a multiclass case, a list of \code{num_class} elements is returned, where each element is
+#' such an array.
+#'
 #' @seealso
 #' \code{\link{xgb.train}}.
 #' 
@@ -269,7 +284,8 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' @rdname predict.xgb.Booster
 #' @export
 predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FALSE, ntreelimit = NULL,
-                                predleaf = FALSE, predcontrib = FALSE, approxcontrib = FALSE, reshape = FALSE, ...) {
+                                predleaf = FALSE, predcontrib = FALSE, approxcontrib = FALSE, predinteraction = FALSE,
+                                reshape = FALSE, ...) {
 
   object <- xgb.Booster.complete(object, saveraw = FALSE)
   if (!inherits(newdata, "xgb.DMatrix"))
@@ -285,7 +301,8 @@ predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FA
   if (ntreelimit < 0)
     stop("ntreelimit cannot be negative")
 
-  option <- 0L + 1L * as.logical(outputmargin) + 2L * as.logical(predleaf) + 4L * as.logical(predcontrib) + 8L * as.logical(approxcontrib)
+  option <- 0L + 1L * as.logical(outputmargin) + 2L * as.logical(predleaf) + 4L * as.logical(predcontrib) +
+    8L * as.logical(approxcontrib) + 16L * as.logical(predinteraction)
 
   ret <- .Call(XGBoosterPredict_R, object$handle, newdata, option[1], as.integer(ntreelimit))
 
@@ -305,17 +322,28 @@ predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FA
   } else if (predcontrib) {
     n_col1 <- ncol(newdata) + 1
     n_group <- npred_per_case / n_col1
-    dnames <- if (!is.null(colnames(newdata))) list(NULL, c(colnames(newdata), "BIAS")) else NULL
+    cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "BIAS") else NULL
     ret <- if (n_ret == n_row) {
-      matrix(ret, ncol = 1, dimnames = dnames)
+      matrix(ret, ncol = 1, dimnames = list(NULL, cnames))
     } else if (n_group == 1) {
-      matrix(ret, nrow = n_row, byrow = TRUE, dimnames = dnames)
+      matrix(ret, nrow = n_row, byrow = TRUE, dimnames = list(NULL, cnames))
     } else {
-      grp_mask <- rep(seq_len(n_col1), n_row) +
-        rep((seq_len(n_row) - 1) * n_col1 * n_group, each = n_col1)
-      lapply(seq_len(n_group), function(g) {
-        matrix(ret[grp_mask + n_col1 * (g - 1)], nrow = n_row, byrow = TRUE, dimnames = dnames)
-      })
+      arr <- array(ret, c(n_col1, n_group, n_row),
+                   dimnames = list(cnames, NULL, NULL)) %>% aperm(c(2,3,1)) # [group, row, col]
+      lapply(seq_len(n_group), function(g) arr[g,,])
+    }
+  } else if (predinteraction) {
+    n_col1 <- ncol(newdata) + 1
+    n_group <- npred_per_case / n_col1^2
+    cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "BIAS") else NULL
+    ret <- if (n_ret == n_row) {
+      matrix(ret, ncol = 1, dimnames = list(NULL, cnames))
+    } else if (n_group == 1) {
+      array(ret, c(n_col1, n_col1, n_row), dimnames = list(cnames, cnames, NULL)) %>% aperm(c(3,1,2))
+    } else {
+      arr <- array(ret, c(n_col1, n_col1, n_group, n_row),
+                   dimnames = list(cnames, cnames, NULL, NULL)) %>% aperm(c(3,4,1,2)) # [group, row, col1, col2]
+      lapply(seq_len(n_group), function(g) arr[g,,,])
     }
   } else if (reshape && npred_per_case > 1) {
     ret <- matrix(ret, nrow = n_row, byrow = TRUE)

R-package/R/xgb.create.features.R

@@ -52,9 +52,9 @@
 #' dtest <- xgb.DMatrix(data = agaricus.test$data, label = agaricus.test$label)
 #'
 #' param <- list(max_depth=2, eta=1, silent=1, objective='binary:logistic')
-#' nround = 4
+#' nrounds = 4
 #'
-#' bst = xgb.train(params = param, data = dtrain, nrounds = nround, nthread = 2)
+#' bst = xgb.train(params = param, data = dtrain, nrounds = nrounds, nthread = 2)
 #' 
 #' # Model accuracy without new features
 #' accuracy.before <- sum((predict(bst, agaricus.test$data) >= 0.5) == agaricus.test$label) /
@@ -68,7 +68,7 @@
 #' new.dtrain <- xgb.DMatrix(data = new.features.train, label = agaricus.train$label)
 #' new.dtest <- xgb.DMatrix(data = new.features.test, label = agaricus.test$label)
 #' watchlist <- list(train = new.dtrain)
-#' bst <- xgb.train(params = param, data = new.dtrain, nrounds = nround, nthread = 2)
+#' bst <- xgb.train(params = param, data = new.dtrain, nrounds = nrounds, nthread = 2)
 #' 
 #' # Model accuracy with new features
 #' accuracy.after <- sum((predict(bst, new.dtest) >= 0.5) == agaricus.test$label) /

R-package/R/xgb.ggplot.R

@@ -22,7 +22,7 @@ xgb.ggplot.importance <- function(importance_matrix = NULL, top_n = NULL, measur
 
   plot <-
     ggplot2::ggplot(importance_matrix, 
-                    ggplot2::aes(x = factor(Feature, levels = rev(Feature)), y = Importance, width = 0.05),
+                    ggplot2::aes(x = factor(Feature, levels = rev(Feature)), y = Importance, width = 0.5),
                     environment = environment()) + 
     ggplot2::geom_bar(ggplot2::aes(fill = Cluster), stat = "identity", position = "identity") + 
     ggplot2::coord_flip() + 

R-package/R/xgb.plot.deepness.R

@@ -27,7 +27,7 @@
 #' a tree's median absolute leaf weight changes through the iterations.
 #'
 #' This function was inspired by the blog post
-#' \url{http://aysent.github.io/2015/11/08/random-forest-leaf-visualization.html}.
+#' \url{https://github.com/aysent/random-forest-leaf-visualization}.
 #' 
 #' @return
 #' 

R-package/R/xgb.train.R

@@ -22,10 +22,11 @@
 #'   \item \code{gamma} minimum loss reduction required to make a further partition on a leaf node of the tree. the larger, the more conservative the algorithm will be. 
 #'   \item \code{max_depth} maximum depth of a tree. Default: 6
 #'   \item \code{min_child_weight} minimum sum of instance weight (hessian) needed in a child. If the tree partition step results in a leaf node with the sum of instance weight less than min_child_weight, then the building process will give up further partitioning. In linear regression mode, this simply corresponds to minimum number of instances needed to be in each node. The larger, the more conservative the algorithm will be. Default: 1
-#'   \item \code{subsample} subsample ratio of the training instance. Setting it to 0.5 means that xgboost randomly collected half of the data instances to grow trees and this will prevent overfitting. It makes computation shorter (because less data to analyse). It is advised to use this parameter with \code{eta} and increase \code{nround}. Default: 1 
+#'   \item \code{subsample} subsample ratio of the training instance. Setting it to 0.5 means that xgboost randomly collected half of the data instances to grow trees and this will prevent overfitting. It makes computation shorter (because less data to analyse). It is advised to use this parameter with \code{eta} and increase \code{nrounds}. Default: 1 
 #'   \item \code{colsample_bytree} subsample ratio of columns when constructing each tree. Default: 1
 #'   \item \code{num_parallel_tree} Experimental parameter. number of trees to grow per round. Useful to test Random Forest through Xgboost (set \code{colsample_bytree < 1}, \code{subsample  < 1}  and \code{round = 1}) accordingly. Default: 1
 #'   \item \code{monotone_constraints} A numerical vector consists of \code{1}, \code{0} and \code{-1} with its length equals to the number of features in the training data. \code{1} is increasing, \code{-1} is decreasing and \code{0} is no constraint.
+#'   \item \code{interaction_constraints} A list of vectors specifying feature indices of permitted interactions. Each item of the list represents one permitted interaction where specified features are allowed to interact with each other. Feature index values should start from \code{0} (\code{0} references the first column).  Leave argument unspecified for no interaction constraints.
 #' }
 #' 
 #' 2.2. Parameter for Linear Booster

R-package/configure

@@ -1667,12 +1667,12 @@ OPENMP_CXXFLAGS=""
 
 if test `uname -s` = "Linux"
 then
-  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CFLAGS)"
+  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CXXFLAGS)"
 fi
 
 if test `uname -s` = "Darwin"
 then
-  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CFLAGS)"
+  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CXXFLAGS)"
   ac_pkg_openmp=no
   { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether OpenMP will work in a package" >&5
 $as_echo_n "checking whether OpenMP will work in a package... " >&6; }

R-package/configure.ac

@@ -8,12 +8,12 @@ OPENMP_CXXFLAGS=""
 
 if test `uname -s` = "Linux"
 then
-  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CFLAGS)"
+  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CXXFLAGS)"
 fi
 
 if test `uname -s` = "Darwin"
 then
-  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CFLAGS)"
+  OPENMP_CXXFLAGS="\$(SHLIB_OPENMP_CXXFLAGS)"
   ac_pkg_openmp=no
   AC_MSG_CHECKING([whether OpenMP will work in a package])
   AC_LANG_CONFTEST(

R-package/demo/00Index

@@ -11,4 +11,5 @@ early_stopping                  Early Stop in training
 poisson_regression              Poisson Regression on count data
 tweedie_regression              Tweddie Regression
 gpu_accelerated                 GPU-accelerated tree building algorithms
+interaction_constraints         Interaction constraints among features
 

R-package/demo/cross_validation.R

@@ -5,20 +5,20 @@ data(agaricus.test, package='xgboost')
 dtrain <- xgb.DMatrix(agaricus.train$data, label = agaricus.train$label)
 dtest <- xgb.DMatrix(agaricus.test$data, label = agaricus.test$label)
 
-nround <- 2
+nrounds <- 2
 param <- list(max_depth=2, eta=1, silent=1, nthread=2, objective='binary:logistic')
 
 cat('running cross validation\n')
 # do cross validation, this will print result out as
 # [iteration]  metric_name:mean_value+std_value
 # std_value is standard deviation of the metric
-xgb.cv(param, dtrain, nround, nfold=5, metrics={'error'})
+xgb.cv(param, dtrain, nrounds, nfold=5, metrics={'error'})
 
 cat('running cross validation, disable standard deviation display\n')
 # do cross validation, this will print result out as
 # [iteration]  metric_name:mean_value+std_value
 # std_value is standard deviation of the metric
-xgb.cv(param, dtrain, nround, nfold=5,
+xgb.cv(param, dtrain, nrounds, nfold=5,
        metrics='error', showsd = FALSE)
 
 ###
@@ -43,9 +43,9 @@ evalerror <- function(preds, dtrain) {
 param <- list(max_depth=2, eta=1, silent=1,
               objective = logregobj, eval_metric = evalerror)
 # train with customized objective
-xgb.cv(params = param, data = dtrain, nrounds = nround, nfold = 5)
+xgb.cv(params = param, data = dtrain, nrounds = nrounds, nfold = 5)
 
 # do cross validation with prediction values for each fold
-res <- xgb.cv(params = param, data = dtrain, nrounds = nround, nfold = 5, prediction = TRUE)
+res <- xgb.cv(params = param, data = dtrain, nrounds = nrounds, nfold = 5, prediction = TRUE)
 res$evaluation_log
 length(res$pred)

R-package/demo/custom_objective.R

@@ -33,7 +33,7 @@ evalerror <- function(preds, dtrain) {
   return(list(metric = "error", value = err))
 }
 
-param <- list(max_depth=2, eta=1, nthread = 2, silent=1, 
+param <- list(max_depth=2, eta=1, nthread = 2, verbosity=0, 
               objective=logregobj, eval_metric=evalerror)
 print ('start training with user customized objective')
 # training with customized objective, we can also do step by step training
@@ -57,7 +57,7 @@ logregobjattr <- function(preds, dtrain) {
   hess <- preds * (1 - preds)
   return(list(grad = grad, hess = hess))
 }
-param <- list(max_depth=2, eta=1, nthread = 2, silent=1, 
+param <- list(max_depth=2, eta=1, nthread = 2, verbosity=0, 
               objective=logregobjattr, eval_metric=evalerror)
 print ('start training with user customized objective, with additional attributes in DMatrix')
 # training with customized objective, we can also do step by step training

R-package/demo/early_stopping.R

@@ -7,7 +7,7 @@ dtest <- xgb.DMatrix(agaricus.test$data, label = agaricus.test$label)
 # note: for customized objective function, we leave objective as default
 # note: what we are getting is margin value in prediction
 # you must know what you are doing
-param <- list(max_depth=2, eta=1, nthread = 2, silent=1)
+param <- list(max_depth=2, eta=1, nthread=2, verbosity=0)
 watchlist <- list(eval = dtest)
 num_round <- 20
 # user define objective function, given prediction, return gradient and second order gradient
@@ -32,9 +32,9 @@ evalerror <- function(preds, dtrain) {
 }
 print ('start training with early Stopping setting')
 
-bst <- xgb.train(param, dtrain, num_round, watchlist, 
+bst <- xgb.train(param, dtrain, num_round, watchlist,
                  objective = logregobj, eval_metric = evalerror, maximize = FALSE,
                  early_stopping_round = 3)
-bst <- xgb.cv(param, dtrain, num_round, nfold = 5, 
+bst <- xgb.cv(param, dtrain, num_round, nfold = 5,
               objective = logregobj, eval_metric = evalerror,
               maximize = FALSE, early_stopping_rounds = 3)

R-package/demo/interaction_constraints.R

@@ -0,0 +1,105 @@
+library(xgboost)
+library(data.table)
+
+set.seed(1024)
+
+# Function to obtain a list of interactions fitted in trees, requires input of maximum depth
+treeInteractions <- function(input_tree, input_max_depth){
+  trees <- copy(input_tree)  # copy tree input to prevent overwriting
+  if (input_max_depth < 2) return(list())  # no interactions if max depth < 2
+  if (nrow(input_tree) == 1) return(list())
+
+  # Attach parent nodes
+  for (i in 2:input_max_depth){
+    if (i == 2) trees[, ID_merge:=ID] else trees[, ID_merge:=get(paste0('parent_',i-2))]
+    parents_left <- trees[!is.na(Split), list(i.id=ID, i.feature=Feature, ID_merge=Yes)]
+    parents_right <- trees[!is.na(Split), list(i.id=ID, i.feature=Feature, ID_merge=No)]
+
+    setorderv(trees, 'ID_merge')
+    setorderv(parents_left, 'ID_merge')
+    setorderv(parents_right, 'ID_merge')
+
+    trees <- merge(trees, parents_left, by='ID_merge', all.x=T)
+    trees[!is.na(i.id), c(paste0('parent_', i-1), paste0('parent_feat_', i-1)):=list(i.id, i.feature)]
+    trees[, c('i.id','i.feature'):=NULL]
+
+    trees <- merge(trees, parents_right, by='ID_merge', all.x=T)
+    trees[!is.na(i.id), c(paste0('parent_', i-1), paste0('parent_feat_', i-1)):=list(i.id, i.feature)]
+    trees[, c('i.id','i.feature'):=NULL]
+  }
+
+  # Extract nodes with interactions
+  interaction_trees <- trees[!is.na(Split) & !is.na(parent_1), 
+                             c('Feature',paste0('parent_feat_',1:(input_max_depth-1))), with=F]
+  interaction_trees_split <- split(interaction_trees, 1:nrow(interaction_trees))
+  interaction_list <- lapply(interaction_trees_split, as.character)
+
+  # Remove NAs (no parent interaction)
+  interaction_list <- lapply(interaction_list, function(x) x[!is.na(x)])
+
+  # Remove non-interactions (same variable)
+  interaction_list <- lapply(interaction_list, unique)  # remove same variables
+  interaction_length <- sapply(interaction_list, length)
+  interaction_list <- interaction_list[interaction_length > 1]
+  interaction_list <- unique(lapply(interaction_list, sort))
+  return(interaction_list)
+}
+
+# Generate sample data
+x <- list()
+for (i in 1:10){
+  x[[i]] = i*rnorm(1000, 10)
+}
+x <- as.data.table(x)
+
+y = -1*x[, rowSums(.SD)] + x[['V1']]*x[['V2']] + x[['V3']]*x[['V4']]*x[['V5']] + rnorm(1000, 0.001) + 3*sin(x[['V7']])
+
+train = as.matrix(x)
+
+# Interaction constraint list (column names form)
+interaction_list <- list(c('V1','V2'),c('V3','V4','V5'))
+
+# Convert interaction constraint list into feature index form
+cols2ids <- function(object, col_names) {
+  LUT <- seq_along(col_names) - 1
+  names(LUT) <- col_names
+  rapply(object, function(x) LUT[x], classes="character", how="replace")
+}
+interaction_list_fid = cols2ids(interaction_list, colnames(train))
+
+# Fit model with interaction constraints
+bst = xgboost(data = train, label = y, max_depth = 4,
+              eta = 0.1, nthread = 2, nrounds = 1000,
+              interaction_constraints = interaction_list_fid)
+
+bst_tree <- xgb.model.dt.tree(colnames(train), bst)
+bst_interactions <- treeInteractions(bst_tree, 4)  # interactions constrained to combinations of V1*V2 and V3*V4*V5
+
+# Fit model without interaction constraints
+bst2 = xgboost(data = train, label = y, max_depth = 4,
+               eta = 0.1, nthread = 2, nrounds = 1000)
+
+bst2_tree <- xgb.model.dt.tree(colnames(train), bst2)
+bst2_interactions <- treeInteractions(bst2_tree, 4)  # much more interactions
+
+# Fit model with both interaction and monotonicity constraints
+bst3 = xgboost(data = train, label = y, max_depth = 4,
+               eta = 0.1, nthread = 2, nrounds = 1000,
+               interaction_constraints = interaction_list_fid,
+               monotone_constraints = c(-1,0,0,0,0,0,0,0,0,0))
+
+bst3_tree <- xgb.model.dt.tree(colnames(train), bst3)
+bst3_interactions <- treeInteractions(bst3_tree, 4)  # interactions still constrained to combinations of V1*V2 and V3*V4*V5
+
+# Show monotonic constraints still apply by checking scores after incrementing V1
+x1 <- sort(unique(x[['V1']]))
+for (i in 1:length(x1)){
+  testdata <- copy(x[, -c('V1')])
+  testdata[['V1']] <- x1[i]
+  testdata <- testdata[, paste0('V',1:10), with=F]
+  pred <- predict(bst3, as.matrix(testdata))
+  
+  # Should not print out anything due to monotonic constraints
+  if (i > 1) if (any(pred > prev_pred)) print(i)
+  prev_pred <- pred 
+}

R-package/demo/predict_first_ntree.R

@@ -7,10 +7,10 @@ dtest <- xgb.DMatrix(agaricus.test$data, label = agaricus.test$label)
 
 param <- list(max_depth=2, eta=1, silent=1, objective='binary:logistic')
 watchlist <- list(eval = dtest, train = dtrain)
-nround = 2
+nrounds = 2
 
 # training the model for two rounds
-bst = xgb.train(param, dtrain, nround, nthread = 2, watchlist)
+bst = xgb.train(param, dtrain, nrounds, nthread = 2, watchlist)
 cat('start testing prediction from first n trees\n')
 labels <- getinfo(dtest,'label')
 

R-package/demo/predict_leaf_indices.R

@@ -11,10 +11,10 @@ dtrain <- xgb.DMatrix(data = agaricus.train$data, label = agaricus.train$label)
 dtest <- xgb.DMatrix(data = agaricus.test$data, label = agaricus.test$label)
 
 param <- list(max_depth=2, eta=1, silent=1, objective='binary:logistic')
-nround = 4
+nrounds = 4
 
 # training the model for two rounds
-bst = xgb.train(params = param, data = dtrain, nrounds = nround, nthread = 2)
+bst = xgb.train(params = param, data = dtrain, nrounds = nrounds, nthread = 2)
 
 # Model accuracy without new features
 accuracy.before <- sum((predict(bst, agaricus.test$data) >= 0.5) == agaricus.test$label) / length(agaricus.test$label)
@@ -43,7 +43,7 @@ new.features.test <- create.new.tree.features(bst, agaricus.test$data)
 new.dtrain <- xgb.DMatrix(data = new.features.train, label = agaricus.train$label)
 new.dtest <- xgb.DMatrix(data = new.features.test, label = agaricus.test$label)
 watchlist <- list(train = new.dtrain)
-bst <- xgb.train(params = param, data = new.dtrain, nrounds = nround, nthread = 2)
+bst <- xgb.train(params = param, data = new.dtrain, nrounds = nrounds, nthread = 2)
 
 # Model accuracy with new features
 accuracy.after <- sum((predict(bst, new.dtest) >= 0.5) == agaricus.test$label) / length(agaricus.test$label)

R-package/man/cb.reset.parameters.Rd

@@ -22,7 +22,7 @@ This is a "pre-iteration" callback function used to reset booster's parameters
 at the beginning of each iteration.
 
 Note that when training is resumed from some previous model, and a function is used to 
-reset a parameter value, the \code{nround} argument in this function would be the 
+reset a parameter value, the \code{nrounds} argument in this function would be the 
 the number of boosting rounds in the current training.
 
 Callback function expects the following values to be set in its calling frame:

R-package/man/predict.xgb.Booster.Rd

@@ -7,7 +7,8 @@
 \usage{
 \method{predict}{xgb.Booster}(object, newdata, missing = NA,
   outputmargin = FALSE, ntreelimit = NULL, predleaf = FALSE,
-  predcontrib = FALSE, approxcontrib = FALSE, reshape = FALSE, ...)
+  predcontrib = FALSE, approxcontrib = FALSE,
+  predinteraction = FALSE, reshape = FALSE, ...)
 
 \method{predict}{xgb.Booster.handle}(object, ...)
 }
@@ -26,14 +27,17 @@ logistic regression would result in predictions for log-odds instead of probabil
 \item{ntreelimit}{limit the number of model's trees or boosting iterations used in prediction (see Details).
 It will use all the trees by default (\code{NULL} value).}
 
-\item{predleaf}{whether predict leaf index instead.}
+\item{predleaf}{whether predict leaf index.}
 
-\item{predcontrib}{whether to return feature contributions to individual predictions instead (see Details).}
+\item{predcontrib}{whether to return feature contributions to individual predictions (see Details).}
 
 \item{approxcontrib}{whether to use a fast approximation for feature contributions (see Details).}
 
+\item{predinteraction}{whether to return contributions of feature interactions to individual predictions (see Details).}
+
 \item{reshape}{whether to reshape the vector of predictions to a matrix form when there are several
-prediction outputs per case. This option has no effect when \code{predleaf = TRUE}.}
+prediction outputs per case. This option has no effect when either of predleaf, predcontrib,
+or predinteraction flags is TRUE.}
 
 \item{...}{Parameters passed to \code{predict.xgb.Booster}}
 }
@@ -51,6 +55,14 @@ When \code{predcontrib = TRUE} and it is not a multiclass setting, the output is
 For a multiclass case, a list of \code{num_class} elements is returned, where each element is
 such a matrix. The contribution values are on the scale of untransformed margin
 (e.g., for binary classification would mean that the contributions are log-odds deviations from bias).
+
+When \code{predinteraction = TRUE} and it is not a multiclass setting, the output is a 3d array with
+dimensions \code{c(nrow, num_features + 1, num_features + 1)}. The off-diagonal (in the last two dimensions)
+elements represent different features interaction contributions. The array is symmetric WRT the last
+two dimensions. The "+ 1" columns corresponds to bias. Summing this array along the last dimension should
+produce practically the same result as predict with \code{predcontrib = TRUE}.
+For a multiclass case, a list of \code{num_class} elements is returned, where each element is
+such an array.
 }
 \description{
 Predicted values based on either xgboost model or model handle object.
@@ -76,6 +88,11 @@ values (Lundberg 2017) that sum to the difference between the expected output
 of the model and the current prediction (where the hessian weights are used to compute the expectations).
 Setting \code{approxcontrib = TRUE} approximates these values following the idea explained
 in \url{http://blog.datadive.net/interpreting-random-forests/}.
+
+With \code{predinteraction = TRUE}, SHAP values of contributions of interaction of each pair of features
+are computed. Note that this operation might be rather expensive in terms of compute and memory.
+Since it quadratically depends on the number of features, it is recommended to perfom selection
+of the most important features first. See below about the format of the returned results.
 }
 \examples{
 ## binary classification:

R-package/man/xgb.create.features.Rd

@@ -63,9 +63,9 @@ dtrain <- xgb.DMatrix(data = agaricus.train$data, label = agaricus.train$label)
 dtest <- xgb.DMatrix(data = agaricus.test$data, label = agaricus.test$label)
 
 param <- list(max_depth=2, eta=1, silent=1, objective='binary:logistic')
-nround = 4
+nrounds = 4
 
-bst = xgb.train(params = param, data = dtrain, nrounds = nround, nthread = 2)
+bst = xgb.train(params = param, data = dtrain, nrounds = nrounds, nthread = 2)
 
 # Model accuracy without new features
 accuracy.before <- sum((predict(bst, agaricus.test$data) >= 0.5) == agaricus.test$label) /
@@ -79,7 +79,7 @@ new.features.test <- xgb.create.features(model = bst, agaricus.test$data)
 new.dtrain <- xgb.DMatrix(data = new.features.train, label = agaricus.train$label)
 new.dtest <- xgb.DMatrix(data = new.features.test, label = agaricus.test$label)
 watchlist <- list(train = new.dtrain)
-bst <- xgb.train(params = param, data = new.dtrain, nrounds = nround, nthread = 2)
+bst <- xgb.train(params = param, data = new.dtrain, nrounds = nrounds, nthread = 2)
 
 # Model accuracy with new features
 accuracy.after <- sum((predict(bst, new.dtest) >= 0.5) == agaricus.test$label) /

R-package/man/xgb.cv.Rd

@@ -4,11 +4,12 @@
 \alias{xgb.cv}
 \title{Cross Validation}
 \usage{
-xgb.cv(params = list(), data, nrounds, nfold, label = NULL, missing = NA,
-  prediction = FALSE, showsd = TRUE, metrics = list(), obj = NULL,
-  feval = NULL, stratified = TRUE, folds = NULL, verbose = TRUE,
-  print_every_n = 1L, early_stopping_rounds = NULL, maximize = NULL,
-  callbacks = list(), ...)
+xgb.cv(params = list(), data, nrounds, nfold, label = NULL,
+  missing = NA, prediction = FALSE, showsd = TRUE,
+  metrics = list(), obj = NULL, feval = NULL, stratified = TRUE,
+  folds = NULL, verbose = TRUE, print_every_n = 1L,
+  early_stopping_rounds = NULL, maximize = NULL, callbacks = list(),
+  ...)
 }
 \arguments{
 \item{params}{the list of parameters. Commonly used ones are:

R-package/man/xgb.dump.Rd

@@ -44,8 +44,8 @@ test <- agaricus.test
 bst <- xgboost(data = train$data, label = train$label, max_depth = 2, 
                eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic")
 # save the model in file 'xgb.model.dump'
-dump.path = file.path(tempdir(), 'model.dump')
-xgb.dump(bst, dump.path, with_stats = TRUE)
+dump_path = file.path(tempdir(), 'model.dump')
+xgb.dump(bst, dump_path, with_stats = TRUE)
 
 # print the model without saving it to a file
 print(xgb.dump(bst, with_stats = TRUE))

R-package/man/xgb.plot.deepness.Rd

@@ -5,11 +5,11 @@
 \alias{xgb.plot.deepness}
 \title{Plot model trees deepness}
 \usage{
-xgb.ggplot.deepness(model = NULL, which = c("2x1", "max.depth", "med.depth",
-  "med.weight"))
+xgb.ggplot.deepness(model = NULL, which = c("2x1", "max.depth",
+  "med.depth", "med.weight"))
 
-xgb.plot.deepness(model = NULL, which = c("2x1", "max.depth", "med.depth",
-  "med.weight"), plot = TRUE, ...)
+xgb.plot.deepness(model = NULL, which = c("2x1", "max.depth",
+  "med.depth", "med.weight"), plot = TRUE, ...)
 }
 \arguments{
 \item{model}{either an \code{xgb.Booster} model generated by the \code{xgb.train} function
@@ -50,7 +50,7 @@ per tree with respect to tree number are created. And \code{which="med.weight"}
 a tree's median absolute leaf weight changes through the iterations.
 
 This function was inspired by the blog post
-\url{http://aysent.github.io/2015/11/08/random-forest-leaf-visualization.html}.
+\url{https://github.com/aysent/random-forest-leaf-visualization}.
 }
 \examples{
 

R-package/man/xgb.plot.importance.Rd

@@ -9,8 +9,8 @@ xgb.ggplot.importance(importance_matrix = NULL, top_n = NULL,
   measure = NULL, rel_to_first = FALSE, n_clusters = c(1:10), ...)
 
 xgb.plot.importance(importance_matrix = NULL, top_n = NULL,
-  measure = NULL, rel_to_first = FALSE, left_margin = 10, cex = NULL,
-  plot = TRUE, ...)
+  measure = NULL, rel_to_first = FALSE, left_margin = 10,
+  cex = NULL, plot = TRUE, ...)
 }
 \arguments{
 \item{importance_matrix}{a \code{data.table} returned by \code{\link{xgb.importance}}.}

R-package/man/xgb.plot.shap.Rd

@@ -6,8 +6,8 @@
 \usage{
 xgb.plot.shap(data, shap_contrib = NULL, features = NULL, top_n = 1,
   model = NULL, trees = NULL, target_class = NULL,
-  approxcontrib = FALSE, subsample = NULL, n_col = 1, col = rgb(0, 0, 1,
-  0.2), pch = ".", discrete_n_uniq = 5, discrete_jitter = 0.01,
+  approxcontrib = FALSE, subsample = NULL, n_col = 1, col = rgb(0,
+  0, 1, 0.2), pch = ".", discrete_n_uniq = 5, discrete_jitter = 0.01,
   ylab = "SHAP", plot_NA = TRUE, col_NA = rgb(0.7, 0, 1, 0.6),
   pch_NA = ".", pos_NA = 1.07, plot_loess = TRUE, col_loess = 2,
   span_loess = 0.5, which = c("1d", "2d"), plot = TRUE, ...)

R-package/man/xgb.train.Rd

@@ -5,15 +5,17 @@
 \alias{xgboost}
 \title{eXtreme Gradient Boosting Training}
 \usage{
-xgb.train(params = list(), data, nrounds, watchlist = list(), obj = NULL,
-  feval = NULL, verbose = 1, print_every_n = 1L,
+xgb.train(params = list(), data, nrounds, watchlist = list(),
+  obj = NULL, feval = NULL, verbose = 1, print_every_n = 1L,
   early_stopping_rounds = NULL, maximize = NULL, save_period = NULL,
-  save_name = "xgboost.model", xgb_model = NULL, callbacks = list(), ...)
+  save_name = "xgboost.model", xgb_model = NULL, callbacks = list(),
+  ...)
 
 xgboost(data = NULL, label = NULL, missing = NA, weight = NULL,
   params = list(), nrounds, verbose = 1, print_every_n = 1L,
   early_stopping_rounds = NULL, maximize = NULL, save_period = NULL,
-  save_name = "xgboost.model", xgb_model = NULL, callbacks = list(), ...)
+  save_name = "xgboost.model", xgb_model = NULL, callbacks = list(),
+  ...)
 }
 \arguments{
 \item{params}{the list of parameters. 
@@ -35,7 +37,7 @@ xgboost(data = NULL, label = NULL, missing = NA, weight = NULL,
   \item \code{gamma} minimum loss reduction required to make a further partition on a leaf node of the tree. the larger, the more conservative the algorithm will be. 
   \item \code{max_depth} maximum depth of a tree. Default: 6
   \item \code{min_child_weight} minimum sum of instance weight (hessian) needed in a child. If the tree partition step results in a leaf node with the sum of instance weight less than min_child_weight, then the building process will give up further partitioning. In linear regression mode, this simply corresponds to minimum number of instances needed to be in each node. The larger, the more conservative the algorithm will be. Default: 1
-  \item \code{subsample} subsample ratio of the training instance. Setting it to 0.5 means that xgboost randomly collected half of the data instances to grow trees and this will prevent overfitting. It makes computation shorter (because less data to analyse). It is advised to use this parameter with \code{eta} and increase \code{nround}. Default: 1 
+  \item \code{subsample} subsample ratio of the training instance. Setting it to 0.5 means that xgboost randomly collected half of the data instances to grow trees and this will prevent overfitting. It makes computation shorter (because less data to analyse). It is advised to use this parameter with \code{eta} and increase \code{nrounds}. Default: 1 
   \item \code{colsample_bytree} subsample ratio of columns when constructing each tree. Default: 1
   \item \code{num_parallel_tree} Experimental parameter. number of trees to grow per round. Useful to test Random Forest through Xgboost (set \code{colsample_bytree < 1}, \code{subsample  < 1}  and \code{round = 1}) accordingly. Default: 1
   \item \code{monotone_constraints} A numerical vector consists of \code{1}, \code{0} and \code{-1} with its length equals to the number of features in the training data. \code{1} is increasing, \code{-1} is decreasing and \code{0} is no constraint.

R-package/src/Makevars.in

@@ -12,13 +12,13 @@ XGB_RFLAGS = -DXGBOOST_STRICT_R_MODE=1 -DDMLC_LOG_BEFORE_THROW=0\
 
 # disable the use of thread_local for 32 bit windows:
 ifeq ($(R_OSTYPE)$(WIN),windows)
-    XGB_RFLAGS += -DDMLC_CXX11_THREAD_LOCAL=0 -msse2 -mfpmath=sse
+    XGB_RFLAGS += -DDMLC_CXX11_THREAD_LOCAL=0
 endif
 $(foreach v, $(XGB_RFLAGS), $(warning $(v)))
 
 PKG_CPPFLAGS=  -I$(PKGROOT)/include -I$(PKGROOT)/dmlc-core/include -I$(PKGROOT)/rabit/include -I$(PKGROOT) $(XGB_RFLAGS)
-PKG_CXXFLAGS= @OPENMP_CXXFLAGS@ $(SHLIB_PTHREAD_FLAGS)
-PKG_LIBS = @OPENMP_CXXFLAGS@ $(SHLIB_PTHREAD_FLAGS)
+PKG_CXXFLAGS= @OPENMP_CXXFLAGS@ -pthread
+PKG_LIBS = @OPENMP_CXXFLAGS@ -pthread
 OBJECTS= ./xgboost_R.o ./xgboost_custom.o ./xgboost_assert.o ./init.o\
          $(PKGROOT)/amalgamation/xgboost-all0.o $(PKGROOT)/amalgamation/dmlc-minimum0.o\
          $(PKGROOT)/rabit/src/engine_empty.o $(PKGROOT)/rabit/src/c_api.o

R-package/src/Makevars.win

@@ -24,13 +24,13 @@ XGB_RFLAGS = -DXGBOOST_STRICT_R_MODE=1 -DDMLC_LOG_BEFORE_THROW=0\
 
 # disable the use of thread_local for 32 bit windows:
 ifeq ($(R_OSTYPE)$(WIN),windows)
-    XGB_RFLAGS += -DDMLC_CXX11_THREAD_LOCAL=0 -msse2 -mfpmath=sse
+    XGB_RFLAGS += -DDMLC_CXX11_THREAD_LOCAL=0
 endif
 $(foreach v, $(XGB_RFLAGS), $(warning $(v)))
 
 PKG_CPPFLAGS=  -I$(PKGROOT)/include -I$(PKGROOT)/dmlc-core/include -I$(PKGROOT)/rabit/include -I$(PKGROOT) $(XGB_RFLAGS)
-PKG_CXXFLAGS= $(SHLIB_OPENMP_CFLAGS) $(SHLIB_PTHREAD_FLAGS)
-PKG_LIBS = $(SHLIB_OPENMP_CFLAGS) $(SHLIB_PTHREAD_FLAGS)
+PKG_CXXFLAGS= $(SHLIB_OPENMP_CXXFLAGS) $(SHLIB_PTHREAD_FLAGS)
+PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS) $(SHLIB_PTHREAD_FLAGS)
 OBJECTS= ./xgboost_R.o ./xgboost_custom.o ./xgboost_assert.o ./init.o\
          $(PKGROOT)/amalgamation/xgboost-all0.o $(PKGROOT)/amalgamation/dmlc-minimum0.o\
          $(PKGROOT)/rabit/src/engine_empty.o $(PKGROOT)/rabit/src/c_api.o

R-package/src/init.c

@@ -1,5 +1,5 @@
 /* Copyright (c) 2015 by Contributors
- * 
+ *
  * This file was initially generated using the following R command:
  * tools::package_native_routine_registration_skeleton('.', con = 'src/init.c', character_only = F)
  * and edited to conform to xgboost C linter requirements. For details, see
@@ -10,7 +10,7 @@
 #include <stdlib.h>
 #include <R_ext/Rdynload.h>
 
-/* FIXME: 
+/* FIXME:
 Check these declarations against the C/Fortran source code.
 */
 
@@ -70,7 +70,7 @@ static const R_CallMethodDef CallEntries[] = {
 
 #if defined(_WIN32)
 __declspec(dllexport)
-#endif
+#endif  // defined(_WIN32)
 void R_init_xgboost(DllInfo *dll) {
   R_registerRoutines(dll, NULL, CallEntries, NULL, NULL);
   R_useDynamicSymbols(dll, FALSE);

R-package/src/xgboost_custom.cc

@@ -32,7 +32,10 @@ extern "C" {
 
 namespace xgboost {
 ConsoleLogger::~ConsoleLogger() {
-  dmlc::CustomLogMessage::Log(log_stream_.str());
+  if (cur_verbosity_ == LogVerbosity::kIgnore ||
+      cur_verbosity_ <= global_verbosity_) {
+    dmlc::CustomLogMessage::Log(log_stream_.str());
+  }
 }
 TrackerLogger::~TrackerLogger() {
   dmlc::CustomLogMessage::Log(log_stream_.str());
@@ -46,10 +49,11 @@ namespace common {
 bool CheckNAN(double v) {
   return ISNAN(v);
 }
+#if !defined(XGBOOST_USE_CUDA)
 double LogGamma(double v) {
   return lgammafn(v);
 }
-
+#endif  // !defined(XGBOOST_USE_CUDA)
 // customize random engine.
 void CustomGlobalRandomEngine::seed(CustomGlobalRandomEngine::result_type val) {
   // ignore the seed

R-package/tests/testthat/test_basic.R

@@ -182,7 +182,7 @@ test_that("xgb.cv works", {
   expect_is(cv, 'xgb.cv.synchronous')
   expect_false(is.null(cv$evaluation_log))
   expect_lt(cv$evaluation_log[, min(test_error_mean)], 0.03)
-  expect_lt(cv$evaluation_log[, min(test_error_std)], 0.004)
+  expect_lt(cv$evaluation_log[, min(test_error_std)], 0.008)
   expect_equal(cv$niter, 2)
   expect_false(is.null(cv$folds) && is.list(cv$folds))
   expect_length(cv$folds, 5)
@@ -223,3 +223,42 @@ test_that("train and predict with non-strict classes", {
   expect_error(pr <- predict(bst, train_dense), regexp = NA)
   expect_equal(pr0, pr)
 })
+
+test_that("max_delta_step works", {
+  dtrain <- xgb.DMatrix(agaricus.train$data, label = agaricus.train$label)
+  watchlist <- list(train = dtrain)
+  param <- list(objective = "binary:logistic", eval_metric="logloss", max_depth = 2, nthread = 2, eta = 0.5)
+  nrounds = 5
+  # model with no restriction on max_delta_step
+  bst1 <- xgb.train(param, dtrain, nrounds, watchlist, verbose = 1)
+  # model with restricted max_delta_step
+  bst2 <- xgb.train(param, dtrain, nrounds, watchlist, verbose = 1, max_delta_step = 1)
+  # the no-restriction model is expected to have consistently lower loss during the initial interations
+  expect_true(all(bst1$evaluation_log$train_logloss < bst2$evaluation_log$train_logloss))
+  expect_lt(mean(bst1$evaluation_log$train_logloss)/mean(bst2$evaluation_log$train_logloss), 0.8)
+})
+
+test_that("colsample_bytree works", {
+  # Randomly generate data matrix by sampling from uniform distribution [-1, 1]
+  set.seed(1)
+  train_x <- matrix(runif(1000, min = -1, max = 1), ncol = 100)
+  train_y <- as.numeric(rowSums(train_x) > 0)
+  test_x <- matrix(runif(1000, min = -1, max = 1), ncol = 100)
+  test_y <- as.numeric(rowSums(test_x) > 0)
+  colnames(train_x) <- paste0("Feature_", sprintf("%03d", 1:100))
+  colnames(test_x) <- paste0("Feature_", sprintf("%03d", 1:100))
+   dtrain <- xgb.DMatrix(train_x, label = train_y)
+  dtest <- xgb.DMatrix(test_x, label = test_y)
+  watchlist <- list(train = dtrain, eval = dtest)
+   # Use colsample_bytree = 0.01, so that roughly one out of 100 features is
+  # chosen for each tree
+  param <- list(max_depth = 2, eta = 0, silent = 1, nthread = 2,
+                colsample_bytree = 0.01, objective = "binary:logistic",
+                eval_metric = "auc")
+   set.seed(2)
+  bst <- xgb.train(param, dtrain, nrounds = 100, watchlist, verbose = 0)
+  xgb.importance(model = bst)
+  # If colsample_bytree works properly, a variety of features should be used
+  # in the 100 trees
+  expect_gte(nrow(xgb.importance(model = bst)), 30)
+})

R-package/tests/testthat/test_callbacks.R

@@ -282,7 +282,7 @@ test_that("prediction in xgb.cv works for gblinear too", {
 })
 
 test_that("prediction in early-stopping xgb.cv works", {
-  set.seed(1)
+  set.seed(11)
   expect_output(
     cv <- xgb.cv(param, dtrain, nfold = 5, eta = 0.1, nrounds = 20,
                  early_stopping_rounds = 5, maximize = FALSE, prediction = TRUE)

R-package/tests/testthat/test_gc_safety.R

@@ -9,7 +9,7 @@ test_that("train and prediction when gctorture is on", {
   test <- agaricus.test
   gctorture(TRUE)
   bst <- xgboost(data = train$data, label = train$label, max.depth = 2,
-  eta = 1, nthread = 2, nround = 2, objective = "binary:logistic")
+  eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic")
   pred <- predict(bst, test$data)
   gctorture(FALSE)
 })

R-package/tests/testthat/test_helpers.R

@@ -7,6 +7,9 @@ require(vcd, quietly = TRUE)
 
 float_tolerance = 5e-6
 
+# disable some tests for Win32
+win32_flag = .Platform$OS.type == "windows" && .Machine$sizeof.pointer != 8
+
 set.seed(1982)
 data(Arthritis)
 df <- data.table(Arthritis, keep.rownames = F)
@@ -41,7 +44,8 @@ mbst.GLM <- xgboost(data = as.matrix(iris[, -5]), label = mlabel, verbose = 0,
 
 
 test_that("xgb.dump works", {
-  expect_length(xgb.dump(bst.Tree), 200)
+  if (!win32_flag)
+    expect_length(xgb.dump(bst.Tree), 200)
   dump_file = file.path(tempdir(), 'xgb.model.dump')
   expect_true(xgb.dump(bst.Tree, dump_file, with_stats = T))
   expect_true(file.exists(dump_file))
@@ -50,7 +54,8 @@ test_that("xgb.dump works", {
   # JSON format
   dmp <- xgb.dump(bst.Tree, dump_format = "json")
   expect_length(dmp, 1)
-  expect_length(grep('nodeid', strsplit(dmp, '\n')[[1]]), 188)
+  if (!win32_flag)
+    expect_length(grep('nodeid', strsplit(dmp, '\n')[[1]]), 188)
 })
 
 test_that("xgb.dump works for gblinear", {
@@ -210,7 +215,8 @@ test_that("xgb.model.dt.tree works with and without feature names", {
   names.dt.trees <- c("Tree", "Node", "ID", "Feature", "Split", "Yes", "No", "Missing", "Quality", "Cover")
   dt.tree <- xgb.model.dt.tree(feature_names = feature.names, model = bst.Tree)
   expect_equal(names.dt.trees, names(dt.tree))
-  expect_equal(dim(dt.tree), c(188, 10))
+  if (!win32_flag)
+    expect_equal(dim(dt.tree), c(188, 10))
   expect_output(str(dt.tree), 'Feature.*\\"Age\\"')
 
   dt.tree.0 <- xgb.model.dt.tree(model = bst.Tree)
@@ -236,7 +242,8 @@ test_that("xgb.model.dt.tree throws error for gblinear", {
 
 test_that("xgb.importance works with and without feature names", {
   importance.Tree <- xgb.importance(feature_names = feature.names, model = bst.Tree)
-  expect_equal(dim(importance.Tree), c(7, 4))
+  if (!win32_flag)
+    expect_equal(dim(importance.Tree), c(7, 4))
   expect_equal(colnames(importance.Tree), c("Feature", "Gain", "Cover", "Frequency"))
   expect_output(str(importance.Tree), 'Feature.*\\"Age\\"')
 

R-package/tests/testthat/test_interaction_constraints.R

@@ -0,0 +1,38 @@
+require(xgboost)
+
+context("interaction constraints")
+
+set.seed(1024)
+x1 <- rnorm(1000, 1)
+x2 <- rnorm(1000, 1)
+x3 <- sample(c(1,2,3), size=1000, replace=TRUE)
+y <- x1 + x2 + x3 + x1*x2*x3 + rnorm(1000, 0.001) + 3*sin(x1)
+train <- matrix(c(x1,x2,x3), ncol = 3)
+
+test_that("interaction constraints for regression", {
+  # Fit a model that only allows interaction between x1 and x2
+  bst <- xgboost(data = train, label = y, max_depth = 3,
+                 eta = 0.1, nthread = 2, nrounds = 100, verbose = 0,
+                 interaction_constraints = list(c(0,1)))
+  
+  # Set all observations to have the same x3 values then increment
+  #  by the same amount
+	preds <- lapply(c(1,2,3), function(x){
+		tmat <- matrix(c(x1,x2,rep(x,1000)), ncol=3)
+		return(predict(bst, tmat))
+	})
+
+  # Check incrementing x3 has the same effect on all observations
+  #   since x3 is constrained to be independent of x1 and x2
+  #   and all observations start off from the same x3 value
+	diff1 <- preds[[2]] - preds[[1]]
+	test1 <- all(abs(diff1 - diff1[1]) < 1e-4)
+	
+	diff2 <- preds[[3]] - preds[[2]]
+	test2 <- all(abs(diff2 - diff2[1]) < 1e-4)
+	
+  expect_true({
+    test1 & test2
+  }, "Interaction Contraint Satisfied")
+  
+})

R-package/tests/testthat/test_interactions.R

@@ -0,0 +1,141 @@
+context('Test prediction of feature interactions')
+
+require(xgboost)
+require(magrittr)
+
+set.seed(123)
+
+test_that("predict feature interactions works", {
+  # simulate some binary data and a linear outcome with an interaction term
+  N <- 1000
+  P <- 5
+  X <- matrix(rbinom(N * P, 1, 0.5), ncol=P, dimnames = list(NULL, letters[1:P]))
+  # center the data (as contributions are computed WRT feature means)
+  X <- scale(X, scale=FALSE)
+
+  # outcome without any interactions, without any noise:
+  f <- function(x) 2 * x[, 1] - 3 * x[, 2]
+  # outcome with interactions, without noise:
+  f_int <- function(x) f(x) + 2 * x[, 2] * x[, 3]
+  # outcome with interactions, with noise:
+  #f_int_noise <- function(x) f_int(x) + rnorm(N, 0, 0.3)
+
+  y <- f_int(X)
+
+  dm <- xgb.DMatrix(X, label = y)
+  param <- list(eta=0.1, max_depth=4, base_score=mean(y), lambda=0, nthread=2)
+  b <- xgb.train(param, dm, 100)
+  
+  pred = predict(b, dm, outputmargin=TRUE)
+
+  # SHAP contributions:
+  cont <- predict(b, dm, predcontrib=TRUE)
+  expect_equal(dim(cont), c(N, P+1))
+  # make sure for each row they add up to marginal predictions
+  max(abs(rowSums(cont) - pred)) %>% expect_lt(0.001)
+  # Hand-construct the 'ground truth' feature contributions:
+  gt_cont <- cbind(
+      2. * X[, 1],
+     -3. * X[, 2] + 1. * X[, 2] * X[, 3], # attribute a HALF of the interaction term to feature #2
+      1. * X[, 2] * X[, 3]               # and another HALF of the interaction term to feature #3
+     )
+  gt_cont <- cbind(gt_cont, matrix(0, nrow=N, ncol=P + 1 - 3))
+  # These should be relatively close:
+  expect_lt(max(abs(cont - gt_cont)), 0.05)
+
+
+  # SHAP interaction contributions:
+  intr <- predict(b, dm, predinteraction=TRUE)
+  expect_equal(dim(intr), c(N, P+1, P+1))
+  # check assigned colnames
+  cn <- c(letters[1:P], "BIAS")
+  expect_equal(dimnames(intr), list(NULL, cn, cn))
+
+  # check the symmetry
+  max(abs(aperm(intr, c(1,3,2)) - intr)) %>% expect_lt(0.00001)
+
+  # sums WRT columns must be close to feature contributions
+  max(abs(apply(intr, c(1,2), sum) - cont)) %>% expect_lt(0.00001)
+
+  # diagonal terms for features 3,4,5 must be close to zero
+  Reduce(max, sapply(3:P, function(i) max(abs(intr[, i, i])))) %>% expect_lt(0.05)
+
+  # BIAS must have no interactions
+  max(abs(intr[, 1:P, P+1])) %>% expect_lt(0.00001)
+
+  # interactions other than 2 x 3 must be close to zero
+  intr23 <- intr
+  intr23[,2,3] <- 0
+  Reduce(max, sapply(1:P, function(i) max(abs(intr23[, i, (i+1):(P+1)])))) %>% expect_lt(0.05)
+
+  # Construct the 'ground truth' contributions of interactions directly from the linear terms:
+  gt_intr <- array(0, c(N, P+1, P+1))
+  gt_intr[,2,3] <- 1. * X[, 2] * X[, 3] # attribute a HALF of the interaction term to each symmetric element
+  gt_intr[,3,2] <- gt_intr[, 2, 3]
+  # merge-in the diagonal based on 'ground truth' feature contributions
+  intr_diag = gt_cont - apply(gt_intr, c(1,2), sum)
+  for(j in seq_len(P)) {
+    gt_intr[,j,j] = intr_diag[,j]
+  }
+  # These should be relatively close:
+  expect_lt(max(abs(intr - gt_intr)), 0.1)
+})
+
+test_that("SHAP contribution values are not NAN", {
+  d <- data.frame(
+    x1 = c(-2.3, 1.4, 5.9, 2, 2.5, 0.3, -3.6, -0.2, 0.5, -2.8, -4.6, 3.3, -1.2,
+           -1.1, -2.3, 0.4, -1.5, -0.2, -1, 3.7),
+    x2 = c(291.179171, 269.198331, 289.942097, 283.191669, 269.673332,
+           294.158346, 287.255835, 291.530838, 285.899586, 269.290833,
+           268.649586, 291.530841, 280.074593, 269.484168, 293.94042,
+           294.327506, 296.20709, 295.441669, 283.16792, 270.227085),
+    y = c(9, 15, 5.7, 9.2, 22.4, 5, 9, 3.2, 7.2, 13.1, 7.8, 16.9, 6.5, 22.1,
+          5.3, 10.4, 11.1, 13.9, 11, 20.5),
+    fold = c(2, 2, 2, 1, 2, 2, 1, 2, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2))
+
+  ivs <- c("x1", "x2")
+
+  fit <- xgboost(
+    verbose = 0,
+    params = list(
+      objective = "reg:linear",
+      eval_metric = "rmse"),
+    data = as.matrix(subset(d, fold == 2)[, ivs]),
+    label = subset(d, fold == 2)$y,
+    nthread = 1,
+    nrounds = 3)
+
+  shaps <- as.data.frame(predict(fit,
+    newdata = as.matrix(subset(d, fold == 1)[, ivs]),
+    predcontrib = T))
+  result <- cbind(shaps, sum = rowSums(shaps), pred = predict(fit,
+      newdata = as.matrix(subset(d, fold == 1)[, ivs])))
+
+  expect_true(identical(TRUE, all.equal(result$sum, result$pred, tol = 1e-6)))
+})
+
+
+test_that("multiclass feature interactions work", {
+  dm <- xgb.DMatrix(as.matrix(iris[,-5]), label=as.numeric(iris$Species)-1)
+  param <- list(eta=0.1, max_depth=4, objective='multi:softprob', num_class=3)
+  b <- xgb.train(param, dm, 40)
+  pred = predict(b, dm, outputmargin=TRUE) %>% array(c(3, 150)) %>% t
+
+  # SHAP contributions:
+  cont <- predict(b, dm, predcontrib=TRUE)
+  expect_length(cont, 3)
+  # rewrap them as a 3d array
+  cont <- unlist(cont) %>% array(c(150, 5, 3))
+  # make sure for each row they add up to marginal predictions
+  max(abs(apply(cont, c(1,3), sum) - pred)) %>% expect_lt(0.001)
+
+  # SHAP interaction contributions:
+  intr <- predict(b, dm, predinteraction=TRUE)
+  expect_length(intr, 3)
+  # rewrap them as a 4d array
+  intr <- unlist(intr) %>% array(c(150, 5, 5, 3)) %>% aperm(c(4, 1, 2, 3)) # [grp, row, col, col]
+  # check the symmetry
+  max(abs(aperm(intr, c(1,2,4,3)) - intr)) %>% expect_lt(0.00001)
+  # sums WRT columns must be close to feature contributions
+  max(abs(apply(intr, c(1,2,3), sum) - aperm(cont, c(3,1,2)))) %>% expect_lt(0.00001)
+})

R-package/tests/testthat/test_update.R

@@ -7,6 +7,10 @@ data(agaricus.test, package = 'xgboost')
 dtrain <- xgb.DMatrix(agaricus.train$data, label = agaricus.train$label)
 dtest <- xgb.DMatrix(agaricus.test$data, label = agaricus.test$label)
 
+# Disable flaky tests for 32-bit Windows.
+# See https://github.com/dmlc/xgboost/issues/3720
+win32_flag = .Platform$OS.type == "windows" && .Machine$sizeof.pointer != 8
+
 test_that("updating the model works", {
   watchlist = list(train = dtrain, test = dtest)
 
@@ -29,7 +33,9 @@ test_that("updating the model works", {
   tr1r <- xgb.model.dt.tree(model = bst1r)
   # all should be the same when no subsampling
   expect_equal(bst1$evaluation_log, bst1r$evaluation_log)
-  expect_equal(tr1, tr1r, tolerance = 0.00001, check.attributes = FALSE)
+  if (!win32_flag) {
+    expect_equal(tr1, tr1r, tolerance = 0.00001, check.attributes = FALSE)
+  }
 
   # the same boosting with subsampling with an extra 'refresh' updater:
   p2r <- modifyList(p2, list(updater = 'grow_colmaker,prune,refresh', refresh_leaf = FALSE))
@@ -38,7 +44,9 @@ test_that("updating the model works", {
   tr2r <- xgb.model.dt.tree(model = bst2r)
   # should be the same evaluation but different gains and larger cover
   expect_equal(bst2$evaluation_log, bst2r$evaluation_log)
-  expect_equal(tr2[Feature == 'Leaf']$Quality, tr2r[Feature == 'Leaf']$Quality)
+  if (!win32_flag) {
+    expect_equal(tr2[Feature == 'Leaf']$Quality, tr2r[Feature == 'Leaf']$Quality)
+  }
   expect_gt(sum(abs(tr2[Feature != 'Leaf']$Quality - tr2r[Feature != 'Leaf']$Quality)), 100)
   expect_gt(sum(tr2r$Cover) / sum(tr2$Cover), 1.5)
 
@@ -61,7 +69,9 @@ test_that("updating the model works", {
   expect_gt(sum(tr2u$Cover) / sum(tr2$Cover), 1.5)
   # the results should be the same as for the model with an extra 'refresh' updater
   expect_equal(bst2r$evaluation_log, bst2u$evaluation_log)
-  expect_equal(tr2r, tr2u, tolerance = 0.00001, check.attributes = FALSE)
+  if (!win32_flag) {
+    expect_equal(tr2r, tr2u, tolerance = 0.00001, check.attributes = FALSE)
+  }
   
   # process type 'update' for no-subsampling model, refreshing only the tree stats from TEST data:
   p1ut <- modifyList(p1, list(process_type = 'update', updater = 'refresh', refresh_leaf = FALSE))

README.md

@@ -1,5 +1,6 @@
 <img src=https://raw.githubusercontent.com/dmlc/dmlc.github.io/master/img/logo-m/xgboost.png width=135/>  eXtreme Gradient Boosting
 ===========
+[![Build Status](https://xgboost-ci.net/job/xgboost/job/master/badge/icon?style=plastic)](https://xgboost-ci.net/blue/organizations/jenkins/xgboost/activity)
 [![Build Status](https://travis-ci.org/dmlc/xgboost.svg?branch=master)](https://travis-ci.org/dmlc/xgboost)
 [![Build Status](https://ci.appveyor.com/api/projects/status/5ypa8vaed6kpmli8?svg=true)](https://ci.appveyor.com/project/tqchen/xgboost)
 [![Documentation Status](https://readthedocs.org/projects/xgboost/badge/?version=latest)](https://xgboost.readthedocs.org)

amalgamation/xgboost-all0.cc

@@ -48,7 +48,7 @@
 #include "../src/tree/tree_model.cc"
 #include "../src/tree/tree_updater.cc"
 #include "../src/tree/updater_colmaker.cc"
-#include "../src/tree/updater_fast_hist.cc"
+#include "../src/tree/updater_quantile_hist.cc"
 #include "../src/tree/updater_prune.cc"
 #include "../src/tree/updater_refresh.cc"
 #include "../src/tree/updater_sync.cc"

appveyor.yml

@@ -44,12 +44,12 @@ install:
     - set DO_PYTHON=off
     - if /i "%target%" == "mingw" set DO_PYTHON=on
     - if /i "%target%_%ver%_%configuration%" == "msvc_2015_Release" set DO_PYTHON=on
-    - if /i "%DO_PYTHON%" == "on" conda install -y numpy scipy pandas matplotlib nose scikit-learn graphviz python-graphviz
+    - if /i "%DO_PYTHON%" == "on" conda install -y numpy scipy pandas matplotlib pytest scikit-learn graphviz python-graphviz
     # R: based on https://github.com/krlmlr/r-appveyor
     - ps: |
         if($env:target -eq 'rmingw' -or $env:target -eq 'rmsvc') {
           #$ErrorActionPreference = "Stop"
-          Invoke-WebRequest http://raw.github.com/krlmlr/r-appveyor/master/scripts/appveyor-tool.ps1 -OutFile "$Env:TEMP\appveyor-tool.ps1"
+          Invoke-WebRequest https://raw.githubusercontent.com/krlmlr/r-appveyor/master/scripts/appveyor-tool.ps1 -OutFile "$Env:TEMP\appveyor-tool.ps1"
           Import-Module "$Env:TEMP\appveyor-tool.ps1"
           Bootstrap
           $DEPS = "c('data.table','magrittr','stringi','ggplot2','DiagrammeR','Ckmeans.1d.dp','vcd','testthat','lintr','knitr','rmarkdown')"
@@ -96,7 +96,7 @@ build_script:
 
 test_script:
     - cd %APPVEYOR_BUILD_FOLDER%
-    - if /i "%DO_PYTHON%" == "on" python -m nose tests/python
+    - if /i "%DO_PYTHON%" == "on" python -m pytest tests/python
     # mingw R package: run the R check (which includes unit tests), and also keep the built binary package
     - if /i "%target%" == "rmingw" (
         set _R_CHECK_CRAN_INCOMING_=FALSE&&

cmake/build_config.h.in

@@ -0,0 +1,11 @@
+/*!
+ * Copyright 2019 by Contributors
+ * \file build_config.h
+ */
+#ifndef XGBOOST_BUILD_CONFIG_H_
+#define XGBOOST_BUILD_CONFIG_H_
+
+#cmakedefine XGBOOST_MM_PREFETCH_PRESENT
+#cmakedefine XGBOOST_BUILTIN_PREFETCH_PRESENT
+
+#endif  // XGBOOST_BUILD_CONFIG_H_

cmake/modules/FindASan.cmake

@@ -1,8 +1,8 @@
 set(ASan_LIB_NAME ASan)
 
 find_library(ASan_LIBRARY
-  NAMES libasan.so libasan.so.4
-  PATHS /usr/lib64 /usr/lib /usr/local/lib64 /usr/local/lib)
+  NAMES libasan.so libasan.so.4 libasan.so.3 libasan.so.2 libasan.so.1 libasan.so.0
+  PATHS ${SANITIZER_PATH} /usr/lib64 /usr/lib /usr/local/lib64 /usr/local/lib ${CMAKE_PREFIX_PATH}/lib)
 
 include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(ASan DEFAULT_MSG

cmake/modules/FindLSan.cmake

@@ -2,7 +2,7 @@ set(LSan_LIB_NAME lsan)
 
 find_library(LSan_LIBRARY
   NAMES liblsan.so liblsan.so.0 liblsan.so.0.0.0
-  PATHS /usr/lib64 /usr/lib /usr/local/lib64 /usr/local/lib)
+  PATHS ${SANITIZER_PATH} /usr/lib64 /usr/lib /usr/local/lib64 /usr/local/lib ${CMAKE_PREFIX_PATH}/lib)
 
 include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(LSan DEFAULT_MSG

cmake/modules/FindTSan.cmake

@@ -2,7 +2,7 @@ set(TSan_LIB_NAME tsan)
 
 find_library(TSan_LIBRARY
   NAMES libtsan.so libtsan.so.0 libtsan.so.0.0.0
-  PATHS /usr/lib64 /usr/lib /usr/local/lib64 /usr/local/lib)
+  PATHS ${SANITIZER_PATH} /usr/lib64 /usr/lib /usr/local/lib64 /usr/local/lib ${CMAKE_PREFIX_PATH}/lib)
 
 include(FindPackageHandleStandardArgs)
 find_package_handle_standard_args(TSan DEFAULT_MSG

demo/README.md

@@ -135,6 +135,7 @@ Send a PR to add a one sentence description:)
 
 ## Awards
 - [John Chambers Award](http://stat-computing.org/awards/jmc/winners.html) - 2016 Winner: XGBoost R Package, by Tong He (Simon Fraser University) and Tianqi Chen (University of Washington)
+- [InfoWorld’s 2019 Technology of the Year Award](https://www.infoworld.com/article/3336072/application-development/infoworlds-2019-technology-of-the-year-award-winners.html)
 
 ## Windows Binaries
 Unofficial windows binaries and instructions on how to use them are hosted on [Guido Tapia's blog](http://www.picnet.com.au/blogs/guido/post/2016/09/22/xgboost-windows-x64-binaries-for-download/)

demo/binary_classification/README.md

@@ -62,7 +62,7 @@ test:data = "agaricus.txt.test"
 We use the tree booster and logistic regression objective in our setting. This indicates that we accomplish our task using classic gradient boosting regression tree(GBRT), which is a promising method for binary classification.
 
 The parameters shown in the example gives the most common ones that are needed to use xgboost.
-If you are interested in more parameter settings, the complete parameter settings and detailed descriptions are [here](../../doc/parameter.md). Besides putting the parameters in the configuration file, we can set them by passing them as arguments as below:
+If you are interested in more parameter settings, the complete parameter settings and detailed descriptions are [here](../../doc/parameter.rst). Besides putting the parameters in the configuration file, we can set them by passing them as arguments as below:
 
 ```
 ../../xgboost mushroom.conf max_depth=6
@@ -80,12 +80,6 @@ booster = gblinear
 # L2 regularization term on weights, default 0
 lambda = 0.01
 # L1 regularization term on weights, default 0
-If ```agaricus.txt.test.buffer``` exists, and automatically loads from binary buffer if possible, this can speedup training process when you do training many times. You can disable it by setting ```use_buffer=0```.
-  - Buffer file can also be used as standalone input, i.e if buffer file exists, but original agaricus.txt.test was removed, xgboost will still run
-* Deviation from LibSVM input format: xgboost is compatible with LibSVM format, with the following minor differences:
-  - xgboost allows feature index starts from 0
-  - for binary classification, the label is 1 for positive, 0 for negative, instead of +1,-1
-  - the feature indices in each line *do not* need to be sorted
 alpha = 0.01
 # L2 regularization term on bias, default 0
 lambda_bias = 0.01
@@ -102,7 +96,7 @@ After training, we can use the output model to get the prediction of the test da
 For binary classification, the output predictions are probability confidence scores in [0,1], corresponds to the probability of the label to be positive.
 
 #### Dump Model
-This is a preliminary feature, so far only tree model support text dump. XGBoost can display the tree models in text files and we can scan the model in an easy way:
+This is a preliminary feature, so only tree models support text dump. XGBoost can display the tree models in text or JSON files, and we can scan the model in an easy way:
 ```
 ../../xgboost mushroom.conf task=dump model_in=0002.model name_dump=dump.raw.txt
 ../../xgboost mushroom.conf task=dump model_in=0002.model fmap=featmap.txt name_dump=dump.nice.txt

demo/binary_classification/mapfeat.py

@@ -18,7 +18,7 @@ def loadfmap( fname ):
             if it.strip() == '':
                 continue
             k , v = it.split('=')
-            fmap[ idx ][ v ] = len(nmap) + 1
+            fmap[ idx ][ v ] = len(nmap)
             nmap[ len(nmap) ] = ftype+'='+k
     return fmap, nmap
 

demo/guide-python/custom_objective.py

@@ -33,10 +33,10 @@ def logregobj(preds, dtrain):
 # Take this in mind when you use the customization, and maybe you need write customized evaluation function
 def evalerror(preds, dtrain):
     labels = dtrain.get_label()
-    # return a pair metric_name, result
+    # return a pair metric_name, result. The metric name must not contain a colon (:) or a space
     # since preds are margin(before logistic transformation, cutoff at 0)
-    return 'error', float(sum(labels != (preds > 0.0))) / len(labels)
+    return 'my-error', float(sum(labels != (preds > 0.0))) / len(labels)
 
 # training with customized objective, we can also do step by step training
 # simply look at xgboost.py's implementation of train
-bst = xgb.train(param, dtrain, num_round, watchlist, logregobj, evalerror)
+bst = xgb.train(param, dtrain, num_round, watchlist, obj=logregobj, feval=evalerror)

demo/guide-python/runall.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-export PYTHONPATH=PYTHONPATH:../../python-package
+export PYTHONPATH=$PYTHONPATH:../../python-package
 python basic_walkthrough.py
 python custom_objective.py
 python boost_from_prediction.py

demo/kaggle-higgs/higgs-train.R

@@ -24,9 +24,9 @@ param <- list("objective" = "binary:logitraw",
               "silent" = 1,
               "nthread" = 16)
 watchlist <- list("train" = xgmat)
-nround = 120
+nrounds = 120
 print ("loading data end, start to boost trees")
-bst = xgb.train(param, xgmat, nround, watchlist );
+bst = xgb.train(param, xgmat, nrounds, watchlist );
 # save out model
 xgb.save(bst, "higgs.model")
 print ('finish training')

demo/kaggle-higgs/speedtest.R

@@ -39,9 +39,9 @@ for (i in 1:length(threads)){
                   "silent" = 1,
                   "nthread" = thread)
     watchlist <- list("train" = xgmat)
-    nround = 120
+    nrounds = 120
     print ("loading data end, start to boost trees")
-    bst = xgb.train(param, xgmat, nround, watchlist );
+    bst = xgb.train(param, xgmat, nrounds, watchlist );
     # save out model
     xgb.save(bst, "higgs.model")
     print ('finish training')

demo/kaggle-otto/README.MD

@@ -1,4 +1,4 @@
-Benckmark for Otto Group Competition
+Benchmark for Otto Group Competition
 =========
 
 This is a folder containing the benchmark for the [Otto Group Competition on Kaggle](http://www.kaggle.com/c/otto-group-product-classification-challenge).
@@ -20,5 +20,3 @@ devtools::install_github('tqchen/xgboost',subdir='R-package')
 ```
 
 Windows users may need to install [RTools](http://cran.r-project.org/bin/windows/Rtools/) first.
-
-

demo/kaggle-otto/otto_train_pred.R

@@ -23,13 +23,13 @@ param <- list("objective" = "multi:softprob",
               "nthread" = 8)
 
 # Run Cross Validation
-cv.nround = 50
+cv.nrounds = 50
 bst.cv = xgb.cv(param=param, data = x[trind,], label = y, 
-                nfold = 3, nrounds=cv.nround)
+                nfold = 3, nrounds=cv.nrounds)
 
 # Train the model
-nround = 50
-bst = xgboost(param=param, data = x[trind,], label = y, nrounds=nround)
+nrounds = 50
+bst = xgboost(param=param, data = x[trind,], label = y, nrounds=nrounds)
 
 # Make prediction
 pred = predict(bst,x[teind,])

demo/kaggle-otto/understandingXGBoostModel.Rmd

@@ -121,19 +121,19 @@ param <- list("objective" = "multi:softprob",
               "eval_metric" = "mlogloss",
               "num_class" = numberOfClasses)
 
-cv.nround <- 5
+cv.nrounds <- 5
 cv.nfold <- 3
 
 bst.cv = xgb.cv(param=param, data = trainMatrix, label = y,
-                nfold = cv.nfold, nrounds = cv.nround)
+                nfold = cv.nfold, nrounds = cv.nrounds)
 ```
 > As we can see the error rate is low on the test dataset (for a 5mn trained model).
 
 Finally, we are ready to train the real model!!!
 
 ```{r modelTraining}
-nround = 50
-bst = xgboost(param=param, data = trainMatrix, label = y, nrounds=nround)
+nrounds = 50
+bst = xgboost(param=param, data = trainMatrix, label = y, nrounds=nrounds)
 ```
 
 Model understanding
@@ -142,7 +142,7 @@ Model understanding
 Feature importance
 ------------------
 
-So far, we have built a model made of **`r nround`** trees.
+So far, we have built a model made of **`r nrounds`** trees.
 
 To build a tree, the dataset is divided recursively several times. At the end of the process, you get groups of observations (here, these observations are properties regarding **Otto** products).
 

demo/multiclass_classification/README.md

@@ -1,10 +1,10 @@
 Demonstrating how to use XGBoost accomplish Multi-Class classification task on [UCI Dermatology dataset](https://archive.ics.uci.edu/ml/datasets/Dermatology)
 
-Make sure you make make xgboost python module in ../../python
+Make sure you make xgboost python module in ../../python
 
 1. Run runexp.sh
 ```bash
 ./runexp.sh
 ```
 
-
+**R version** please see the `train.R`.

demo/multiclass_classification/train.R

@@ -0,0 +1,64 @@
+library(data.table)
+library(xgboost)
+
+if (!file.exists("./dermatology.data")) {
+  download.file(
+    "https://archive.ics.uci.edu/ml/machine-learning-databases/dermatology/dermatology.data",
+    "dermatology.data",
+    method = "curl"
+  )
+}
+
+df <- fread("dermatology.data", sep = ",", header = FALSE)
+
+df[, `:=`(V34 = as.integer(ifelse(V34 == "?", 0L, V34)),
+          V35 = V35 - 1L)]
+
+idx <- sample(nrow(df), size = round(0.7 * nrow(df)), replace = FALSE)
+
+train <- df[idx,]
+test <- df[-idx,]
+
+train_x <- train[, 1:34]
+train_y <- train[, V35]
+
+test_x <- test[, 1:34]
+test_y <- test[, V35]
+
+xg_train <- xgb.DMatrix(data = as.matrix(train_x), label = train_y)
+xg_test = xgb.DMatrix(as.matrix(test_x), label = test_y)
+
+params <- list(
+  objective = 'multi:softmax',
+  num_class = 6,
+  max_depth = 6,
+  nthread = 4,
+  eta = 0.1
+)
+
+watchlist = list(train = xg_train, test = xg_test)
+
+bst <- xgb.train(
+  params = params,
+  data = xg_train,
+  watchlist = watchlist,
+  nrounds = 5
+)
+
+pred <- predict(bst, xg_test)
+error_rate <- sum(pred != test_y) / length(test_y)
+print(paste("Test error using softmax =", error_rate))
+
+# do the same thing again, but output probabilities
+params$objective <- 'multi:softprob'
+bst <- xgb.train(params, xg_train, nrounds = 5, watchlist)
+
+pred_prob <- predict(bst, xg_test)
+
+pred_mat <- matrix(pred_prob, ncol = 6, byrow = TRUE)
+# validation
+# rowSums(pred_mat)
+
+pred_label <- apply(pred_mat, 1, which.max) - 1L
+error_rate = sum(pred_label != test_y) / length(test_y)
+print(paste("Test error using softprob =", error_rate))

demo/rank/README.md

@@ -1,6 +1,6 @@
 Learning to rank
 ====
-XGBoost supports accomplishing ranking tasks. In ranking scenario, data are often grouped and we need the [group information file](../../doc/input_format.md#group-input-format) to specify ranking tasks. The model used in XGBoost for ranking is the LambdaRank, this function is not yet completed. Currently, we provide pairwise rank.
+XGBoost supports accomplishing ranking tasks. In ranking scenario, data are often grouped and we need the [group information file](../../doc/tutorials/input_format.rst#group-input-format) to specify ranking tasks. The model used in XGBoost for ranking is the LambdaRank, this function is not yet completed. Currently, we provide pairwise rank.
 
 ### Parameters
 The configuration setting is similar to the regression and binary classification setting, except user need to specify the objectives:
@@ -15,14 +15,27 @@ For more usage details please refer to the [binary classification demo](../binar
 Instructions
 ====
 The dataset for ranking demo is from LETOR04 MQ2008 fold1.
-You can use the following command to run the example:
+Before running the examples, you need to get the data by running:
 
-Get the data:
 ```
 ./wgetdata.sh
 ```
 
+### Command Line
 Run the example:
 ```
 ./runexp.sh
 ```
+
+### Python
+There are two ways of doing ranking in python.  
+
+Run the example using `xgboost.train`:
+```
+python rank.py
+```
+
+Run the example using `XGBRanker`:
+```
+python rank_sklearn.py
+```

demo/rank/rank.py

@@ -0,0 +1,41 @@
+#!/usr/bin/python
+import xgboost as xgb
+from xgboost import DMatrix
+from sklearn.datasets import load_svmlight_file
+
+
+#  This script demonstrate how to do ranking with xgboost.train
+x_train, y_train = load_svmlight_file("mq2008.train")
+x_valid, y_valid = load_svmlight_file("mq2008.vali")
+x_test, y_test = load_svmlight_file("mq2008.test")
+
+group_train = []
+with open("mq2008.train.group", "r") as f:
+    data = f.readlines()
+    for line in data:
+        group_train.append(int(line.split("\n")[0]))
+
+group_valid = []
+with open("mq2008.vali.group", "r") as f:
+    data = f.readlines()
+    for line in data:
+        group_valid.append(int(line.split("\n")[0]))
+
+group_test = []
+with open("mq2008.test.group", "r") as f:
+    data = f.readlines()
+    for line in data:
+        group_test.append(int(line.split("\n")[0]))
+
+train_dmatrix = DMatrix(x_train, y_train)
+valid_dmatrix = DMatrix(x_valid, y_valid)
+test_dmatrix = DMatrix(x_test)
+
+train_dmatrix.set_group(group_train)
+valid_dmatrix.set_group(group_valid)
+
+params = {'objective': 'rank:pairwise', 'eta': 0.1, 'gamma': 1.0,
+               'min_child_weight': 0.1, 'max_depth': 6}
+xgb_model = xgb.train(params, train_dmatrix, num_boost_round=4,
+                           evals=[(valid_dmatrix, 'validation')])
+pred = xgb_model.predict(test_dmatrix)

demo/rank/rank_sklearn.py

@@ -0,0 +1,35 @@
+#!/usr/bin/python
+import xgboost as xgb
+from sklearn.datasets import load_svmlight_file
+
+
+#  This script demonstrate how to do ranking with XGBRanker
+x_train, y_train = load_svmlight_file("mq2008.train")
+x_valid, y_valid = load_svmlight_file("mq2008.vali")
+x_test, y_test = load_svmlight_file("mq2008.test")
+
+group_train = []
+with open("mq2008.train.group", "r") as f:
+    data = f.readlines()
+    for line in data:
+        group_train.append(int(line.split("\n")[0]))
+
+group_valid = []
+with open("mq2008.vali.group", "r") as f:
+    data = f.readlines()
+    for line in data:
+        group_valid.append(int(line.split("\n")[0]))
+
+group_test = []
+with open("mq2008.test.group", "r") as f:
+    data = f.readlines()
+    for line in data:
+        group_test.append(int(line.split("\n")[0]))
+
+params = {'objective': 'rank:pairwise', 'learning_rate': 0.1,
+          'gamma': 1.0, 'min_child_weight': 0.1,
+          'max_depth': 6, 'n_estimators': 4}
+model = xgb.sklearn.XGBRanker(**params)
+model.fit(x_train, y_train, group_train,
+          eval_set=[(x_valid, y_valid)], eval_group=[group_valid])
+pred = model.predict(x_test)

demo/rank/runexp.sh

@@ -1,11 +1,5 @@
-python trans_data.py train.txt mq2008.train mq2008.train.group
-
-python trans_data.py test.txt mq2008.test mq2008.test.group
-
-python trans_data.py vali.txt mq2008.vali mq2008.vali.group
+#!/bin/bash
 
 ../../xgboost mq2008.conf
 
 ../../xgboost mq2008.conf task=pred model_in=0004.model
-
-

demo/rank/wgetdata.sh

@@ -2,3 +2,9 @@
 wget https://s3-us-west-2.amazonaws.com/xgboost-examples/MQ2008.rar
 unrar x MQ2008.rar
 mv -f MQ2008/Fold1/*.txt .
+
+python trans_data.py train.txt mq2008.train mq2008.train.group
+
+python trans_data.py test.txt mq2008.test mq2008.test.group
+
+python trans_data.py vali.txt mq2008.vali mq2008.vali.group

doc/R-package/discoverYourData.md

@@ -222,7 +222,7 @@ The code below is very usual. For more information, you can look at the document
 
 ```r
 bst <- xgboost(data = sparse_matrix, label = output_vector, max.depth = 4,
-               eta = 1, nthread = 2, nround = 10,objective = "binary:logistic")
+               eta = 1, nthread = 2, nrounds = 10,objective = "binary:logistic")
 ```
 
 ```
@@ -244,7 +244,7 @@ A model which fits too well may [overfit](http://en.wikipedia.org/wiki/Overfitti
 
 > Here you can see the numbers decrease until line 7 and then increase.
 >
-> It probably means we are overfitting. To fix that I should reduce the number of rounds to `nround = 4`. I will let things like that because I don't really care for the purpose of this example :-)
+> It probably means we are overfitting. To fix that I should reduce the number of rounds to `nrounds = 4`. I will let things like that because I don't really care for the purpose of this example :-)
 
 Feature importance
 ------------------
@@ -448,7 +448,7 @@ train <- agaricus.train
 test <- agaricus.test
 
 #Random Forest™ - 1000 trees
-bst <- xgboost(data = train$data, label = train$label, max.depth = 4, num_parallel_tree = 1000, subsample = 0.5, colsample_bytree =0.5, nround = 1, objective = "binary:logistic")
+bst <- xgboost(data = train$data, label = train$label, max.depth = 4, num_parallel_tree = 1000, subsample = 0.5, colsample_bytree =0.5, nrounds = 1, objective = "binary:logistic")
 ```
 
 ```
@@ -457,7 +457,7 @@ bst <- xgboost(data = train$data, label = train$label, max.depth = 4, num_parall
 
 ```r
 #Boosting - 3 rounds
-bst <- xgboost(data = train$data, label = train$label, max.depth = 4, nround = 3, objective = "binary:logistic")
+bst <- xgboost(data = train$data, label = train$label, max.depth = 4, nrounds = 3, objective = "binary:logistic")
 ```
 
 ```

doc/R-package/xgboostPresentation.md

@@ -176,13 +176,13 @@ In a *sparse* matrix, cells containing `0` are not stored in memory. Therefore,
 We will train decision tree model using the following parameters:
 
 * `objective = "binary:logistic"`: we will train a binary classification model ;
-* `max.deph = 2`: the trees won't be deep, because our case is very simple ;
+* `max.depth = 2`: the trees won't be deep, because our case is very simple ;
 * `nthread = 2`: the number of cpu threads we are going to use;
-* `nround = 2`: there will be two passes on the data, the second one will enhance the model by further reducing the difference between ground truth and prediction.
+* `nrounds = 2`: there will be two passes on the data, the second one will enhance the model by further reducing the difference between ground truth and prediction.
 
 
 ```r
-bstSparse <- xgboost(data = train$data, label = train$label, max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic")
+bstSparse <- xgboost(data = train$data, label = train$label, max.depth = 2, eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic")
 ```
 
 ```
@@ -200,7 +200,7 @@ Alternatively, you can put your dataset in a *dense* matrix, i.e. a basic **R**
 
 
 ```r
-bstDense <- xgboost(data = as.matrix(train$data), label = train$label, max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic")
+bstDense <- xgboost(data = as.matrix(train$data), label = train$label, max.depth = 2, eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic")
 ```
 
 ```
@@ -215,7 +215,7 @@ bstDense <- xgboost(data = as.matrix(train$data), label = train$label, max.depth
 
 ```r
 dtrain <- xgb.DMatrix(data = train$data, label = train$label)
-bstDMatrix <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic")
+bstDMatrix <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic")
 ```
 
 ```
@@ -232,13 +232,13 @@ One of the simplest way to see the training progress is to set the `verbose` opt
 
 ```r
 # verbose = 0, no message
-bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic", verbose = 0)
+bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic", verbose = 0)
 ```
 
 
 ```r
 # verbose = 1, print evaluation metric
-bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic", verbose = 1)
+bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic", verbose = 1)
 ```
 
 ```
@@ -249,7 +249,7 @@ bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nround = 2, o
 
 ```r
 # verbose = 2, also print information about tree
-bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nround = 2, objective = "binary:logistic", verbose = 2)
+bst <- xgboost(data = dtrain, max.depth = 2, eta = 1, nthread = 2, nrounds = 2, objective = "binary:logistic", verbose = 2)
 ```
 
 ```
@@ -372,7 +372,7 @@ For the purpose of this example, we use `watchlist` parameter. It is a list of `
 ```r
 watchlist <- list(train=dtrain, test=dtest)
 
-bst <- xgb.train(data=dtrain, max.depth=2, eta=1, nthread = 2, nround=2, watchlist=watchlist, objective = "binary:logistic")
+bst <- xgb.train(data=dtrain, max.depth=2, eta=1, nthread = 2, nrounds=2, watchlist=watchlist, objective = "binary:logistic")
 ```
 
 ```
@@ -380,7 +380,7 @@ bst <- xgb.train(data=dtrain, max.depth=2, eta=1, nthread = 2, nround=2, watchli
 ## [1]	train-error:0.022263	test-error:0.021726
 ```
 
-**XGBoost** has computed at each round the same average error metric than seen above (we set `nround` to 2, that is why we have two lines). Obviously, the `train-error` number is related to the training dataset (the one the algorithm learns from) and the `test-error` number to the test dataset.
+**XGBoost** has computed at each round the same average error metric than seen above (we set `nrounds` to 2, that is why we have two lines). Obviously, the `train-error` number is related to the training dataset (the one the algorithm learns from) and the `test-error` number to the test dataset.
 
 Both training and test error related metrics are very similar, and in some way, it makes sense: what we have learned from the training dataset matches the observations from the test dataset.
 
@@ -390,7 +390,7 @@ For a better understanding of the learning progression, you may want to have som
 
 
 ```r
-bst <- xgb.train(data=dtrain, max.depth=2, eta=1, nthread = 2, nround=2, watchlist=watchlist, eval.metric = "error", eval.metric = "logloss", objective = "binary:logistic")
+bst <- xgb.train(data=dtrain, max.depth=2, eta=1, nthread = 2, nrounds=2, watchlist=watchlist, eval.metric = "error", eval.metric = "logloss", objective = "binary:logistic")
 ```
 
 ```
@@ -407,7 +407,7 @@ Until now, all the learnings we have performed were based on boosting trees. **X
 
 
 ```r
-bst <- xgb.train(data=dtrain, booster = "gblinear", max.depth=2, nthread = 2, nround=2, watchlist=watchlist, eval.metric = "error", eval.metric = "logloss", objective = "binary:logistic")
+bst <- xgb.train(data=dtrain, booster = "gblinear", max.depth=2, nthread = 2, nrounds=2, watchlist=watchlist, eval.metric = "error", eval.metric = "logloss", objective = "binary:logistic")
 ```
 
 ```
@@ -445,7 +445,7 @@ dtrain2 <- xgb.DMatrix("dtrain.buffer")
 ```
 
 ```r
-bst <- xgb.train(data=dtrain2, max.depth=2, eta=1, nthread = 2, nround=2, watchlist=watchlist, objective = "binary:logistic")
+bst <- xgb.train(data=dtrain2, max.depth=2, eta=1, nthread = 2, nrounds=2, watchlist=watchlist, objective = "binary:logistic")
 ```
 
 ```
@@ -576,8 +576,8 @@ print(class(rawVec))
 bst3 <- xgb.load(rawVec)
 pred3 <- predict(bst3, test$data)
 
-# pred2 should be identical to pred
-print(paste("sum(abs(pred3-pred))=", sum(abs(pred2-pred))))
+# pred3 should be identical to pred
+print(paste("sum(abs(pred3-pred))=", sum(abs(pred3-pred))))
 ```
 
 ```

doc/build.rst

@@ -13,7 +13,7 @@ Installation Guide
     #   * xgboost-{version}-py2.py3-none-win_amd64.whl
     pip3 install xgboost
 
-  * The binary wheel will support GPU algorithms (`gpu_exact`, `gpu_hist`) on machines with NVIDIA GPUs. **However, it will not support multi-GPU training; only single GPU will be used.** To enable multi-GPU training, download and install the binary wheel from `this page <https://s3-us-west-2.amazonaws.com/xgboost-wheels/list.html>`_.
+  * The binary wheel will support GPU algorithms (`gpu_exact`, `gpu_hist`) on machines with NVIDIA GPUs. Please note that **training with multiple GPUs is only supported for Linux platform**. See :doc:`gpu/index`.
   * Currently, we provide binary wheels for 64-bit Linux and Windows.
 
 ****************************
@@ -70,19 +70,21 @@ Our goal is to build the shared library:
 The minimal building requirement is
 
 - A recent C++ compiler supporting C++11 (g++-4.8 or higher)
-
-We can edit ``make/config.mk`` to change the compile options, and then build by
-``make``. If everything goes well, we can go to the specific language installation section.
+- CMake 3.2 or higher
 
 Building on Ubuntu/Debian
 =========================
 
-On Ubuntu, one builds XGBoost by running
+On Ubuntu, one builds XGBoost by running CMake:
 
 .. code-block:: bash
 
   git clone --recursive https://github.com/dmlc/xgboost
-  cd xgboost; make -j4
+  cd xgboost
+  mkdir build
+  cd build
+  cmake ..
+  make -j4
 
 Building on OSX
 ===============
@@ -90,11 +92,11 @@ Building on OSX
 Install with pip: simple method
 --------------------------------
 
-First, make sure you obtained ``gcc-5`` (newer version does not work with this method yet). Note: installation of ``gcc`` can take a while (~ 30 minutes).
+First, obtain ``gcc-8`` with Homebrew (https://brew.sh/) to enable multi-threading (i.e. using multiple CPU threads for training). The default Apple Clang compiler does not support OpenMP, so using the default compiler would have disabled multi-threading.
 
 .. code-block:: bash
 
-  brew install gcc@5
+  brew install gcc@8
 
 Then install XGBoost with ``pip``:
 
@@ -102,42 +104,30 @@ Then install XGBoost with ``pip``:
 
   pip3 install xgboost
 
-You might need to run the command with ``sudo`` if you run into permission errors.
+You might need to run the command with ``--user`` flag if you run into permission errors.
 
 Build from the source code - advanced method
 --------------------------------------------
 
-First, obtain ``gcc-7`` with homebrew (https://brew.sh/) if you want multi-threaded version. Clang is okay if multithreading is not required. Note: installation of ``gcc`` can take a while (~ 30 minutes).
+Obtain ``gcc-8`` from Homebrew:
 
 .. code-block:: bash
 
-  brew install gcc@7
+  brew install gcc@8
 
-Now, clone the repository:
+Now clone the repository:
 
 .. code-block:: bash
 
   git clone --recursive https://github.com/dmlc/xgboost
-  cd xgboost; cp make/config.mk ./config.mk
 
-Open ``config.mk`` and uncomment these two lines:
-
-.. code-block:: bash
-
-  export CC = gcc
-  export CXX = g++
-
-and replace these two lines as follows: (specify the GCC version)
-
-.. code-block:: bash
-
-  export CC = gcc-7
-  export CXX = g++-7
-
-Now, you may build XGBoost using the following command:
+Create the ``build/`` directory and invoke CMake. Make sure to add ``CC=gcc-8 CXX=g++-8`` so that Homebrew GCC is selected. After invoking CMake, you can build XGBoost with ``make``:
 
 .. code-block:: bash
 
+  mkdir build
+  cd build
+  CC=gcc-8 CXX=g++-8 cmake ..
   make -j4
 
 You may now continue to `Python Package Installation`_.
@@ -154,6 +144,20 @@ We recommend you use `Git for Windows <https://git-for-windows.github.io/>`_, as
 
 XGBoost support compilation with Microsoft Visual Studio and MinGW.
 
+Compile XGBoost with Microsoft Visual Studio
+--------------------------------------------
+To build with Visual Studio, we will need CMake. Make sure to install a recent version of CMake. Then run the following from the root of the XGBoost directory:
+
+.. code-block:: bash
+
+  mkdir build
+  cd build
+  cmake .. -G"Visual Studio 14 2015 Win64"
+
+This specifies an out of source build using the Visual Studio 64 bit generator. (Change the ``-G`` option appropriately if you have a different version of Visual Studio installed.) Open the ``.sln`` file in the build directory and build with Visual Studio.
+
+After the build process successfully ends, you will find a ``xgboost.dll`` library file inside ``./lib/`` folder.
+
 Compile XGBoost using MinGW
 ---------------------------
 After installing `Git for Windows <https://git-for-windows.github.io/>`_, you should have a shortcut named ``Git Bash``. You should run all subsequent steps in ``Git Bash``.
@@ -173,21 +177,7 @@ To build with MinGW, type:
 
   cp make/mingw64.mk config.mk; make -j4
 
-Compile XGBoost with Microsoft Visual Studio
---------------------------------------------
-To build with Visual Studio, we will need CMake. Make sure to install a recent version of CMake. Then run the following from the root of the XGBoost directory:
-
-.. code-block:: bash
-
-  mkdir build
-  cd build
-  cmake .. -G"Visual Studio 12 2013 Win64"
-
-This specifies an out of source build using the MSVC 12 64 bit generator. Open the ``.sln`` file in the build directory and build with Visual Studio. To use the Python module you can copy ``xgboost.dll`` into ``python-package/xgboost``.
-
-After the build process successfully ends, you will find a ``xgboost.dll`` library file inside ``./lib/`` folder, copy this file to the the API package folder like ``python-package/xgboost`` if you are using Python API.
-
-Unofficial windows binaries and instructions on how to use them are hosted on `Guido Tapia's blog <http://www.picnet.com.au/blogs/guido/post/2016/09/22/xgboost-windows-x64-binaries-for-download/>`_.
+See :ref:`mingw_python` for buildilng XGBoost for Python.
 
 .. _build_gpu_support:
 
@@ -204,7 +194,7 @@ From the command line on Linux starting from the XGBoost directory:
   mkdir build
   cd build
   cmake .. -DUSE_CUDA=ON
-  make -j
+  make -j4
 
 .. note:: Enabling multi-GPU training
 
@@ -214,16 +204,10 @@ From the command line on Linux starting from the XGBoost directory:
 
     mkdir build
     cd build
-    cmake .. -DUSE_CUDA=ON -DUSE_NCCL=ON
-    make -j
+    cmake .. -DUSE_CUDA=ON -DUSE_NCCL=ON -DNCCL_ROOT=/path/to/nccl2
+    make -j4
 
-On Windows, see what options for generators you have for CMake, and choose one with ``[arch]`` replaced with Win64:
-
-.. code-block:: bash
-
-  cmake -help
-
-Then run CMake as follows:
+On Windows, run CMake as follows:
 
 .. code-block:: bash
 
@@ -231,13 +215,15 @@ Then run CMake as follows:
   cd build
   cmake .. -G"Visual Studio 14 2015 Win64" -DUSE_CUDA=ON
 
+(Change the ``-G`` option appropriately if you have a different version of Visual Studio installed.)
+
 .. note:: Visual Studio 2017 Win64 Generator may not work
 
   Choosing the Visual Studio 2017 generator may cause compilation failure. When it happens, specify the 2015 compiler by adding the ``-T`` option:
 
   .. code-block:: bash
 
-    make .. -G"Visual Studio 15 2017 Win64" -T v140,cuda=8.0 -DR_LIB=ON -DUSE_CUDA=ON
+    make .. -G"Visual Studio 15 2017 Win64" -T v140,cuda=8.0 -DUSE_CUDA=ON
 
 To speed up compilation, the compute version specific to your GPU could be passed to cmake as, e.g., ``-DGPU_COMPUTE_VER=50``.
 The above cmake configuration run will create an ``xgboost.sln`` solution file in the build directory. Build this solution in release mode as a x64 build, either from Visual studio or from command line:
@@ -251,7 +237,9 @@ To speed up compilation, run multiple jobs in parallel by appending option ``--
 Customized Building
 ===================
 
-The configuration file ``config.mk`` modifies several compilation flags:
+We recommend the use of CMake for most use cases. See the full range of building options in CMakeLists.txt.
+
+Alternatively, you may use Makefile. The Makefile uses a configuration file ``config.mk``, which lets you modify several compilation flags:
 - Whether to enable support for various distributed filesystems such as HDFS and Amazon S3
 - Which compiler to use
 - And some more
@@ -261,7 +249,7 @@ To customize, first copy ``make/config.mk`` to the project root and then modify
 Python Package Installation
 ===========================
 
-The python package is located at ``python-package/``.
+The Python package is located at ``python-package/``.
 There are several ways to install the package:
 
 1. Install system-wide, which requires root permission:
@@ -271,7 +259,7 @@ There are several ways to install the package:
   cd python-package; sudo python setup.py install
 
 You will however need Python ``distutils`` module for this to
-work. It is often part of the core python package or it can be installed using your
+work. It is often part of the core Python package or it can be installed using your
 package manager, e.g. in Debian use
 
 .. code-block:: bash
@@ -282,10 +270,10 @@ package manager, e.g. in Debian use
 
   If you recompiled XGBoost, then you need to reinstall it again to make the new library take effect.
 
-2. Only set the environment variable ``PYTHONPATH`` to tell python where to find
-   the library. For example, assume we cloned `xgboost` on the home directory
-   `~`. then we can added the following line in `~/.bashrc`.
-   This option is **recommended for developers** who change the code frequently. The changes will be immediately reflected once you pulled the code and rebuild the project (no need to call ``setup`` again)
+2. Only set the environment variable ``PYTHONPATH`` to tell Python where to find
+   the library. For example, assume we cloned ``xgboost`` on the home directory
+   ``~``. then we can added the following line in ``~/.bashrc``.
+   This option is **recommended for developers** who change the code frequently. The changes will be immediately reflected once you pulled the code and rebuild the project (no need to call ``setup`` again).
 
 .. code-block:: bash
 
@@ -297,12 +285,23 @@ package manager, e.g. in Debian use
 
   cd python-package; python setup.py develop --user
 
-4. If you are installing the latest XGBoost version which requires compilation, add MinGW to the system PATH:
+.. _mingw_python:
 
-.. code-block:: bash
+Building XGBoost library for Python for Windows with MinGW-w64 (Advanced)
+-------------------------------------------------------------------------
 
-    import os
-    os.environ['PATH'] = os.environ['PATH'] + ';C:\\Program Files\\mingw-w64\\x86_64-5.3.0-posix-seh-rt_v4-rev0\\mingw64\\bin'
+Windows versions of Python are built with Microsoft Visual Studio. Usually Python binary modules are built with the same compiler the interpreter is built with. However, you may not be able to use Visual Studio, for following reasons:
+
+1. VS is proprietary and commercial software. Microsoft provides a freeware "Community" edition, but its licensing terms impose restrictions as to where and how it can be used.
+2. Visual Studio contains telemetry, as documented in `Microsoft Visual Studio Licensing Terms <https://visualstudio.microsoft.com/license-terms/mt736442/>`_. Running software with telemetry may be against the policy of your organization.
+
+So you may want to build XGBoost with GCC own your own risk. This presents some difficulties because MSVC uses Microsoft runtime and MinGW-w64 uses own runtime, and the runtimes have different incompatible memory allocators. But in fact this setup is usable if you know how to deal with it. Here is some experience.
+
+1. The Python interpreter will crash on exit if XGBoost was used. This is usually not a big issue.
+2. ``-O3`` is OK.
+3. ``-mtune=native`` is also OK.
+4. Don't use ``-march=native`` gcc flag. Using it causes the Python interpreter to crash if the DLL was actually used.
+5. You may need to provide the lib with the runtime libs. If ``mingw32/bin`` is not in ``PATH``, build a wheel (``python setup.py bdist_wheel``), open it with an archiver and put the needed dlls to the directory where ``xgboost.dll`` is situated. Then you can install the wheel with ``pip``.
 
 R Package Installation
 ======================
@@ -316,35 +315,13 @@ You can install xgboost from CRAN just like any other R package:
 
   install.packages("xgboost")
 
-Or you can install it from our weekly updated drat repo:
-
-.. code-block:: R
-
-  install.packages("drat", repos="https://cran.rstudio.com")
-  drat:::addRepo("dmlc")
-  install.packages("xgboost", repos="http://dmlc.ml/drat/", type = "source")
-
-For OSX users, single threaded version will be installed. To install multi-threaded version,
-first follow `Building on OSX`_ to get the OpenMP enabled compiler. Then
-
-- Set the ``Makevars`` file in highest piority for R.
-
-  The point is, there are three ``Makevars`` : ``~/.R/Makevars``, ``xgboost/R-package/src/Makevars``, and ``/usr/local/Cellar/r/3.2.0/R.framework/Resources/etc/Makeconf`` (the last one obtained by running ``file.path(R.home("etc"), "Makeconf")`` in R), and ``SHLIB_OPENMP_CXXFLAGS`` is not set by default!! After trying, it seems that the first one has highest piority (surprise!).
-
-  Then inside R, run
-
-.. code-block:: R
-
-  install.packages("drat", repos="https://cran.rstudio.com")
-  drat:::addRepo("dmlc")
-  install.packages("xgboost", repos="http://dmlc.ml/drat/", type = "source")
+For OSX users, single-threaded version will be installed. So only one thread will be used for training. To enable use of multiple threads (and utilize capacity of multi-core CPUs), see the section :ref:`osx_multithread` to install XGBoost from source.
 
 Installing the development version
 ----------------------------------
 
 Make sure you have installed git and a recent C++ compiler supporting C++11 (e.g., g++-4.8 or higher).
-On Windows, Rtools must be installed, and its bin directory has to be added to PATH during the installation.
-And see the previous subsection for an OSX tip.
+On Windows, Rtools must be installed, and its bin directory has to be added to ``PATH`` during the installation.
 
 Due to the use of git-submodules, ``devtools::install_github`` can no longer be used to install the latest version of R package.
 Thus, one has to run git to check out the code first:
@@ -366,10 +343,39 @@ In this case, just start R as you would normally do and run the following:
   setwd('wherever/you/cloned/it/xgboost/R-package/')
   install.packages('.', repos = NULL, type="source")
 
-The package could also be built and installed with cmake (and Visual C++ 2015 on Windows) using instructions from the next section, but without GPU support (omit the ``-DUSE_CUDA=ON`` cmake parameter).
+The package could also be built and installed with CMake (and Visual C++ 2015 on Windows) using instructions from :ref:`r_gpu_support`, but without GPU support (omit the ``-DUSE_CUDA=ON`` cmake parameter).
 
 If all fails, try `Building the shared library`_ to see whether a problem is specific to R package or not.
 
+.. _osx_multithread:
+
+Installing R package on Mac OSX with multi-threading
+----------------------------------------------------
+
+First, obtain ``gcc-8`` with Homebrew (https://brew.sh/) to enable multi-threading (i.e. using multiple CPU threads for training). The default Apple Clang compiler does not support OpenMP, so using the default compiler would have disabled multi-threading.
+
+.. code-block:: bash
+
+  brew install gcc@8
+
+Now, clone the repository:
+
+.. code-block:: bash
+
+  git clone --recursive https://github.com/dmlc/xgboost
+
+Create the ``build/`` directory and invoke CMake with option ``R_LIB=ON``. Make sure to add ``CC=gcc-8 CXX=g++-8`` so that Homebrew GCC is selected. After invoking CMake, you can install the R package by running ``make`` and ``make install``:
+
+.. code-block:: bash
+
+  mkdir build
+  cd build
+  CC=gcc-7 CXX=g++-7 cmake .. -DR_LIB=ON
+  make -j4
+  make install
+
+.. _r_gpu_support:
+
 Installing R package with GPU support
 -------------------------------------
 
@@ -385,9 +391,9 @@ On Linux, starting from the XGBoost directory type:
   make install -j
 
 When default target is used, an R package shared library would be built in the ``build`` area.
-The ``install`` target, in addition, assembles the package files with this shared library under ``build/R-package``, and runs ``R CMD INSTALL``.
+The ``install`` target, in addition, assembles the package files with this shared library under ``build/R-package`` and runs ``R CMD INSTALL``.
 
-On Windows, cmake with Visual C++ Build Tools (or Visual Studio) has to be used to build an R package with GPU support. Rtools must also be installed (perhaps, some other MinGW distributions with ``gendef.exe`` and ``dlltool.exe`` would work, but that was not tested).
+On Windows, CMake with Visual C++ Build Tools (or Visual Studio) has to be used to build an R package with GPU support. Rtools must also be installed (perhaps, some other MinGW distributions with ``gendef.exe`` and ``dlltool.exe`` would work, but that was not tested).
 
 .. code-block:: bash
 
@@ -396,8 +402,8 @@ On Windows, cmake with Visual C++ Build Tools (or Visual Studio) has to be used
   cmake .. -G"Visual Studio 14 2015 Win64" -DUSE_CUDA=ON -DR_LIB=ON
   cmake --build . --target install --config Release
 
-When ``--target xgboost`` is used, an R package dll would be built under ``build/Release``.
-The ``--target install``, in addition, assembles the package files with this dll under ``build/R-package``, and runs ``R CMD INSTALL``.
+When ``--target xgboost`` is used, an R package DLL would be built under ``build/Release``.
+The ``--target install``, in addition, assembles the package files with this dll under ``build/R-package`` and runs ``R CMD INSTALL``.
 
 If cmake can't find your R during the configuration step, you might provide the location of its executable to cmake like this: ``-DLIBR_EXECUTABLE="C:/Program Files/R/R-3.4.1/bin/x64/R.exe"``.
 

doc/conf.py

@@ -14,6 +14,7 @@
 from subprocess import call
 from sh.contrib import git
 import urllib.request
+from urllib.error import HTTPError
 from recommonmark.parser import CommonMarkParser
 import sys
 import re
@@ -24,8 +25,11 @@ import guzzle_sphinx_theme
 git_branch = [re.sub(r'origin/', '', x.lstrip(' ')) for x in str(git.branch('-r', '--contains', 'HEAD')).rstrip('\n').split('\n')]
 git_branch = [x for x in git_branch if 'HEAD' not in x]
 print('git_branch = {}'.format(git_branch[0]))
-filename, _ = urllib.request.urlretrieve('https://s3-us-west-2.amazonaws.com/xgboost-docs/{}.tar.bz2'.format(git_branch[0]))
-call('if [ -d tmp ]; then rm -rf tmp; fi; mkdir -p tmp/jvm; cd tmp/jvm; tar xvf {}'.format(filename), shell=True)
+try:
+  filename, _ = urllib.request.urlretrieve('https://s3-us-west-2.amazonaws.com/xgboost-docs/{}.tar.bz2'.format(git_branch[0]))
+  call('if [ -d tmp ]; then rm -rf tmp; fi; mkdir -p tmp/jvm; cd tmp/jvm; tar xvf {}'.format(filename), shell=True)
+except HTTPError:
+  print('JVM doc not found. Skipping...')
 
 # If extensions (or modules to document with autodoc) are in another directory,
 # add these directories to sys.path here. If the directory is relative to the
@@ -37,7 +41,7 @@ sys.path.insert(0, curr_path)
 
 # -- mock out modules
 import mock
-MOCK_MODULES = ['numpy', 'scipy', 'scipy.sparse', 'sklearn', 'matplotlib', 'pandas', 'graphviz']
+MOCK_MODULES = ['scipy', 'scipy.sparse', 'sklearn', 'pandas']
 for mod_name in MOCK_MODULES:
   sys.modules[mod_name] = mock.Mock()
 
@@ -58,6 +62,7 @@ release = xgboost.__version__
 # Add any Sphinx extension module names here, as strings. They can be
 # extensions coming with Sphinx (named 'sphinx.ext.*') or your custom ones
 extensions = [
+    'matplotlib.sphinxext.plot_directive',
     'sphinx.ext.autodoc',
     'sphinx.ext.napoleon',
     'sphinx.ext.mathjax',
@@ -65,6 +70,11 @@ extensions = [
     'breathe'
 ]
 
+graphviz_output_format = 'png'
+plot_formats = [('svg', 300), ('png', 100), ('hires.png', 300)]
+plot_html_show_source_link = False
+plot_html_show_formats = False
+
 # Breathe extension variables
 breathe_projects = {"xgboost": "doxyxml/"}
 breathe_default_project = "xgboost"

doc/contribute.rst

@@ -19,6 +19,7 @@ Everyone is more than welcome to contribute. It is a way to make the project bet
 * `Documents`_
 * `Testcases`_
 * `Sanitizers`_
+* `clang-tidy`_
 * `Examples`_
 * `Core Library`_
 * `Python Package`_
@@ -149,6 +150,14 @@ sanitizer is not compatible with the other two sanitizers.
 
     cmake -DUSE_SANITIZER=ON -DENABLED_SANITIZERS="address;leak" /path/to/xgboost
 
+By default, CMake will search regular system paths for sanitizers, you can also
+supply a specified SANITIZER_PATH.
+
+  .. code-block:: bash
+
+    cmake -DUSE_SANITIZER=ON -DENABLED_SANITIZERS="address;leak" \
+    -DSANITIZER_PATH=/path/to/sanitizers /path/to/xgboost
+
 How to use sanitizers with CUDA support
 =======================================
 Runing XGBoost on CUDA with address sanitizer (asan) will raise memory error.
@@ -161,6 +170,31 @@ environment variable:
 
 For details, please consult `official documentation <https://github.com/google/sanitizers/wiki>`_ for sanitizers.
 
+**********
+clang-tidy
+**********
+To run clang-tidy on both C++ and CUDA source code,  run the following command
+from the top level source tree:
+
+  .. code-black:: bash
+    cd /path/to/xgboost/
+    python3 tests/ci_build/tidy.py --gtest-path=/path/to/google-test
+
+The script requires the full path of Google Test library via the ``--gtest-path`` argument.
+
+Also, the script accepts two optional integer arguments, namely ``--cpp`` and ``--cuda``.
+By default they are both set to 1.  If you want to exclude CUDA source from
+linting, use:
+
+  .. code-black:: bash
+    cd /path/to/xgboost/
+    python3 tests/ci_build/tidy.py --cuda=0
+
+Similarly, if you want to exclude C++ source from linting:
+
+  .. code-black:: bash
+    cd /path/to/xgboost/
+    python3 tests/ci_build/tidy.py --cpp=0
 
 ********
 Examples

doc/get_started.rst

@@ -42,7 +42,7 @@ R
   train <- agaricus.train
   test <- agaricus.test
   # fit model
-  bst <- xgboost(data = train$data, label = train$label, max.depth = 2, eta = 1, nround = 2,
+  bst <- xgboost(data = train$data, label = train$label, max.depth = 2, eta = 1, nrounds = 2,
                  nthread = 2, objective = "binary:logistic")
   # predict
   pred <- predict(bst, test$data)

doc/gpu/index.rst

@@ -5,6 +5,12 @@ XGBoost GPU Support
 This page contains information about GPU algorithms supported in XGBoost.
 To install GPU support, checkout the :doc:`/build`.
 
+.. note:: CUDA 8.0, Compute Capability 3.5 required
+
+  The GPU algorithms in XGBoost require a graphics card with compute capability 3.5 or higher, with
+  CUDA toolkits 8.0 or later.
+  (See `this list <https://en.wikipedia.org/wiki/CUDA#GPUs_supported>`_ to look up compute capability of your GPU card.)
+
 *********************************************
 CUDA Accelerated Tree Construction Algorithms
 *********************************************
@@ -12,7 +18,7 @@ Tree construction (training) and prediction can be accelerated with CUDA-capable
 
 Usage
 =====
-Specify the ``tree_method`` parameter as one of the following algorithms. 
+Specify the ``tree_method`` parameter as one of the following algorithms.
 
 Algorithms
 ----------
@@ -25,39 +31,43 @@ Algorithms
 | gpu_hist     | Equivalent to the XGBoost fast histogram algorithm. Much faster and uses considerably less memory. NOTE: Will run very slowly on GPUs older than Pascal architecture. |
 +--------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 
-Supported parameters 
+Supported parameters
 --------------------
 
-.. |tick| unicode:: U+2714 
-.. |cross| unicode:: U+2718 
+.. |tick| unicode:: U+2714
+.. |cross| unicode:: U+2718
 
-+--------------------------+---------------+--------------+
-| parameter                | ``gpu_exact`` | ``gpu_hist`` |
-+==========================+===============+==============+
-| ``subsample``            | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
-| ``colsample_bytree``     | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
-| ``colsample_bylevel``    | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
-| ``max_bin``              | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
-| ``gpu_id``               | |tick|        | |tick|       |
-+--------------------------+---------------+--------------+
-| ``n_gpus``               | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
-| ``predictor``            | |tick|        | |tick|       |
-+--------------------------+---------------+--------------+
-| ``grow_policy``          | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
-| ``monotone_constraints`` | |cross|       | |tick|       |
-+--------------------------+---------------+--------------+
++--------------------------------+---------------+--------------+
+| parameter                      | ``gpu_exact`` | ``gpu_hist`` |
++================================+===============+==============+
+| ``subsample``                  | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``colsample_bytree``           | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``colsample_bylevel``          | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``max_bin``                    | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``gpu_id``                     | |tick|        | |tick|       |
++--------------------------------+---------------+--------------+
+| ``n_gpus``                     | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``predictor``                  | |tick|        | |tick|       |
++--------------------------------+---------------+--------------+
+| ``grow_policy``                | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``monotone_constraints``       | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
+| ``single_precision_histogram`` | |cross|       | |tick|       |
++--------------------------------+---------------+--------------+
 
 GPU accelerated prediction is enabled by default for the above mentioned ``tree_method`` parameters but can be switched to CPU prediction by setting ``predictor`` to ``cpu_predictor``. This could be useful if you want to conserve GPU memory. Likewise when using CPU algorithms, GPU accelerated prediction can be enabled by setting ``predictor`` to ``gpu_predictor``.
 
+The experimental parameter ``single_precision_histogram`` can be set to True to enable building histograms using single precision. This may improve speed, in particular on older architectures.
+
 The device ordinal can be selected using the ``gpu_id`` parameter, which defaults to 0.
 
-Multiple GPUs can be used with the ``gpu_hist`` tree method using the ``n_gpus`` parameter. which defaults to 1. If this is set to -1 all available GPUs will be used.  If ``gpu_id`` is specified as non-zero, the gpu device order is ``mod(gpu_id + i) % n_visible_devices`` for ``i=0`` to ``n_gpus-1``.  As with GPU vs. CPU, multi-GPU will not always be faster than a single GPU due to PCI bus bandwidth that can limit performance.
+Multiple GPUs can be used with the ``gpu_hist`` tree method using the ``n_gpus`` parameter. which defaults to 1. If this is set to -1 all available GPUs will be used.  If ``gpu_id`` is specified as non-zero, the selected gpu devices will be from ``gpu_id`` to ``gpu_id+n_gpus``, please note that ``gpu_id+n_gpus`` must be less than or equal to the number of available GPUs on your system.  As with GPU vs. CPU, multi-GPU will not always be faster than a single GPU due to PCI bus bandwidth that can limit performance.
 
 .. note:: Enabling multi-GPU training
 
@@ -72,6 +82,95 @@ The GPU algorithms currently work with CLI, Python and R packages. See :doc:`/bu
   param['max_bin'] = 16
   param['tree_method'] = 'gpu_hist'
 
+Objective functions
+===================
+Most of the objective functions implemented in XGBoost can be run on GPU.  Following table shows current support status.
+
+.. |tick| unicode:: U+2714
+.. |cross| unicode:: U+2718
+
++-----------------+-------------+
+| Objectives      | GPU support |
++-----------------+-------------+
+| reg:linear      | |tick|      |
++-----------------+-------------+
+| reg:logistic    | |tick|      |
++-----------------+-------------+
+| binary:logistic | |tick|      |
++-----------------+-------------+
+| binary:logitraw | |tick|      |
++-----------------+-------------+
+| binary:hinge    | |tick|      |
++-----------------+-------------+
+| count:poisson   | |tick|      |
++-----------------+-------------+
+| reg:gamma       | |tick|      |
++-----------------+-------------+
+| reg:tweedie     | |tick|      |
++-----------------+-------------+
+| multi:softmax   | |tick|      |
++-----------------+-------------+
+| multi:softprob  | |tick|      |
++-----------------+-------------+
+| survival:cox    | |cross|     |
++-----------------+-------------+
+| rank:pairwise   | |cross|     |
++-----------------+-------------+
+| rank:ndcg       | |cross|     |
++-----------------+-------------+
+| rank:map        | |cross|     |
++-----------------+-------------+
+
+For multi-gpu support, objective functions also honor the ``n_gpus`` parameter,
+which, by default is set to 1.  To disable running objectives on GPU, just set
+``n_gpus`` to 0.
+
+Metric functions
+===================
+Following table shows current support status for evaluation metrics on the GPU.
+
+.. |tick| unicode:: U+2714
+.. |cross| unicode:: U+2718
+
++-----------------+-------------+
+| Metric          | GPU Support |
++=================+=============+
+| rmse            | |tick|      |
++-----------------+-------------+
+| mae             | |tick|      |
++-----------------+-------------+
+| logloss         | |tick|      |
++-----------------+-------------+
+| error           | |tick|      |
++-----------------+-------------+
+| merror          | |cross|     |
++-----------------+-------------+
+| mlogloss        | |cross|     |
++-----------------+-------------+
+| auc             | |cross|     |
++-----------------+-------------+
+| aucpr           | |cross|     |
++-----------------+-------------+
+| ndcg            | |cross|     |
++-----------------+-------------+
+| map             | |cross|     |
++-----------------+-------------+
+| poisson-nloglik | |tick|      |
++-----------------+-------------+
+| gamma-nloglik   | |tick|      |
++-----------------+-------------+
+| cox-nloglik     | |cross|     |
++-----------------+-------------+
+| gamma-deviance  | |tick|      |
++-----------------+-------------+
+| tweedie-nloglik | |tick|      |
++-----------------+-------------+
+
+As for objective functions, metrics honor the ``n_gpus`` parameter,
+which, by default is set to 1.  To disable running metrics on GPU, just set
+``n_gpus`` to 0.
+
+
 Benchmarks
 ==========
 You can run benchmarks on synthetic data for binary classification:
@@ -103,13 +202,16 @@ References
 
 `Nvidia Parallel Forall: Gradient Boosting, Decision Trees and XGBoost with CUDA <https://devblogs.nvidia.com/parallelforall/gradient-boosting-decision-trees-xgboost-cuda/>`_
 
-Authors
+Contributors
 =======
-* Rory Mitchell
+Many thanks to the following contributors (alphabetical order):
+* Andrey Adinets
+* Jiaming Yuan
 * Jonathan C. McKinney
+* Matthew Jones
+* Philip Cho
+* Rory Mitchell
 * Shankara Rao Thejaswi Nanditale
 * Vinay Deshpande
-* ... and the rest of the H2O.ai and NVIDIA team.
 
 Please report bugs to the user forum https://discuss.xgboost.ai/.
-

doc/jvm/index.rst

@@ -58,9 +58,11 @@ For sbt, please add the repository and dependency in build.sbt as following:
 
 If you want to use XGBoost4J-Spark, replace ``xgboost4j`` with ``xgboost4j-spark``.
 
-.. note:: XGBoost4J-Spark requires Spark 2.3+
+.. note:: XGBoost4J-Spark requires Apache Spark 2.3+
 
-  XGBoost4J-Spark now requires Spark 2.3+. Latest versions of XGBoost4J-Spark uses facilities of `org.apache.spark.ml.param.shared` extensively to provide for a tight integration with Spark MLLIB framework, and these facilities are not fully available on earlier versions of Spark.
+  XGBoost4J-Spark now requires **Apache Spark 2.3+**. Latest versions of XGBoost4J-Spark uses facilities of `org.apache.spark.ml.param.shared` extensively to provide for a tight integration with Spark MLLIB framework, and these facilities are not fully available on earlier versions of Spark.
+
+  Also, make sure to install Spark directly from `Apache website <https://spark.apache.org/>`_. **Upstream XGBoost is not guaranteed to work with third-party distributions of Spark, such as Cloudera Spark.** Consult appropriate third parties to obtain their distribution of XGBoost.
 
 Installation from maven repo
 ============================

doc/jvm/xgboost4j_spark_tutorial.rst

@@ -57,13 +57,21 @@ and then refer to the snapshot dependency by adding:
 
   <dependency>
       <groupId>ml.dmlc</groupId>
-      <artifactId>xgboost4j</artifactId>
+      <artifactId>xgboost4j-spark</artifactId>
       <version>next_version_num-SNAPSHOT</version>
   </dependency>
 
-.. note:: XGBoost4J-Spark requires Spark 2.3+
+.. note:: XGBoost4J-Spark requires Apache Spark 2.3+
 
-  XGBoost4J-Spark now requires Spark 2.3+. Latest versions of XGBoost4J-Spark uses facilities of `org.apache.spark.ml.param.shared` extensively to provide for a tight integration with Spark MLLIB framework, and these facilities are not fully available on earlier versions of Spark.
+  XGBoost4J-Spark now requires **Apache Spark 2.3+**. Latest versions of XGBoost4J-Spark uses facilities of `org.apache.spark.ml.param.shared` extensively to provide for a tight integration with Spark MLLIB framework, and these facilities are not fully available on earlier versions of Spark.
+
+  Also, make sure to install Spark directly from `Apache website <https://spark.apache.org/>`_. **Upstream XGBoost is not guaranteed to work with third-party distributions of Spark, such as Cloudera Spark.** Consult appropriate third parties to obtain their distribution of XGBoost.
+
+Installation from maven repo
+
+.. note:: Use of Python in XGBoost4J-Spark
+
+  By default, we use the tracker in `dmlc-core <https://github.com/dmlc/dmlc-core/tree/master/tracker>`_ to drive the training with XGBoost4J-Spark. It requires Python 2.7+. We also have an experimental Scala version of tracker which can be enabled by passing the parameter ``tracker_conf`` as ``scala``.
 
 Data Preparation
 ================
@@ -183,6 +191,20 @@ After we set XGBoostClassifier parameters and feature/label column, we can build
 
   val xgbClassificationModel = xgbClassifier.fit(xgbInput)
 
+Early Stopping
+----------------
+
+Early stopping is a feature to prevent the unnecessary training iterations. By specifying ``num_early_stopping_rounds`` or directly call ``setNumEarlyStoppingRounds`` over a XGBoostClassifier or XGBoostRegressor, we can define number of rounds if the evaluation metric going away from the best iteration and early stop training iterations.
+
+In additional to ``num_early_stopping_rounds``, you also need to define ``maximize_evaluation_metrics`` or call ``setMaximizeEvaluationMetrics`` to specify whether you want to maximize or minimize the metrics in training.
+
+For example, we need to maximize the evaluation metrics (set ``maximize_evaluation_metrics`` with true), and set ``num_early_stopping_rounds`` with 5. The evaluation metric of 10th iteration is the maximum one until now. In the following iterations, if there is no evaluation metric greater than the 10th iteration's (best one), the traning would be early stopped at 15th iteration.  
+
+Training with Evaluation Sets
+----------------
+
+You can also monitor the performance of the model during training with multiple evaluation datasets. By specifying ``eval_sets`` or call ``setEvalSets`` over a XGBoostClassifier or XGBoostRegressor, you can pass in multiple evaluation datasets typed as a Map from String to DataFrame.
+
 Prediction
 ==========
 
@@ -274,7 +296,7 @@ and then loading the model in another session:
 With regards to ML pipeline save and load, please refer the next section.
 
 Interact with Other Bindings of XGBoost
-------------------------------------
+---------------------------------------
 After we train a model with XGBoost4j-Spark on massive dataset, sometimes we want to do model serving in single machine or integrate it with other single node libraries for further processing. XGBoost4j-Spark supports export model to local by:
 
 .. code-block:: scala

doc/parameter.rst

@@ -23,14 +23,25 @@ General Parameters
 
   - Which booster to use. Can be ``gbtree``, ``gblinear`` or ``dart``; ``gbtree`` and ``dart`` use tree based models while ``gblinear`` uses linear functions.
 
-* ``silent`` [default=0]
+* ``silent`` [default=0] [Deprecated]
 
-  - 0 means printing running messages, 1 means silent mode
+  - Deprecated.  Please use ``verbosity`` instead.
+
+* ``verbosity`` [default=1]
+
+  - Verbosity of printing messages.  Valid values are 0 (silent),
+    1 (warning), 2 (info), 3 (debug).  Sometimes XGBoost tries to change
+    configurations based on heuristics, which is displayed as warning message.
+    If there's unexpected behaviour, please try to increase value of verbosity.
 
 * ``nthread`` [default to maximum number of threads available if not set]
 
   - Number of parallel threads used to run XGBoost
 
+* ``disable_default_eval_metric`` [default=0]
+
+  - Flag to disable default metric. Set to >0 to disable.
+
 * ``num_pbuffer`` [set automatically by XGBoost, no need to be set by user]
 
   - Size of prediction buffer, normally set to number of training instances. The buffers are used to save the prediction results of last boosting step.
@@ -53,8 +64,8 @@ Parameters for Tree Booster
 
 * ``max_depth`` [default=6]
 
-  - Maximum depth of a tree. Increasing this value will make the model more complex and more likely to overfit. 0 indicates no limit. Note that limit is required when ``grow_policy`` is set of ``depthwise``.
-  - range: [0,∞]
+  - Maximum depth of a tree. Increasing this value will make the model more complex and more likely to overfit. 0 is only accepted in ``lossguided`` growing policy when tree_method is set as ``hist`` and it indicates no limit on depth. Beware that XGBoost aggressively consumes memory when training a deep tree.
+  - range: [0,∞] (0 is only accepted in ``lossguided`` growing policy when tree_method is set as ``hist``)
 
 * ``min_child_weight`` [default=1]
 
@@ -71,15 +82,22 @@ Parameters for Tree Booster
   - Subsample ratio of the training instances. Setting it to 0.5 means that XGBoost would randomly sample half of the training data prior to growing trees. and this will prevent overfitting. Subsampling will occur once in every boosting iteration.
   - range: (0,1]
 
-* ``colsample_bytree`` [default=1]
-
-  - Subsample ratio of columns when constructing each tree. Subsampling will occur once in every boosting iteration.
-  - range: (0,1]
-
-* ``colsample_bylevel`` [default=1]
-
-  - Subsample ratio of columns for each split, in each level. Subsampling will occur each time a new split is made. This paramter has no effect when ``tree_method`` is set to ``hist``.
-  - range: (0,1]
+* ``colsample_bytree``, ``colsample_bylevel``, ``colsample_bynode`` [default=1]
+  - This is a family of parameters for subsampling of columns.
+  - All ``colsample_by*`` parameters have a range of (0, 1], the default value of 1, and
+    specify the fraction of columns to be subsampled.
+  - ``colsample_bytree`` is the subsample ratio of columns when constructing each
+    tree. Subsampling occurs once for every tree constructed.
+  - ``colsample_bylevel`` is the subsample ratio of columns for each level. Subsampling
+    occurs once for every new depth level reached in a tree. Columns are subsampled from
+    the set of columns chosen for the current tree.
+  - ``colsample_bynode`` is the subsample ratio of columns for each node
+    (split). Subsampling occurs once every time a new split is evaluated. Columns are
+    subsampled from the set of columns chosen for the current level.
+  - ``colsample_by*`` parameters work cumulatively. For instance,
+    the combination ``{'colsample_bytree':0.5, 'colsample_bylevel':0.5,
+    'colsample_bynode':0.5}`` with 64 features will leave 4 features to choose from at
+    each split.
 
 * ``lambda`` [default=1, alias: ``reg_lambda``]
 
@@ -92,7 +110,7 @@ Parameters for Tree Booster
 * ``tree_method`` string [default= ``auto``]
 
   - The tree construction algorithm used in XGBoost. See description in the `reference paper <http://arxiv.org/abs/1603.02754>`_.
-  - Distributed and external memory version only support ``tree_method=approx``.
+  - XGBoost supports ``hist`` and ``approx`` for distributed training and only support ``approx`` for external memory version.
   - Choices: ``auto``, ``exact``, ``approx``, ``hist``, ``gpu_exact``, ``gpu_hist``
 
     - ``auto``: Use heuristic to choose the fastest method.
@@ -119,7 +137,7 @@ Parameters for Tree Booster
 
 * ``scale_pos_weight`` [default=1]
 
-  - Control the balance of positive and negative weights, useful for unbalanced classes. A typical value to consider: ``sum(negative instances) / sum(positive instances)``. See `Parameters Tuning </tutorials/param_tuning>`_ for more discussion. Also, see Higgs Kaggle competition demo for examples: `R <https://github.com/dmlc/xgboost/blob/master/demo/kaggle-higgs/higgs-train.R>`_, `py1 <https://github.com/dmlc/xgboost/blob/master/demo/kaggle-higgs/higgs-numpy.py>`_, `py2 <https://github.com/dmlc/xgboost/blob/master/demo/kaggle-higgs/higgs-cv.py>`_, `py3 <https://github.com/dmlc/xgboost/blob/master/demo/guide-python/cross_validation.py>`_.
+  - Control the balance of positive and negative weights, useful for unbalanced classes. A typical value to consider: ``sum(negative instances) / sum(positive instances)``. See :doc:`Parameters Tuning </tutorials/param_tuning>` for more discussion. Also, see Higgs Kaggle competition demo for examples: `R <https://github.com/dmlc/xgboost/blob/master/demo/kaggle-higgs/higgs-train.R>`_, `py1 <https://github.com/dmlc/xgboost/blob/master/demo/kaggle-higgs/higgs-numpy.py>`_, `py2 <https://github.com/dmlc/xgboost/blob/master/demo/kaggle-higgs/higgs-cv.py>`_, `py3 <https://github.com/dmlc/xgboost/blob/master/demo/guide-python/cross_validation.py>`_.
 
 * ``updater`` [default= ``grow_colmaker,prune``]
 
@@ -134,7 +152,7 @@ Parameters for Tree Booster
     - ``refresh``: refreshes tree's statistics and/or leaf values based on the current data. Note that no random subsampling of data rows is performed.
     - ``prune``: prunes the splits where loss < min_split_loss (or gamma).
 
-  - In a distributed setting, the implicit updater sequence value would be adjusted to ``grow_histmaker,prune``.
+  - In a distributed setting, the implicit updater sequence value would be adjusted to ``grow_histmaker,prune`` by default, and you can set ``tree_method`` as ``hist`` to use ``grow_histmaker``. 
 
 * ``refresh_leaf`` [default=1]
 
@@ -152,7 +170,7 @@ Parameters for Tree Booster
 
   - Controls a way new nodes are added to the tree.
   - Currently supported only if ``tree_method`` is set to ``hist``.
-  - Choices: ``depthwise``, ```lossguide``
+  - Choices: ``depthwise``, ``lossguide``
 
     - ``depthwise``: split at nodes closest to the root.
     - ``lossguide``: split at nodes with highest loss change.
@@ -174,6 +192,10 @@ Parameters for Tree Booster
     - ``cpu_predictor``: Multicore CPU prediction algorithm.
     - ``gpu_predictor``: Prediction using GPU. Default when ``tree_method`` is ``gpu_exact`` or ``gpu_hist``.
 
+* ``num_parallel_tree``, [default=1]
+  - Number of parallel trees constructed during each iteration. This
+     option is used to support boosted random forest
+
 Additional parameters for Dart Booster (``booster=dart``)
 =========================================================
 
@@ -241,8 +263,22 @@ Parameters for Linear Booster (``booster=gblinear``)
 
   - Choice of algorithm to fit linear model
 
-    - ``shotgun``: Parallel coordinate descent algorithm based on shotgun algorithm. Uses 'hogwild' parallelism and therefore produces a nondeterministic solution on each run. 
-    - ``coord_descent``: Ordinary coordinate descent algorithm. Also multithreaded but still produces a deterministic solution. 
+    - ``shotgun``: Parallel coordinate descent algorithm based on shotgun algorithm. Uses 'hogwild' parallelism and therefore produces a nondeterministic solution on each run.
+    - ``coord_descent``: Ordinary coordinate descent algorithm. Also multithreaded but still produces a deterministic solution.
+
+* ``feature_selector`` [default= ``cyclic``]
+
+  - Feature selection and ordering method
+
+    * ``cyclic``: Deterministic selection by cycling through features one at a time.
+    * ``shuffle``: Similar to ``cyclic`` but with random feature shuffling prior to each update.
+    * ``random``: A random (with replacement) coordinate selector.
+    * ``greedy``: Select coordinate with the greatest gradient magnitude.  It has ``O(num_feature^2)`` complexity. It is fully deterministic. It allows restricting the selection to ``top_k`` features per group with the largest magnitude of univariate weight change, by setting the ``top_k`` parameter. Doing so would reduce the complexity to ``O(num_feature*top_k)``.
+    * ``thrifty``: Thrifty, approximately-greedy feature selector. Prior to cyclic updates, reorders features in descending magnitude of their univariate weight changes. This operation is multithreaded and is a linear complexity approximation of the quadratic greedy selection. It allows restricting the selection to ``top_k`` features per group with the largest magnitude of univariate weight change, by setting the ``top_k`` parameter.
+
+* ``top_k`` [default=0]
+
+  - The number of top features to select in ``greedy`` and ``thrifty`` feature selector. The value of 0 means using all the features.
 
 Parameters for Tweedie Regression (``objective=reg:tweedie``)
 =============================================================
@@ -265,9 +301,6 @@ Specify the learning task and the corresponding learning objective. The objectiv
   - ``binary:logistic``: logistic regression for binary classification, output probability
   - ``binary:logitraw``: logistic regression for binary classification, output score before logistic transformation
   - ``binary:hinge``: hinge loss for binary classification. This makes predictions of 0 or 1, rather than producing probabilities.
-  - ``gpu:reg:linear``, ``gpu:reg:logistic``, ``gpu:binary:logistic``, ``gpu:binary:logitraw``: versions
-    of the corresponding objective functions evaluated on the GPU; note that like the GPU histogram algorithm,
-    they can only be used when the entire training session uses the same dataset
   - ``count:poisson`` --poisson regression for count data, output mean of poisson distribution
 
     - ``max_delta_step`` is set to 0.7 by default in poisson regression (used to safeguard optimization)
@@ -276,7 +309,9 @@ Specify the learning task and the corresponding learning objective. The objectiv
     Note that predictions are returned on the hazard ratio scale (i.e., as HR = exp(marginal_prediction) in the proportional hazard function ``h(t) = h0(t) * HR``).
   - ``multi:softmax``: set XGBoost to do multiclass classification using the softmax objective, you also need to set num_class(number of classes)
   - ``multi:softprob``: same as softmax, but output a vector of ``ndata * nclass``, which can be further reshaped to ``ndata * nclass`` matrix. The result contains predicted probability of each data point belonging to each class.
-  - ``rank:pairwise``: set XGBoost to do ranking task by minimizing the pairwise loss
+  - ``rank:pairwise``: Use LambdaMART to perform pairwise ranking where the pairwise loss is minimized
+  - ``rank:ndcg``: Use LambdaMART to perform list-wise ranking where `Normalized Discounted Cumulative Gain (NDCG) <http://en.wikipedia.org/wiki/NDCG>`_ is maximized
+  - ``rank:map``: Use LambdaMART to perform list-wise ranking where `Mean Average Precision (MAP) <http://en.wikipedia.org/wiki/Mean_average_precision#Mean_average_precision>`_ is maximized
   - ``reg:gamma``: gamma regression with log-link. Output is a mean of gamma distribution. It might be useful, e.g., for modeling insurance claims severity, or for any outcome that might be `gamma-distributed <https://en.wikipedia.org/wiki/Gamma_distribution#Applications>`_.
   - ``reg:tweedie``: Tweedie regression with log-link. It might be useful, e.g., for modeling total loss in insurance, or for any outcome that might be `Tweedie-distributed <https://en.wikipedia.org/wiki/Tweedie_distribution#Applications>`_.
 
@@ -299,8 +334,9 @@ Specify the learning task and the corresponding learning objective. The objectiv
     - ``merror``: Multiclass classification error rate. It is calculated as ``#(wrong cases)/#(all cases)``.
     - ``mlogloss``: `Multiclass logloss <http://scikit-learn.org/stable/modules/generated/sklearn.metrics.log_loss.html>`_.
     - ``auc``: `Area under the curve <http://en.wikipedia.org/wiki/Receiver_operating_characteristic#Area_under_curve>`_
+    - ``aucpr``: `Area under the PR curve <https://en.wikipedia.org/wiki/Precision_and_recall>`_
     - ``ndcg``: `Normalized Discounted Cumulative Gain <http://en.wikipedia.org/wiki/NDCG>`_
-    - ``map``: `Mean average precision <http://en.wikipedia.org/wiki/Mean_average_precision#Mean_average_precision>`_
+    - ``map``: `Mean Average Precision <http://en.wikipedia.org/wiki/Mean_average_precision#Mean_average_precision>`_
     - ``ndcg@n``, ``map@n``: 'n' can be assigned as an integer to cut off the top positions in the lists for evaluation.
     - ``ndcg-``, ``map-``, ``ndcg@n-``, ``map@n-``: In XGBoost, NDCG and MAP will evaluate the score of a list without any positive samples as 1. By adding "-" in the evaluation metric XGBoost will evaluate these score as 0 to be consistent under some conditions.
     - ``poisson-nloglik``: negative log-likelihood for Poisson regression
@@ -318,10 +354,6 @@ Command Line Parameters
 ***********************
 The following parameters are only used in the console version of XGBoost
 
-* ``use_buffer`` [default=1]
-
-  - Whether to create a binary buffer from text input. Doing so normally will speed up loading times
-
 * ``num_round``
 
   - The number of rounds for boosting
@@ -361,6 +393,10 @@ The following parameters are only used in the console version of XGBoost
 
   - Feature map, used for dumping model
 
+* ``dump_format`` [default= ``text``] options: ``text``, ``json``
+
+  - Format of model dump file
+
 * ``name_dump`` [default= ``dump.txt``]
 
   - Name of model dump file

doc/python/python_api.rst

@@ -2,6 +2,10 @@ Python API Reference
 ====================
 This page gives the Python API reference of xgboost, please also refer to Python Package Introduction for more information about python package.
 
+.. contents::
+  :backlinks: none
+  :local:
+
 Core Data Structure
 -------------------
 .. automodule:: xgboost.core
@@ -29,9 +33,15 @@ Scikit-Learn API
 .. automodule:: xgboost.sklearn
 .. autoclass:: xgboost.XGBRegressor
     :members:
+    :inherited-members:
     :show-inheritance:
 .. autoclass:: xgboost.XGBClassifier
     :members:
+    :inherited-members:
+    :show-inheritance:
+.. autoclass:: xgboost.XGBRanker
+    :members:
+    :inherited-members:
     :show-inheritance:
 
 Plotting API
@@ -43,3 +53,15 @@ Plotting API
 .. autofunction:: xgboost.plot_tree
 
 .. autofunction:: xgboost.to_graphviz
+
+.. _callback_api:
+
+Callback API
+------------
+.. autofunction:: xgboost.callback.print_evaluation
+
+.. autofunction:: xgboost.callback.record_evaluation
+
+.. autofunction:: xgboost.callback.reset_learning_rate
+
+.. autofunction:: xgboost.callback.early_stop

doc/python/python_intro.rst

@@ -48,9 +48,15 @@ The data is stored in a :py:class:`DMatrix <xgboost.DMatrix>` object.
     dtrain = xgb.DMatrix('train.csv?format=csv&label_column=0')
     dtest = xgb.DMatrix('test.csv?format=csv&label_column=0')
 
-  (Note that XGBoost does not support categorical features; if your data contains
-  categorical features, load it as a NumPy array first and then perform
-  `one-hot encoding <http://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OneHotEncoder.html>`_.)
+  .. note:: Categorical features not supported
+
+    Note that XGBoost does not support categorical features; if your data contains
+    categorical features, load it as a NumPy array first and then perform
+    `one-hot encoding <http://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OneHotEncoder.html>`_.
+
+  .. note:: Use Pandas to load CSV files with headers
+
+    Currently, the DMLC data parser cannot parse CSV files with headers. Use Pandas (see below) to read CSV files with headers.
 
 * To load a NumPy array into :py:class:`DMatrix <xgboost.DMatrix>`:
 
@@ -95,6 +101,10 @@ The data is stored in a :py:class:`DMatrix <xgboost.DMatrix>` object.
     w = np.random.rand(5, 1)
     dtrain = xgb.DMatrix(data, label=label, missing=-999.0, weight=w)
 
+When performing ranking tasks, the number of weights should be equal
+to number of groups.
+
+
 Setting Parameters
 ------------------
 XGBoost can use either a list of pairs or a dictionary to set :doc:`parameters </parameter>`. For instance:
@@ -155,6 +165,10 @@ A saved model can be loaded as follows:
   bst = xgb.Booster({'nthread': 4})  # init model
   bst.load_model('model.bin')  # load data
 
+Methods including `update` and `boost` from `xgboost.Booster` are designed for
+internal usage only.  The wrapper function `xgboost.train` does some
+pre-configuration including setting up caches and some other parameters.
+
 Early Stopping
 --------------
 If you have a validation set, you can use early stopping to find the optimal number of boosting rounds.
@@ -209,4 +223,3 @@ When you use ``IPython``, you can use the :py:meth:`xgboost.to_graphviz` functio
 .. code-block:: python
 
   xgb.to_graphviz(bst, num_trees=2)
-

doc/requirements.txt

@@ -3,3 +3,6 @@ mock
 guzzle_sphinx_theme
 breathe
 sh>=1.12.14
+matplotlib>=2.1
+graphviz
+numpy

doc/tutorials/aws_yarn.rst

@@ -1,216 +1,8 @@
 ###############################
 Distributed XGBoost YARN on AWS
 ###############################
-This is a step-by-step tutorial on how to setup and run distributed `XGBoost <https://github.com/dmlc/xgboost>`_
-on an AWS EC2 cluster. Distributed XGBoost runs on various platforms such as MPI, SGE and Hadoop YARN.
-In this tutorial, we use YARN as an example since this is a widely used solution for distributed computing.
+[This page is under construction.]
 
 .. note:: XGBoost with Spark
 
   If you are preprocessing training data with Spark, consider using :doc:`XGBoost4J-Spark </jvm/xgboost4j_spark_tutorial>`.
-
-************
-Prerequisite
-************
-We need to get a `AWS key-pair <http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-key-pairs.html>`_
-to access the AWS services. Let us assume that we are using a key ``mykey`` and  the corresponding permission file ``mypem.pem``.
-
-We also need `AWS credentials <https://docs.aws.amazon.com/cli/latest/userguide/cli-chap-getting-started.html>`_,
-which includes an ``ACCESS_KEY_ID`` and a ``SECRET_ACCESS_KEY``.
-
-Finally, we will need a S3 bucket to host the data and the model, ``s3://mybucket/``
-
-***************************
-Setup a Hadoop YARN Cluster
-***************************
-This sections shows how to start a Hadoop YARN cluster from scratch.
-You can skip this step if you have already have one.
-We will be using `yarn-ec2 <https://github.com/tqchen/yarn-ec2>`_ to start the cluster.
-
-We can first clone the yarn-ec2 script by the following command.
-
-.. code-block:: bash
-
-  git clone https://github.com/tqchen/yarn-ec2
-
-To use the script, we must set the environment variables ``AWS_ACCESS_KEY_ID`` and
-``AWS_SECRET_ACCESS_KEY`` properly. This can be done by adding the following two lines in
-``~/.bashrc`` (replacing the strings with the correct ones)
-
-.. code-block:: bash
-
-  export AWS_ACCESS_KEY_ID=[your access ID]
-  export AWS_SECRET_ACCESS_KEY=[your secret access key]
-
-Now we can launch a master machine of the cluster from EC2:
-
-.. code-block:: bash
-
-  ./yarn-ec2 -k mykey -i mypem.pem launch xgboost
-
-Wait a few mininutes till the master machine gets up.
-
-After the master machine gets up, we can query the public DNS of the master machine using the following command.
-
-.. code-block:: bash
-
-  ./yarn-ec2 -k mykey -i mypem.pem get-master xgboost
-
-It will show the public DNS of the master machine like ``ec2-xx-xx-xx.us-west-2.compute.amazonaws.com``
-Now we can open the browser, and type (replace the DNS with the master DNS)
-
-.. code-block:: none
-
-  ec2-xx-xx-xx.us-west-2.compute.amazonaws.com:8088
-
-This will show the job tracker of the YARN cluster. Note that we may have to wait a few minutes before the master finishes bootstrapping and starts the
-job tracker.
-
-After the master machine gets up, we can freely add more slave machines to the cluster.
-The following command add m3.xlarge instances to the cluster.
-
-.. code-block:: bash
-
-  ./yarn-ec2 -k mykey -i mypem.pem -t m3.xlarge -s 2 addslave xgboost
-
-We can also choose to add two spot instances
-
-.. code-block:: bash
-
-  ./yarn-ec2 -k mykey -i mypem.pem -t m3.xlarge -s 2 addspot xgboost
-
-The slave machines will start up, bootstrap  and report to the master.
-You can check if the slave machines are connected by clicking on the Nodes link on the job tracker.
-Or simply type the following URL (replace DNS ith the master DNS)
-
-.. code-block:: none
-
-  ec2-xx-xx-xx.us-west-2.compute.amazonaws.com:8088/cluster/nodes
-
-One thing we should note is that not all the links in the job tracker work.
-This is due to that many of them use the private IP of AWS, which can only be accessed by EC2.
-We can use ssh proxy to access these packages.
-Now that we have set up a cluster with one master and two slaves, we are ready to run the experiment.
-
-*********************
-Build XGBoost with S3
-*********************
-We can log into the master machine by the following command.
-
-.. code-block:: bash
-
-  ./yarn-ec2 -k mykey -i mypem.pem login xgboost
-
-We will be using S3 to host the data and the result model, so the data won't get lost after the cluster shutdown.
-To do so, we will need to build XGBoost with S3 support. The only thing we need to do is to set ``USE_S3``
-variable to be true. This can be achieved by the following command.
-
-.. code-block:: bash
-
-  git clone --recursive https://github.com/dmlc/xgboost
-  cd xgboost
-  cp make/config.mk config.mk
-  echo "USE_S3=1" >> config.mk
-  make -j4
-
-Now we have built the XGBoost with S3 support. You can also enable HDFS support if you plan to store data on HDFS by turning on ``USE_HDFS`` option.
-XGBoost also relies on the environment variable to access S3, so you will need to add the following two lines to ``~/.bashrc`` (replacing the strings with the correct ones)
-on the master machine as well.
-
-.. code-block:: bash
-
-  export AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE
-  export AWS_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
-  export BUCKET=mybucket
-
-*******************
-Host the Data on S3
-*******************
-In this example, we will copy the example dataset in XGBoost to the S3 bucket as input.
-In normal usecases, the dataset is usually created from existing distributed processing pipeline.
-We can use `s3cmd <http://s3tools.org/s3cmd>`_ to copy the data into mybucket (replace ``${BUCKET}`` with the real bucket name).
-
-.. code-block:: bash
-
-  cd xgboost
-  s3cmd put demo/data/agaricus.txt.train s3://${BUCKET}/xgb-demo/train/
-  s3cmd put demo/data/agaricus.txt.test s3://${BUCKET}/xgb-demo/test/
-
-***************
-Submit the Jobs
-***************
-Now everything is ready, we can submit the XGBoost distributed job to the YARN cluster.
-We will use the `dmlc-submit <https://github.com/dmlc/dmlc-core/tree/master/tracker>`_ script to submit the job.
-
-Now we can run the following script in the distributed training folder (replace ``${BUCKET}`` with the real bucket name)
-
-.. code-block:: bash
-
-  cd xgboost/demo/distributed-training
-  # Use dmlc-submit to submit the job.
-  ../../dmlc-core/tracker/dmlc-submit --cluster=yarn --num-workers=2 --worker-cores=2\
-      ../../xgboost mushroom.aws.conf nthread=2\
-      data=s3://${BUCKET}/xgb-demo/train\
-      eval[test]=s3://${BUCKET}/xgb-demo/test\
-      model_dir=s3://${BUCKET}/xgb-demo/model
-
-All the configurations such as ``data`` and ``model_dir`` can also be directly written into the configuration file.
-Note that we only specified the folder path to the file, instead of the file name.
-XGBoost will read in all the files in that folder as the training and evaluation data.
-
-In this command, we are using two workers, and each worker uses two running threads.
-XGBoost can benefit from using multiple cores in each worker.
-A common choice of working cores can range from 4 to 8.
-The trained model will be saved into the specified model folder. You can browse the model folder.
-
-.. code-block:: bash
-
-  s3cmd ls s3://${BUCKET}/xgb-demo/model/
-
-The following is an example output from distributed training.
-
-.. code-block:: none
-
-  16/02/26 05:41:59 INFO dmlc.Client: jobname=DMLC[nworker=2]:xgboost,username=ubuntu
-  16/02/26 05:41:59 INFO dmlc.Client: Submitting application application_1456461717456_0015
-  16/02/26 05:41:59 INFO impl.YarnClientImpl: Submitted application application_1456461717456_0015
-  2016-02-26 05:42:05,230 INFO @tracker All of 2 nodes getting started
-  2016-02-26 05:42:14,027 INFO [05:42:14] [0]  test-error:0.016139        train-error:0.014433
-  2016-02-26 05:42:14,186 INFO [05:42:14] [1]  test-error:0.000000        train-error:0.001228
-  2016-02-26 05:42:14,947 INFO @tracker All nodes finishes job
-  2016-02-26 05:42:14,948 INFO @tracker 9.71754479408 secs between node start and job finish
-  Application application_1456461717456_0015 finished with state FINISHED at 1456465335961
-
-*****************
-Analyze the Model
-*****************
-After the model is trained, we can analyse the learnt model and use it for future prediction tasks.
-XGBoost is a portable framework, meaning the models in all platforms are *exchangeable*.
-This means we can load the trained model in python/R/Julia and take benefit of data science pipelines
-in these languages to do model analysis and prediction.
-
-For example, you can use `this IPython notebook <https://github.com/dmlc/xgboost/tree/master/demo/distributed-training/plot_model.ipynb>`_
-to plot feature importance and visualize the learnt model.
-
-***************
-Troubleshooting
-***************
-
-If you encounter a problem, the best way might be to use the following command
-to get logs of stdout and stderr of the containers and check what causes the problem.
-
-.. code-block:: bash
-
-  yarn logs -applicationId yourAppId
-
-*****************
-Future Directions
-*****************
-You have learned to use distributed XGBoost on YARN in this tutorial.
-XGBoost is a portable and scalable framework for gradient boosting.
-You can check out more examples and resources in the `resources page <https://github.com/dmlc/xgboost/blob/master/demo/README.md>`_.
-
-The project goal is to make the best scalable machine learning solution available to all platforms.
-The API is designed to be able to portable, and the same code can also run on other platforms such as MPI and SGE.
-XGBoost is actively evolving and we are working on even more exciting features
-such as distributed XGBoost python/R package.

doc/tutorials/external_memory.rst

@@ -13,6 +13,10 @@ The external memory version takes in the following filename format:
 The ``filename`` is the normal path to libsvm file you want to load in, and ``cacheprefix`` is a
 path to a cache file that XGBoost will use for external memory cache.
 
+.. note:: External memory is not available with GPU algorithms
+
+  External memory is not available when ``tree_method`` is set to ``gpu_exact`` or ``gpu_hist``.
+
 The following code was extracted from `demo/guide-python/external_memory.py <https://github.com/dmlc/xgboost/blob/master/demo/guide-python/external_memory.py>`_:
 
 .. code-block:: python

doc/tutorials/feature_interaction_constraint.rst

@@ -0,0 +1,177 @@
+###############################
+Feature Interaction Constraints
+###############################
+
+The decision tree is a powerful tool to discover interaction among independent
+variables (features). Variables that appear together in a traversal path
+are interacting with one another, since the condition of a child node is
+predicated on the condition of the parent node. For example, the highlighted
+red path in the diagram below contains three variables: :math:`x_1`, :math:`x_7`,
+and :math:`x_{10}`, so the highlighted prediction (at the highlighted leaf node)
+is the product of interaction between :math:`x_1`, :math:`x_7`, and
+:math:`x_{10}`.
+
+.. plot::
+  :nofigs:
+
+  from graphviz import Source
+  source = r"""
+    digraph feature_interaction_illustration1 {
+      graph [fontname = "helvetica"];
+      node [fontname = "helvetica"];
+      edge [fontname = "helvetica"];
+      0 [label=<x<SUB><FONT POINT-SIZE="11">10</FONT></SUB> &lt; -1.5 ?>, shape=box, color=red, fontcolor=red];
+      1 [label=<x<SUB><FONT POINT-SIZE="11">2</FONT></SUB> &lt; 2 ?>, shape=box];
+      2 [label=<x<SUB><FONT POINT-SIZE="11">7</FONT></SUB> &lt; 0.3 ?>, shape=box, color=red, fontcolor=red];
+      3 [label="...", shape=none];
+      4 [label="...", shape=none];
+      5 [label=<x<SUB><FONT POINT-SIZE="11">1</FONT></SUB> &lt; 0.5 ?>, shape=box, color=red, fontcolor=red];
+      6 [label="...", shape=none];
+      7 [label="...", shape=none];
+      8 [label="Predict +1.3", color=red, fontcolor=red];
+      0 -> 1 [labeldistance=2.0, labelangle=45, headlabel="Yes/Missing           "];
+      0 -> 2 [labeldistance=2.0, labelangle=-45,
+              headlabel="No", color=red, fontcolor=red];
+      1 -> 3 [labeldistance=2.0, labelangle=45, headlabel="Yes"];
+      1 -> 4 [labeldistance=2.0, labelangle=-45, headlabel="             No/Missing"];
+      2 -> 5 [labeldistance=2.0, labelangle=-45, headlabel="Yes",
+              color=red, fontcolor=red];
+      2 -> 6 [labeldistance=2.0, labelangle=-45, headlabel="           No/Missing"];
+      5 -> 7;
+      5 -> 8 [color=red];
+    }
+  """
+  Source(source, format='png').render('../_static/feature_interaction_illustration1', view=False)
+  Source(source, format='svg').render('../_static/feature_interaction_illustration1', view=False)
+
+.. raw:: html
+
+  <p>
+  <img src="../_static/feature_interaction_illustration1.svg"
+    onerror="this.src='../_static/feature_interaction_illustration1.png'; this.onerror=null;">
+  </p>
+
+When the tree depth is larger than one, many variables interact on
+the sole basis of minimizing training loss, and the resulting decision tree may
+capture a spurious relationship (noise) rather than a legitimate relationship
+that generalizes across different datasets. **Feature interaction constraints**
+allow users to decide which variables are allowed to interact and which are not.
+
+Potential benefits include:
+
+* Better predictive performance from focusing on interactions that work --
+  whether through domain specific knowledge or algorithms that rank interactions
+* Less noise in predictions; better generalization
+* More control to the user on what the model can fit. For example, the user may
+  want to exclude some interactions even if they perform well due to regulatory
+  constraints
+
+****************
+A Simple Example
+****************
+
+Feature interaction constraints are expressed in terms of groups of variables
+that are allowed to interact. For example, the constraint
+``[0, 1]`` indicates that variables :math:`x_0` and :math:`x_1` are allowed to
+interact with each other but with no other variable. Similarly, ``[2, 3, 4]``
+indicates that :math:`x_2`, :math:`x_3`, and :math:`x_4` are allowed to
+interact with one another but with no other variable. A set of feature
+interaction constraints is expressed as a nested list, e.g.
+``[[0, 1], [2, 3, 4]]``, where each inner list is a group of indices of features
+that are allowed to interact with each other.
+
+In the following diagram, the left decision tree is in violation of the first
+constraint (``[0, 1]``), whereas the right decision tree complies with both the
+first and second constraints (``[0, 1]``, ``[2, 3, 4]``).
+
+.. plot::
+  :nofigs:
+
+  from graphviz import Source
+  source = r"""
+    digraph feature_interaction_illustration2 {
+      graph [fontname = "helvetica"];
+      node [fontname = "helvetica"];
+      edge [fontname = "helvetica"];
+      0 [label=<x<SUB><FONT POINT-SIZE="11">0</FONT></SUB> &lt; 5.0 ?>, shape=box];
+      1 [label=<x<SUB><FONT POINT-SIZE="11">2</FONT></SUB> &lt; -3.0 ?>, shape=box];
+      2 [label="+0.6"];
+      3 [label="-0.4"];
+      4 [label="+1.2"];
+      0 -> 1 [labeldistance=2.0, labelangle=45, headlabel="Yes/Missing           "];
+      0 -> 2 [labeldistance=2.0, labelangle=-45, headlabel="No"];
+      1 -> 3 [labeldistance=2.0, labelangle=45, headlabel="Yes"];
+      1 -> 4 [labeldistance=2.0, labelangle=-45, headlabel="           No/Missing"];
+    }
+  """
+  Source(source, format='png').render('../_static/feature_interaction_illustration2', view=False)
+  Source(source, format='svg').render('../_static/feature_interaction_illustration2', view=False)
+
+.. plot::
+  :nofigs:
+
+  from graphviz import Source
+  source = r"""
+    digraph feature_interaction_illustration3 {
+      graph [fontname = "helvetica"];
+      node [fontname = "helvetica"];
+      edge [fontname = "helvetica"];
+      0 [label=<x<SUB><FONT POINT-SIZE="11">3</FONT></SUB> &lt; 2.5 ?>, shape=box];
+      1 [label="+1.6"];
+      2 [label=<x<SUB><FONT POINT-SIZE="11">2</FONT></SUB> &lt; -1.2 ?>, shape=box];
+      3 [label="+0.1"];
+      4 [label="-0.3"];
+      0 -> 1 [labeldistance=2.0, labelangle=45, headlabel="Yes"];
+      0 -> 2 [labeldistance=2.0, labelangle=-45, headlabel="           No/Missing"];
+      2 -> 3 [labeldistance=2.0, labelangle=45, headlabel="Yes/Missing           "];
+      2 -> 4 [labeldistance=2.0, labelangle=-45, headlabel="No"];
+    }
+  """
+  Source(source, format='png').render('../_static/feature_interaction_illustration3', view=False)
+  Source(source, format='svg').render('../_static/feature_interaction_illustration3', view=False)
+
+.. raw:: html
+
+  <p>
+  <img src="../_static/feature_interaction_illustration2.svg"
+       onerror="this.src='../_static/feature_interaction_illustration2.png'; this.onerror=null;">
+  <img src="../_static/feature_interaction_illustration3.svg"
+       onerror="this.src='../_static/feature_interaction_illustration3.png'; this.onerror=null;">
+  </p>
+
+****************************************************
+Enforcing Feature Interaction Constraints in XGBoost
+****************************************************
+
+It is very simple to enforce feature interaction constraints in XGBoost.  Here we will
+give an example using Python, but the same general idea generalizes to other
+platforms.
+
+Suppose the following code fits your model without feature interaction constraints:
+
+.. code-block:: python
+
+  model_no_constraints = xgb.train(params, dtrain,
+                                   num_boost_round = 1000, evals = evallist,
+                                   early_stopping_rounds = 10)
+
+Then fitting with feature interaction constraints only requires adding a single
+parameter:
+
+.. code-block:: python
+
+  params_constrained = params.copy()
+  # Use nested list to define feature interaction constraints
+  params_constrained['interaction_constraints'] = '[[0, 2], [1, 3, 4], [5, 6]]'
+  # Features 0 and 2 are allowed to interact with each other but with no other feature
+  # Features 1, 3, 4 are allowed to interact with one another but with no other feature
+  # Features 5 and 6 are allowed to interact with each other but with no other feature
+
+  model_with_constraints = xgb.train(params_constrained, dtrain,
+                                     num_boost_round = 1000, evals = evallist,
+                                     early_stopping_rounds = 10)
+
+**Choice of tree construction algorithm**. To use feature interaction
+constraints, be sure to set the ``tree_method`` parameter to either ``exact``
+or ``hist``. Currently, GPU algorithms (``gpu_hist``, ``gpu_exact``) do not
+support feature interaction constraints.

doc/tutorials/index.rst

@@ -14,6 +14,7 @@ See `Awesome XGBoost <https://github.com/dmlc/xgboost/tree/master/demo>`_ for mo
   Distributed XGBoost with XGBoost4J-Spark <https://xgboost.readthedocs.io/en/latest/jvm/xgboost4j_spark_tutorial.html>
   dart
   monotonic
+  feature_interaction_constraint
   input_format
   param_tuning
   external_memory

doc/tutorials/model.rst

@@ -223,7 +223,7 @@ In this equation, :math:`w_j` are independent with respect to each other, the fo
   w_j^\ast &= -\frac{G_j}{H_j+\lambda}\\
   \text{obj}^\ast &= -\frac{1}{2} \sum_{j=1}^T \frac{G_j^2}{H_j+\lambda} + \gamma T
 
-The last equation measures *how good* a tree structure :math:`$q(x)` is.
+The last equation measures *how good* a tree structure :math:`q(x)` is.
 
 .. image:: https://raw.githubusercontent.com/dmlc/web-data/master/xgboost/model/struct_score.png
   :width: 100%

doc/tutorials/monotonic.rst

@@ -82,7 +82,7 @@ Some other examples:
 - ``(1,0)``: An increasing constraint on the first predictor and no constraint on the second.
 - ``(0,-1)``: No constraint on the first predictor and a decreasing constraint on the second.
 
-**Choise of tree construction algorithm**. To use monotonic constraints, be
+**Choice of tree construction algorithm**. To use monotonic constraints, be
 sure to set the ``tree_method`` parameter to one of ``exact``, ``hist``, and
 ``gpu_hist``.
 

include/xgboost/base.h

@@ -16,7 +16,7 @@
  */
 #ifndef XGBOOST_STRICT_R_MODE
 #define XGBOOST_STRICT_R_MODE 0
-#endif
+#endif  // XGBOOST_STRICT_R_MODE
 
 /*!
  * \brief Whether always log console message with time.
@@ -26,21 +26,21 @@
  */
 #ifndef XGBOOST_LOG_WITH_TIME
 #define XGBOOST_LOG_WITH_TIME 1
-#endif
+#endif  // XGBOOST_LOG_WITH_TIME
 
 /*!
  * \brief Whether customize the logger outputs.
  */
 #ifndef XGBOOST_CUSTOMIZE_LOGGER
 #define XGBOOST_CUSTOMIZE_LOGGER XGBOOST_STRICT_R_MODE
-#endif
+#endif  // XGBOOST_CUSTOMIZE_LOGGER
 
 /*!
  * \brief Whether to customize global PRNG.
  */
 #ifndef XGBOOST_CUSTOMIZE_GLOBAL_PRNG
 #define XGBOOST_CUSTOMIZE_GLOBAL_PRNG XGBOOST_STRICT_R_MODE
-#endif
+#endif  // XGBOOST_CUSTOMIZE_GLOBAL_PRNG
 
 /*!
  * \brief Check if alignas(*) keyword is supported. (g++ 4.8 or higher)
@@ -49,7 +49,7 @@
 #define XGBOOST_ALIGNAS(X) alignas(X)
 #else
 #define XGBOOST_ALIGNAS(X)
-#endif
+#endif  // defined(__GNUC__) && ((__GNUC__ == 4 && __GNUC_MINOR__ >= 8) || __GNUC__ > 4)
 
 #if defined(__GNUC__) && ((__GNUC__ == 4 && __GNUC_MINOR__ >= 8) || __GNUC__ > 4) && \
     !defined(__CUDACC__)
@@ -64,7 +64,7 @@
 #else
 #define XGBOOST_PARALLEL_SORT(X, Y, Z) std::sort((X), (Y), (Z))
 #define XGBOOST_PARALLEL_STABLE_SORT(X, Y, Z) std::stable_sort((X), (Y), (Z))
-#endif
+#endif  // GLIBC VERSION
 
 /*!
  * \brief Tag function as usable by device
@@ -73,7 +73,7 @@
 #define XGBOOST_DEVICE __host__ __device__
 #else
 #define XGBOOST_DEVICE
-#endif
+#endif  // defined (__CUDA__) || defined(__NVCC__)
 
 /*! \brief namespace of xgboost*/
 namespace xgboost {
@@ -215,7 +215,11 @@ using bst_omp_uint = dmlc::omp_uint;  // NOLINT
 #if __GNUC__ == 4 && __GNUC_MINOR__ < 8
 #define override
 #define final
-#endif
-#endif
+#endif  // __GNUC__ == 4 && __GNUC_MINOR__ < 8
+#endif  // DMLC_USE_CXX11 && defined(__GNUC__) && !defined(__clang_version__)
 }  // namespace xgboost
+
+/* Always keep this #include at the bottom of xgboost/base.h */
+#include <xgboost/build_config.h>
+
 #endif  // XGBOOST_BASE_H_

include/xgboost/build_config.h

@@ -0,0 +1,18 @@
+/*!
+ * Copyright 2019 by Contributors
+ * \file build_config.h
+ */
+#ifndef XGBOOST_BUILD_CONFIG_H_
+#define XGBOOST_BUILD_CONFIG_H_
+
+/* default logic for software pre-fetching */
+#if (defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_AMD64))) || defined(__INTEL_COMPILER)
+// Enable _mm_prefetch for Intel compiler and MSVC+x86
+  #define XGBOOST_MM_PREFETCH_PRESENT
+  #define XGBOOST_BUILTIN_PREFETCH_PRESENT
+#elif defined(__GNUC__)
+// Enable __builtin_prefetch for GCC
+#define XGBOOST_BUILTIN_PREFETCH_PRESENT
+#endif  // GUARDS
+
+#endif  // XGBOOST_BUILD_CONFIG_H_

include/xgboost/c_api.h

@@ -10,11 +10,12 @@
 #ifdef __cplusplus
 #define XGB_EXTERN_C extern "C"
 #include <cstdio>
+#include <cstdint>
 #else
 #define XGB_EXTERN_C
 #include <stdio.h>
 #include <stdint.h>
-#endif
+#endif  // __cplusplus
 
 // XGBoost C API will include APIs in Rabit C API
 #include <rabit/c_api.h>
@@ -23,7 +24,7 @@
 #define XGB_DLL XGB_EXTERN_C __declspec(dllexport)
 #else
 #define XGB_DLL XGB_EXTERN_C
-#endif
+#endif  // defined(_MSC_VER) || defined(_WIN32)
 
 // manually define unsigned long
 typedef uint64_t bst_ulong;  // NOLINT(*)
@@ -49,7 +50,7 @@ typedef struct {  // NOLINT(*)
   long* offset; // NOLINT(*)
 #else
   int64_t* offset;  // NOLINT(*)
-#endif
+#endif  // __APPLE__
   /*! \brief labels of each instance */
   float* label;
   /*! \brief weight of each instance, can be NULL */
@@ -562,7 +563,7 @@ XGB_DLL int XGBoosterGetAttr(BoosterHandle handle,
  *
  * \param handle handle
  * \param key The key of the attribute.
- * \param value The value to be saved. 
+ * \param value The value to be saved.
  *              If nullptr, the attribute would be deleted.
  * \return 0 when success, -1 when failure happens
  */

include/xgboost/data.h

@@ -9,12 +9,18 @@
 
 #include <dmlc/base.h>
 #include <dmlc/data.h>
+#include <rabit/rabit.h>
 #include <cstring>
 #include <memory>
 #include <numeric>
+#include <algorithm>
 #include <string>
 #include <vector>
 #include "./base.h"
+#include "../../src/common/span.h"
+#include "../../src/common/group_data.h"
+
+#include "../../src/common/host_device_vector.h"
 
 namespace xgboost {
 // forward declare learner.
@@ -40,7 +46,7 @@ class MetaInfo {
   /*! \brief number of nonzero entries in the data */
   uint64_t num_nonzero_{0};
   /*! \brief label of each instance */
-  std::vector<bst_float> labels_;
+  HostDeviceVector<bst_float> labels_;
   /*!
    * \brief specified root index of each instance,
    *  can be used for multi task setting
@@ -52,7 +58,7 @@ class MetaInfo {
    */
   std::vector<bst_uint> group_ptr_;
   /*! \brief weights of each instance, optional */
-  std::vector<bst_float> weights_;
+  HostDeviceVector<bst_float> weights_;
   /*! \brief session-id of each instance, optional */
   std::vector<uint64_t> qids_;
   /*!
@@ -60,7 +66,7 @@ class MetaInfo {
    * if specified, xgboost will start from this init margin
    * can be used to specify initial prediction to boost from.
    */
-  std::vector<bst_float> base_margin_;
+  HostDeviceVector<bst_float> base_margin_;
   /*! \brief version flag, used to check version of this info */
   static const int kVersion = 2;
   /*! \brief version that introduced qid field */
@@ -73,7 +79,7 @@ class MetaInfo {
    * \return The weight.
    */
   inline bst_float GetWeight(size_t i) const {
-    return weights_.size() != 0 ?  weights_[i] : 1.0f;
+    return weights_.Size() != 0 ?  weights_.HostVector()[i] : 1.0f;
   }
   /*!
    * \brief Get the root index of i-th instance.
@@ -85,12 +91,12 @@ class MetaInfo {
   }
   /*! \brief get sorted indexes (argsort) of labels by absolute value (used by cox loss) */
   inline const std::vector<size_t>& LabelAbsSort() const {
-    if (label_order_cache_.size() == labels_.size()) {
+    if (label_order_cache_.size() == labels_.Size()) {
       return label_order_cache_;
     }
-    label_order_cache_.resize(labels_.size());
+    label_order_cache_.resize(labels_.Size());
     std::iota(label_order_cache_.begin(), label_order_cache_.end(), 0);
-    const auto l = labels_;
+    const auto& l = labels_.HostVector();
     XGBOOST_PARALLEL_SORT(label_order_cache_.begin(), label_order_cache_.end(),
               [&l](size_t i1, size_t i2) {return std::abs(l[i1]) < std::abs(l[i2]);});
 
@@ -133,7 +139,7 @@ struct Entry {
   /*!
    * \brief constructor with index and value
    * \param index The feature or row index.
-   * \param fvalue THe feature value.
+   * \param fvalue The feature value.
    */
   Entry(bst_uint index, bst_float fvalue) : index(index), fvalue(fvalue) {}
   /*! \brief reversely compare feature values */
@@ -146,33 +152,34 @@ struct Entry {
 };
 
 /*!
- * \brief in-memory storage unit of sparse batch
+ * \brief In-memory storage unit of sparse batch, stored in CSR format.
  */
 class SparsePage {
  public:
-  std::vector<size_t> offset;
+  // Offset for each row.
+  HostDeviceVector<size_t> offset;
   /*! \brief the data of the segments */
-  std::vector<Entry> data;
+  HostDeviceVector<Entry> data;
 
   size_t base_rowid;
+
   /*! \brief an instance of sparse vector in the batch */
-  struct Inst {
-    /*! \brief pointer to the elements*/
-    const Entry *data{nullptr};
-    /*! \brief length of the instance */
-    bst_uint length{0};
-    /*! \brief constructor */
-    Inst()  = default;
-    Inst(const Entry *data, bst_uint length) : data(data), length(length) {}
-    /*! \brief get i-th pair in the sparse vector*/
-    inline const Entry& operator[](size_t i) const {
-      return data[i];
-    }
-  };
+  using Inst = common::Span<Entry const>;
 
   /*! \brief get i-th row from the batch */
   inline Inst operator[](size_t i) const {
-    return {data.data() + offset[i], static_cast<bst_uint>(offset[i + 1] - offset[i])};
+    const auto& data_vec = data.HostVector();
+    const auto& offset_vec = offset.HostVector();
+    size_t size;
+    // in distributed mode, some partitions may not get any instance for a feature. Therefore
+    // we should set the size as zero
+    if (rabit::IsDistributed() && i + 1 >= offset_vec.size()) {
+      size = 0;
+    } else {
+      size = offset_vec[i + 1] - offset_vec[i];
+    }
+    return {data_vec.data() + offset_vec[i],
+            static_cast<Inst::index_type>(size)};
   }
 
   /*! \brief constructor */
@@ -181,72 +188,153 @@ class SparsePage {
   }
   /*! \return number of instance in the page */
   inline size_t Size() const {
-    return offset.size() - 1;
+    return offset.Size() - 1;
   }
   /*! \return estimation of memory cost of this page */
   inline size_t MemCostBytes() const {
-    return offset.size() * sizeof(size_t) + data.size() * sizeof(Entry);
+    return offset.Size() * sizeof(size_t) + data.Size() * sizeof(Entry);
   }
   /*! \brief clear the page */
   inline void Clear() {
     base_rowid = 0;
-    offset.clear();
-    offset.push_back(0);
-    data.clear();
+    auto& offset_vec = offset.HostVector();
+    offset_vec.clear();
+    offset_vec.push_back(0);
+    data.HostVector().clear();
+  }
+
+  SparsePage GetTranspose(int num_columns) const {
+    SparsePage transpose;
+    common::ParallelGroupBuilder<Entry> builder(&transpose.offset.HostVector(),
+                                                &transpose.data.HostVector());
+    const int nthread = omp_get_max_threads();
+    builder.InitBudget(num_columns, nthread);
+    long batch_size = static_cast<long>(this->Size());  // NOLINT(*)
+#pragma omp parallel for schedule(static)
+    for (long i = 0; i < batch_size; ++i) {  // NOLINT(*)
+      int tid = omp_get_thread_num();
+      auto inst = (*this)[i];
+      for (bst_uint j = 0; j < inst.size(); ++j) {
+        builder.AddBudget(inst[j].index, tid);
+      }
+    }
+    builder.InitStorage();
+#pragma omp parallel for schedule(static)
+    for (long i = 0; i < batch_size; ++i) {  // NOLINT(*)
+      int tid = omp_get_thread_num();
+      auto inst = (*this)[i];
+      for (bst_uint j = 0; j < inst.size(); ++j) {
+        builder.Push(
+            inst[j].index,
+            Entry(static_cast<bst_uint>(this->base_rowid + i), inst[j].fvalue),
+            tid);
+      }
+    }
+    return transpose;
+  }
+
+  void SortRows() {
+    auto ncol = static_cast<bst_omp_uint>(this->Size());
+#pragma omp parallel for schedule(dynamic, 1)
+    for (bst_omp_uint i = 0; i < ncol; ++i) {
+      if (this->offset.HostVector()[i] < this->offset.HostVector()[i + 1]) {
+        std::sort(
+            this->data.HostVector().begin() + this->offset.HostVector()[i],
+            this->data.HostVector().begin() + this->offset.HostVector()[i + 1],
+            Entry::CmpValue);
+      }
+    }
   }
 
   /*!
    * \brief Push row block into the page.
    * \param batch the row batch.
    */
-  inline void Push(const dmlc::RowBlock<uint32_t>& batch) {
-    data.reserve(data.size() + batch.offset[batch.size] - batch.offset[0]);
-    offset.reserve(offset.size() + batch.size);
-    CHECK(batch.index != nullptr);
-    for (size_t i = 0; i < batch.size; ++i) {
-      offset.push_back(offset.back() + batch.offset[i + 1] - batch.offset[i]);
-    }
-    for (size_t i = batch.offset[0]; i < batch.offset[batch.size]; ++i) {
-      uint32_t index = batch.index[i];
-      bst_float fvalue = batch.value == nullptr ? 1.0f : batch.value[i];
-      data.emplace_back(index, fvalue);
-    }
-    CHECK_EQ(offset.back(), data.size());
-  }
+  void Push(const dmlc::RowBlock<uint32_t>& batch);
   /*!
    * \brief Push a sparse page
    * \param batch the row page
    */
-  inline void Push(const SparsePage &batch) {
-    size_t top = offset.back();
-    data.resize(top + batch.data.size());
-    std::memcpy(dmlc::BeginPtr(data) + top,
-                dmlc::BeginPtr(batch.data),
-                sizeof(Entry) * batch.data.size());
-    size_t begin = offset.size();
-    offset.resize(begin + batch.Size());
-    for (size_t i = 0; i < batch.Size(); ++i) {
-      offset[i + begin] = top + batch.offset[i + 1];
-    }
-  }
+  void Push(const SparsePage &batch);
+  /*!
+   * \brief Push a SparsePage stored in CSC format
+   * \param batch The row batch to be pushed
+   */
+  void PushCSC(const SparsePage& batch);
   /*!
    * \brief Push one instance into page
    *  \param inst an instance row
    */
   inline void Push(const Inst &inst) {
-    offset.push_back(offset.back() + inst.length);
-    size_t begin = data.size();
-    data.resize(begin + inst.length);
-    if (inst.length != 0) {
-      std::memcpy(dmlc::BeginPtr(data) + begin, inst.data,
-                  sizeof(Entry) * inst.length);
+    auto& data_vec = data.HostVector();
+    auto& offset_vec = offset.HostVector();
+    offset_vec.push_back(offset_vec.back() + inst.size());
+    size_t begin = data_vec.size();
+    data_vec.resize(begin + inst.size());
+    if (inst.size() != 0) {
+      std::memcpy(dmlc::BeginPtr(data_vec) + begin, inst.data(),
+                  sizeof(Entry) * inst.size());
     }
   }
 
-  size_t Size() { return offset.size() - 1; }
+  size_t Size() { return offset.Size() - 1; }
 };
 
+class BatchIteratorImpl {
+ public:
+  virtual ~BatchIteratorImpl() {}
+  virtual BatchIteratorImpl* Clone() = 0;
+  virtual const SparsePage& operator*() const = 0;
+  virtual void operator++() = 0;
+  virtual bool AtEnd() const = 0;
+};
 
+class BatchIterator {
+ public:
+  using iterator_category = std::forward_iterator_tag;
+  explicit BatchIterator(BatchIteratorImpl* impl) { impl_.reset(impl); }
+
+  BatchIterator(const BatchIterator& other) {
+    if (other.impl_) {
+      impl_.reset(other.impl_->Clone());
+    } else {
+      impl_.reset();
+    }
+  }
+
+  void operator++() {
+    CHECK(impl_ != nullptr);
+    ++(*impl_);
+  }
+
+  const SparsePage& operator*() const {
+    CHECK(impl_ != nullptr);
+    return *(*impl_);
+  }
+
+  bool operator!=(const BatchIterator& rhs) const {
+    CHECK(impl_ != nullptr);
+    return !impl_->AtEnd();
+  }
+
+  bool AtEnd() const {
+    CHECK(impl_ != nullptr);
+    return impl_->AtEnd();
+  }
+
+ private:
+  std::unique_ptr<BatchIteratorImpl> impl_;
+};
+
+class BatchSet {
+ public:
+  explicit BatchSet(BatchIterator begin_iter) : begin_iter_(begin_iter) {}
+  BatchIterator begin() { return begin_iter_; }
+  BatchIterator end() { return BatchIterator(nullptr); }
+
+ private:
+  BatchIterator begin_iter_;
+};
 
 /*!
  * \brief This is data structure that user can pass to DMatrix::Create
@@ -317,32 +405,17 @@ class DMatrix {
   virtual MetaInfo& Info() = 0;
   /*! \brief meta information of the dataset */
   virtual const MetaInfo& Info() const = 0;
-  /*!
-   * \brief get the row iterator, reset to beginning position
-   * \note Only either RowIterator or  column Iterator can be active.
+  /**
+   * \brief Gets row batches. Use range based for loop over BatchSet to access individual batches.
    */
-  virtual dmlc::DataIter<SparsePage>* RowIterator() = 0;
-  /*!\brief get column iterator, reset to the beginning position */
-  virtual dmlc::DataIter<SparsePage>* ColIterator() = 0;
-  /*!
-   * \brief check if column access is supported, if not, initialize column access.
-   * \param max_row_perbatch auxiliary information, maximum row used in each column batch.
-   *         this is a hint information that can be ignored by the implementation.
-   * \param sorted If column features should be in sorted order           
-   * \return Number of column blocks in the column access.
-   */
-  virtual void InitColAccess(size_t max_row_perbatch, bool sorted) = 0;
+  virtual BatchSet GetRowBatches() = 0;
+  virtual BatchSet GetSortedColumnBatches() = 0;
+  virtual BatchSet GetColumnBatches() = 0;
   // the following are column meta data, should be able to answer them fast.
-  /*! \return whether column access is enabled */
-  virtual bool HaveColAccess(bool sorted) const = 0;
   /*! \return Whether the data columns single column block. */
   virtual bool SingleColBlock() const = 0;
-  /*! \brief get number of non-missing entries in column */
-  virtual size_t GetColSize(size_t cidx) const = 0;
   /*! \brief get column density */
-  virtual float GetColDensity(size_t cidx) const = 0;
-  /*! \return reference of buffered rowset, in column access */
-  virtual const RowSet& BufferedRowset() const = 0;
+  virtual float GetColDensity(size_t cidx) = 0;
   /*! \brief virtual destructor */
   virtual ~DMatrix() = default;
   /*!
@@ -389,12 +462,6 @@ class DMatrix {
    */
   static DMatrix* Create(dmlc::Parser<uint32_t>* parser,
                          const std::string& cache_prefix = "");
-
- private:
-  // allow learner class to access this field.
-  friend class LearnerImpl;
-  /*! \brief public field to back ref cached matrix. */
-  LearnerImpl* cache_learner_ptr_{nullptr};
 };
 
 // implementation of inline functions

include/xgboost/learner.h

@@ -10,6 +10,8 @@
 
 #include <rabit/rabit.h>
 #include <utility>
+#include <map>
+#include <memory>
 #include <string>
 #include <vector>
 #include "./base.h"
@@ -178,6 +180,12 @@ class Learner : public rabit::Serializable {
    */
   static Learner* Create(const std::vector<std::shared_ptr<DMatrix> >& cache_data);
 
+  /*!
+   * \brief Get configuration arguments currently stored by the learner
+   * \return Key-value pairs representing configuration arguments
+   */
+  virtual const std::map<std::string, std::string>& GetConfigurationArguments() const = 0;
+
  protected:
   /*! \brief internal base score of the model */
   bst_float base_score_;

include/xgboost/logging.h

@@ -9,8 +9,13 @@
 #define XGBOOST_LOGGING_H_
 
 #include <dmlc/logging.h>
+#include <dmlc/parameter.h>
 #include <dmlc/thread_local.h>
 #include <sstream>
+#include <map>
+#include <string>
+#include <utility>
+#include <vector>
 #include "./base.h"
 
 namespace xgboost {
@@ -20,7 +25,7 @@ class BaseLogger {
   BaseLogger() {
 #if XGBOOST_LOG_WITH_TIME
     log_stream_ << "[" << dmlc::DateLogger().HumanDate() << "] ";
-#endif
+#endif  // XGBOOST_LOG_WITH_TIME
   }
   std::ostream& stream() { return log_stream_; }  // NOLINT
 
@@ -28,8 +33,55 @@ class BaseLogger {
   std::ostringstream log_stream_;
 };
 
+// Parsing both silent and debug_verbose is to provide backward compatibility.
+struct ConsoleLoggerParam : public dmlc::Parameter<ConsoleLoggerParam> {
+  bool silent;  // deprecated.
+  int verbosity;
+
+  DMLC_DECLARE_PARAMETER(ConsoleLoggerParam) {
+    DMLC_DECLARE_FIELD(silent)
+        .set_default(false)
+        .describe("Do not print information during training.");
+    DMLC_DECLARE_FIELD(verbosity)
+        .set_range(0, 3)
+        .set_default(1)  // shows only warning
+        .describe("Flag to print out detailed breakdown of runtime.");
+    DMLC_DECLARE_ALIAS(verbosity, debug_verbose);
+  }
+};
+
 class ConsoleLogger : public BaseLogger {
  public:
+  enum class LogVerbosity {
+    kSilent = 0,
+    kWarning = 1,
+    kInfo = 2,   // information may interests users.
+    kDebug = 3,  // information only interesting to developers.
+    kIgnore = 4  // ignore global setting
+  };
+  using LV = LogVerbosity;
+
+ private:
+  static LogVerbosity global_verbosity_;
+  static ConsoleLoggerParam param_;
+
+  LogVerbosity cur_verbosity_;
+  static void Configure(const std::map<std::string, std::string>& args);
+
+ public:
+  template <typename ArgIter>
+  static void Configure(ArgIter begin, ArgIter end) {
+    std::map<std::string, std::string> args(begin, end);
+    Configure(args);
+  }
+
+  static LogVerbosity GlobalVerbosity();
+  static LogVerbosity DefaultVerbosity();
+  static bool ShouldLog(LogVerbosity verbosity);
+
+  ConsoleLogger() = delete;
+  explicit ConsoleLogger(LogVerbosity cur_verb);
+  ConsoleLogger(const std::string& file, int line, LogVerbosity cur_verb);
   ~ConsoleLogger();
 };
 
@@ -38,6 +90,8 @@ class TrackerLogger : public BaseLogger {
   ~TrackerLogger();
 };
 
+// custom logging callback; disabled for R wrapper
+#if !defined(XGBOOST_STRICT_R_MODE) || XGBOOST_STRICT_R_MODE == 0
 class LogCallbackRegistry {
  public:
   using Callback = void (*)(const char*);
@@ -52,16 +106,57 @@ class LogCallbackRegistry {
  private:
   Callback log_callback_;
 };
+#else
+class LogCallbackRegistry {
+ public:
+  using Callback = void (*)(const char*);
+  LogCallbackRegistry() {}
+  inline void Register(Callback log_callback) {}
+  inline Callback Get() const {
+    return nullptr;
+  }
+};
+#endif  // !defined(XGBOOST_STRICT_R_MODE) || XGBOOST_STRICT_R_MODE == 0
 
 using LogCallbackRegistryStore = dmlc::ThreadLocalStore<LogCallbackRegistry>;
 
+// Redefines LOG_WARNING for controling verbosity
+#if defined(LOG_WARNING)
+#undef  LOG_WARNING
+#endif  // defined(LOG_WARNING)
+#define LOG_WARNING                                                            \
+  if (::xgboost::ConsoleLogger::ShouldLog(                                     \
+          ::xgboost::ConsoleLogger::LV::kWarning))                             \
+  ::xgboost::ConsoleLogger(__FILE__, __LINE__,                                 \
+                           ::xgboost::ConsoleLogger::LogVerbosity::kWarning)
+
+// Redefines LOG_INFO for controling verbosity
+#if defined(LOG_INFO)
+#undef  LOG_INFO
+#endif  // defined(LOG_INFO)
+#define LOG_INFO                                                               \
+  if (::xgboost::ConsoleLogger::ShouldLog(                                     \
+          ::xgboost::ConsoleLogger::LV::kInfo))                                \
+  ::xgboost::ConsoleLogger(__FILE__, __LINE__,                                 \
+                           ::xgboost::ConsoleLogger::LogVerbosity::kInfo)
+
+#if defined(LOG_DEBUG)
+#undef LOG_DEBUG
+#endif  // defined(LOG_DEBUG)
+#define LOG_DEBUG                                                              \
+  if (::xgboost::ConsoleLogger::ShouldLog(                                     \
+          ::xgboost::ConsoleLogger::LV::kDebug))                               \
+  ::xgboost::ConsoleLogger(__FILE__, __LINE__,                                 \
+                           ::xgboost::ConsoleLogger::LogVerbosity::kDebug)
+
 // redefines the logging macro if not existed
 #ifndef LOG
 #define LOG(severity) LOG_##severity.stream()
-#endif
+#endif  // LOG
 
 // Enable LOG(CONSOLE) for print messages to console.
-#define LOG_CONSOLE ::xgboost::ConsoleLogger()
+#define LOG_CONSOLE ::xgboost::ConsoleLogger(           \
+    ::xgboost::ConsoleLogger::LogVerbosity::kIgnore)
 // Enable LOG(TRACKER) for print messages to tracker
 #define LOG_TRACKER ::xgboost::TrackerLogger()
 }  // namespace xgboost.

include/xgboost/metric.h

@@ -11,8 +11,11 @@
 #include <vector>
 #include <string>
 #include <functional>
+#include <utility>
+
 #include "./data.h"
 #include "./base.h"
+#include "../../src/common/host_device_vector.h"
 
 namespace xgboost {
 /*!
@@ -21,6 +24,23 @@ namespace xgboost {
  */
 class Metric {
  public:
+  /*!
+   * \brief Configure the Metric with the specified parameters.
+   * \param args arguments to the objective function.
+   */
+  virtual void Configure(
+      const std::vector<std::pair<std::string, std::string> >& args) {}
+  /*!
+   * \brief set configuration from pair iterators.
+   * \param begin The beginning iterator.
+   * \param end The end iterator.
+   * \tparam PairIter iterator<std::pair<std::string, std::string> >
+   */
+  template<typename PairIter>
+  inline void Configure(PairIter begin, PairIter end) {
+    std::vector<std::pair<std::string, std::string> > vec(begin, end);
+    this->Configure(vec);
+  }
   /*!
    * \brief evaluate a specific metric
    * \param preds prediction
@@ -29,9 +49,9 @@ class Metric {
    *        the average statistics across all the node,
    *        this is only supported by some metrics
    */
-  virtual bst_float Eval(const std::vector<bst_float>& preds,
+  virtual bst_float Eval(const HostDeviceVector<bst_float>& preds,
                          const MetaInfo& info,
-                         bool distributed) const = 0;
+                         bool distributed) = 0;
   /*! \return name of metric */
   virtual const char* Name() const = 0;
   /*! \brief virtual destructor */

include/xgboost/objective.h

@@ -44,7 +44,7 @@ class ObjFunction {
    * \param iteration current iteration number.
    * \param out_gpair output of get gradient, saves gradient and second order gradient in
    */
-  virtual void GetGradient(HostDeviceVector<bst_float>* preds,
+  virtual void GetGradient(const HostDeviceVector<bst_float>& preds,
                            const MetaInfo& info,
                            int iteration,
                            HostDeviceVector<GradientPair>* out_gpair) = 0;