Release version 0.72 (#3337)

* Fix print.xgb.Booster

valid_handle should be TRUE when x$handle is NOT null

* Update xgb.Booster.R

Modify is.null.handle to return TRUE for NULL handle

.clang-tidy

@@ -0,0 +1,22 @@
+Checks: 'modernize-*,-modernize-make-*,-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  }
+  - { key: readability-identifier-naming.TypeAliasCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.TypedefCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.TypeTemplateParameterCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.LocalVariableCase,          value: lower_case  }
+  - { key: readability-identifier-naming.MemberCase,          value: lower_case  }
+  - { key: readability-identifier-naming.PrivateMemberSuffix,           value: '_'  }
+  - { key: readability-identifier-naming.ProtectedMemberSuffix,           value: '_'  }
+  - { key: readability-identifier-naming.EnumCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.EnumConstant,          value: CamelCase  }
+  - { key: readability-identifier-naming.EnumConstantPrefix,          value: k  }
+  - { key: readability-identifier-naming.GlobalConstantCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.GlobalConstantPrefix,          value: k  }
+  - { key: readability-identifier-naming.StaticConstantCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.StaticConstantPrefix,          value: k  }
+  - { key: readability-identifier-naming.ConstexprVariableCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.ConstexprVariablePrefix,          value: k  }
+  - { key: readability-identifier-naming.FunctionCase,          value: CamelCase  }
+  - { key: readability-identifier-naming.NamespaceCase,       value: lower_case }

.gitignore

@@ -15,7 +15,6 @@
 *.Rcheck
 *.rds
 *.tar.gz
-#*txt*
 *conf
 *buffer
 *model
@@ -47,13 +46,12 @@ Debug
 *.cpage.col
 *.cpage
 *.Rproj
-./xgboost
 ./xgboost.mpi
 ./xgboost.mock
 #.Rbuildignore
 R-package.Rproj
 *.cache*
-#java
+# java
 java/xgboost4j/target
 java/xgboost4j/tmp
 java/xgboost4j-demo/target
@@ -68,10 +66,9 @@ nb-configuration*
 .settings/
 build
 config.mk
-xgboost
+/xgboost
 *.data
 build_plugin
-dmlc-core
 .idea
 recommonmark/
 tags

.travis.yml

@@ -44,10 +44,12 @@ matrix:
 addons:
   apt:
     sources:
+      - llvm-toolchain-trusty-5.0
       - ubuntu-toolchain-r-test
       - george-edison55-precise-backports
     packages:
-      - cmake
+      - clang
+      - clang-tidy-5.0
       - cmake-data
       - doxygen
       - wget

CMakeLists.txt

@@ -14,8 +14,8 @@ 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)
-set(GPU_COMPUTE_VER 35;50;52;60;61 CACHE STRING
+set(GPU_COMPUTE_VER "" CACHE STRING
-  "Space separated list of compute versions to be built against")
+  "Space separated list of compute versions to be built against, e.g. '35 61'")
 
 # Deprecation warning
 if(PLUGIN_UPDATER_GPU)
@@ -106,7 +106,7 @@ endif()
 
 # dmlc-core
 add_subdirectory(dmlc-core)
-set(LINK_LIBRARIES dmlccore rabit)
+set(LINK_LIBRARIES dmlc rabit)
 
 
 if(USE_CUDA)
@@ -122,16 +122,13 @@ if(USE_CUDA)
     add_definitions(-DXGBOOST_USE_NCCL)
   endif()
 
-  if((CUDA_VERSION_MAJOR EQUAL 9) OR (CUDA_VERSION_MAJOR GREATER 9))
-    message("CUDA 9.0 detected, adding Volta compute capability (7.0).")
-    set(GPU_COMPUTE_VER "${GPU_COMPUTE_VER};70")
-  endif()
-  
   set(GENCODE_FLAGS "")
   format_gencode_flags("${GPU_COMPUTE_VER}" GENCODE_FLAGS)
+  message("cuda architecture flags: ${GENCODE_FLAGS}")
+
   set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};--expt-extended-lambda;--expt-relaxed-constexpr;${GENCODE_FLAGS};-lineinfo;")
   if(NOT MSVC)
-    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};-Xcompiler -fPIC; -std=c++11")
+    set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};-Xcompiler -fPIC; -Xcompiler -Werror; -std=c++11")
   endif()
 
   if(USE_NCCL)

CONTRIBUTORS.md

@@ -7,8 +7,8 @@ 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.
-* [Tong He](https://github.com/hetong007), Simon Fraser University
+* [Tong He](https://github.com/hetong007), Amazon AI
-  - Tong is a master student working on data mining, 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)
@@ -54,7 +54,8 @@ List of Contributors
 * [Masaaki Horikoshi](https://github.com/sinhrks)
   - Masaaki is the initial creator of xgboost python plotting module.
 * [Hongliang Liu](https://github.com/phunterlau)
-  - Hongliang is the maintainer of xgboost python PyPI package for pip installation.
+* [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.
 * [daiyl0320](https://github.com/daiyl0320)
   - daiyl0320 contributed patch to xgboost distributed version more robust, and scales stably on TB scale datasets.
 * [Huayi Zhang](https://github.com/irachex)

Jenkinsfile

@@ -7,9 +7,9 @@
 dockerRun = 'tests/ci_build/ci_build.sh'
 
 def buildMatrix = [
-    [ "enabled": true,  "os" : "linux", "withGpu": true,  "withOmp": true, "pythonVersion": "2.7" ],
+    [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "9.1" ],
-    [ "enabled": true,  "os" : "linux", "withGpu": false, "withOmp": true, "pythonVersion": "2.7" ],
+    [ "enabled": true,  "os" : "linux", "withGpu": true, "withNccl": true,  "withOmp": true, "pythonVersion": "2.7", "cudaVersion": "8.0" ],
-    [ "enabled": false, "os" : "osx",   "withGpu": false, "withOmp": false, "pythonVersion": "2.7" ],
+    [ "enabled": false,  "os" : "linux", "withGpu": false, "withNccl": false, "withOmp": true, "pythonVersion": "2.7", "cudaVersion": ""  ],
 ]
 
 pipeline {
@@ -69,8 +69,7 @@ def buildFactory(buildName, conf) {
     def os = conf["os"]
     def nodeReq = conf["withGpu"] ? "${os} && gpu" : "${os}"
     def dockerTarget = conf["withGpu"] ? "gpu" : "cpu"
-    [ ("cmake_${buildName}") : { buildPlatformCmake("cmake_${buildName}", conf, nodeReq, dockerTarget) },
+    [ ("${buildName}") : { buildPlatformCmake("${buildName}", conf, nodeReq, dockerTarget) }
-      ("make_${buildName}") : { buildPlatformMake("make_${buildName}", conf, nodeReq, dockerTarget) }
     ]
 }
 
@@ -81,6 +80,10 @@ def buildPlatformCmake(buildName, conf, nodeReq, dockerTarget) {
     def opts = 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
     node(nodeReq) {
         unstash name: 'srcs'
@@ -92,58 +95,33 @@ def buildPlatformCmake(buildName, conf, nodeReq, dockerTarget) {
         """.stripMargin('|')
         // Invoke command inside docker
         sh """
-        ${dockerRun} ${dockerTarget} tests/ci_build/build_via_cmake.sh ${opts}
+        ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/build_via_cmake.sh ${opts}
-        ${dockerRun} ${dockerTarget} tests/ci_build/test_${dockerTarget}.sh
+        ${dockerRun} ${dockerTarget} ${dockerArgs} tests/ci_build/test_${dockerTarget}.sh
-        ${dockerRun} ${dockerTarget} bash -c "cd python-package; python setup.py bdist_wheel"
+        ${dockerRun} ${dockerTarget} ${dockerArgs} bash -c "cd python-package; python setup.py bdist_wheel"
         rm -rf "${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 ${dockerArgs} bash -c " \
+            auditwheel show xgboost-*-py2-none-any.whl
+            pip install --user python-package/dist/xgboost-*-py2-none-any.whl && \
+            python -m nose tests/python"
         """
         archiveArtifacts artifacts: "${distDir}/**/*.*", allowEmptyArchive: true
     }
 }
 
-/**
- * Build platform via make
- */
-def buildPlatformMake(buildName, conf, nodeReq, dockerTarget) {
-    def opts = makeOptions(conf)
-    // Destination dir for artifacts
-    def distDir = "dist/${buildName}"
-    // Build node
-    node(nodeReq) {
-        unstash name: 'srcs'
-        echo """
-        |===== XGBoost Make build =====
-        |  dockerTarget: ${dockerTarget}
-        |  makeOpts    : ${opts}
-        |=========================
-        """.stripMargin('|')
-        // Invoke command inside docker
-        sh """
-        ${dockerRun} ${dockerTarget} tests/ci_build/build_via_make.sh ${opts}
-        """
-    }
-}
-
-def makeOptions(conf) {
-    return ([
-        conf["withGpu"] ? 'PLUGIN_UPDATER_GPU=ON' : 'PLUGIN_UPDATER_GPU=OFF',
-        conf["withOmp"] ? 'USE_OPENMP=1' : 'USE_OPENMP=0']
-        ).join(" ")
-}
-
-
 def cmakeOptions(conf) {
     return ([
-        conf["withGpu"] ? '-DPLUGIN_UPDATER_GPU:BOOL=ON' : '',
+        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'] ? "_gpu" : "_cpu"
+    def gpuLabel = conf['withGpu'] ? "_cuda" + conf['cudaVersion'] : "_cpu"
     def ompLabel = conf['withOmp'] ? "_omp" : ""
     def pyLabel = "_py${conf['pythonVersion']}"
     return "${conf['os']}${gpuLabel}${ompLabel}${pyLabel}"

Makefile

@@ -198,7 +198,11 @@ endif
 clean:
 	$(RM) -rf build build_plugin lib bin *~ */*~ */*/*~ */*/*/*~ */*.o */*/*.o */*/*/*.o #xgboost
 	$(RM) -rf build_tests *.gcov tests/cpp/xgboost_test
-	cd R-package/src; $(RM) -rf rabit src include dmlc-core amalgamation *.so *.dll; cd $(ROOTDIR)
+	if [ -d "R-package/src" ]; then \
+		cd R-package/src; \
+		$(RM) -rf rabit src include dmlc-core amalgamation *.so *.dll; \
+		cd $(ROOTDIR); \
+	fi
 
 clean_all: clean
 	cd $(DMLC_CORE); "$(MAKE)" clean; cd $(ROOTDIR)
@@ -212,16 +216,28 @@ pypack: ${XGBOOST_DYLIB}
 	cp ${XGBOOST_DYLIB} python-package/xgboost
 	cd python-package; tar cf xgboost.tar xgboost; cd ..
 
-# create pip installation pack for PyPI
+# create pip source dist (sdist) pack for PyPI
 pippack: clean_all
 	rm -rf xgboost-python
+# remove symlinked directories in python-package/xgboost
+	rm -rf python-package/xgboost/lib
+	rm -rf python-package/xgboost/dmlc-core
+	rm -rf python-package/xgboost/include
+	rm -rf python-package/xgboost/make
+	rm -rf python-package/xgboost/rabit
+	rm -rf python-package/xgboost/src
 	cp -r python-package xgboost-python
 	cp -r Makefile xgboost-python/xgboost/
 	cp -r make xgboost-python/xgboost/
 	cp -r src xgboost-python/xgboost/
+	cp -r tests xgboost-python/xgboost/
 	cp -r include xgboost-python/xgboost/
 	cp -r dmlc-core xgboost-python/xgboost/
 	cp -r rabit xgboost-python/xgboost/
+# Use setup_pip.py instead of setup.py
+	mv xgboost-python/setup_pip.py xgboost-python/setup.py
+# Build sdist tarball
+	cd xgboost-python; python setup.py sdist; mv dist/*.tar.gz ..; cd ..
 
 # Script to make a clean installable R package.
 Rpack: clean_all
@@ -245,13 +261,15 @@ Rpack: clean_all
 	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
+	bash R-package/remove_warning_suppression_pragma.sh
+	rm xgboost/remove_warning_suppression_pragma.sh
 
 Rbuild: Rpack
 	R CMD build --no-build-vignettes xgboost
 	rm -rf xgboost
 
 Rcheck: Rbuild
-	R CMD check  xgboost*.tar.gz
+	R CMD check xgboost*.tar.gz
 
 -include build/*.d
 -include build/*/*.d

NEWS.md

@@ -3,6 +3,65 @@ XGBoost Change Log
 
 This file records the changes in xgboost library in reverse chronological order.
 
+## v0.72 (2018.06.01)
+* Starting with this release, we plan to make a new release every two months. See #3252 for more details.
+* Fix a pathological behavior (near-zero second-order gradients) in multiclass objective (#3304)
+* Tree dumps now use high precision in storing floating-point values (#3298)
+* Submodules `rabit` and `dmlc-core` have been brought up to date, bringing bug fixes (#3330, #3221).
+* GPU support
+  - Continuous integration tests for GPU code (#3294, #3309)
+  - GPU accelerated coordinate descent algorithm (#3178)
+  - Abstract 1D vector class now works with multiple GPUs (#3287)
+  - Generate PTX code for most recent architecture (#3316)
+  - Fix a memory bug on NVIDIA K80 cards (#3293)
+  - Address performance instability for single-GPU, multi-core machines (#3324)
+* Python package
+  - FreeBSD support (#3247)
+  - Validation of feature names in `Booster.predict()` is now optional (#3323)
+* Updated Sklearn API
+  - Validation sets now support instance weights (#2354)
+  - `XGBClassifier.predict_proba()` should not support `output_margin` option. (#3343) See BREAKING CHANGES below.
+* R package:
+  - Better handling of NULL in `print.xgb.Booster()` (#3338)
+  - Comply with CRAN policy by removing compiler warning suppression (#3329)
+  - Updated CRAN submission
+* JVM packages
+  - JVM packages will now use the same versioning scheme as other packages (#3253)
+  - Update Spark to 2.3 (#3254)
+  - Add scripts to cross-build and deploy artifacts (#3276, #3307)
+  - Fix a compilation error for Scala 2.10 (#3332)
+* BREAKING CHANGES
+  - `XGBClassifier.predict_proba()` no longer accepts paramter `output_margin`. The paramater makes no sense for `predict_proba()` because the method is to predict class probabilities, not raw margin scores.
+
+## v0.71 (2018.04.11)
+* This is a minor release, mainly motivated by issues concerning `pip install`, e.g. #2426, #3189, #3118, and #3194.
+  With this release, users of Linux and MacOS will be able to run `pip install` for the most part.
+* Refactored linear booster class (`gblinear`), so as to support multiple coordinate descent updaters (#3103, #3134). See BREAKING CHANGES below.
+* Fix slow training for multiclass classification with high number of classes (#3109)
+* Fix a corner case in approximate quantile sketch (#3167). Applicable for 'hist' and 'gpu_hist' algorithms
+* Fix memory leak in DMatrix (#3182)
+* New functionality
+  - Better linear booster class (#3103, #3134)
+  - Pairwise SHAP interaction effects (#3043)
+  - Cox loss (#3043)
+  - AUC-PR metric for ranking task (#3172)
+  - Monotonic constraints for 'hist' algorithm (#3085)
+* GPU support
+  - Create an abtract 1D vector class that moves data seamlessly between the main and GPU memory (#2935, #3116, #3068). This eliminates unnecessary PCIe data transfer during training time.
+  - Fix minor bugs (#3051, #3217)
+  - Fix compatibility error for CUDA 9.1 (#3218)
+* Python package:
+  - Correctly handle parameter `verbose_eval=0` (#3115)
+* R package:
+  - Eliminate segmentation fault on 32-bit Windows platform (#2994)
+* JVM packages
+  - Fix a memory bug involving double-freeing Booster objects (#3005, #3011)
+  - Handle empty partition in predict (#3014)
+  - Update docs and unify terminology (#3024)
+  - Delete cache files after job finishes (#3022)
+  - Compatibility fixes for latest Spark versions (#3062, #3093)
+* BREAKING CHANGES: Updated linear modelling algorithms. In particular L1/L2 regularisation penalties are now normalised to number of training examples. This makes the implementation consistent with sklearn/glmnet. L2 regularisation has also been removed from the intercept. To produce linear models with the old regularisation behaviour, the alpha/lambda regularisation parameters can be manually scaled by dividing them by the number of training examples.
+
 ## v0.7 (2017.12.30)
 * **This version represents a major change from the last release (v0.6), which was released one year and half ago.**
 * Updated Sklearn API

R-package/DESCRIPTION

@@ -1,12 +1,34 @@
 Package: xgboost
 Type: Package
 Title: Extreme Gradient Boosting
-Version: 0.6.4.8
+Version: 0.71.1
-Date: 2017-12-05
+Date: 2018-05-11
-Author: Tianqi Chen <tianqi.tchen@gmail.com>, Tong He <hetong007@gmail.com>,
+Authors@R: c(
-    Michael Benesty <michael@benesty.fr>, Vadim Khotilovich <khotilovich@gmail.com>,
+  person("Tianqi", "Chen", role = c("aut"),
-    Yuan Tang <terrytangyuan@gmail.com>
+         email = "tianqi.tchen@gmail.com"),
-Maintainer: Tong He <hetong007@gmail.com>
+  person("Tong", "He", role = c("aut", "cre"),
+         email = "hetong007@gmail.com"),
+  person("Michael", "Benesty", role = c("aut"),
+         email = "michael@benesty.fr"),
+  person("Vadim", "Khotilovich", role = c("aut"),
+         email = "khotilovich@gmail.com"),
+  person("Yuan", "Tang", role = c("aut"),
+         email = "terrytangyuan@gmail.com",
+         comment = c(ORCID = "0000-0001-5243-233X")),
+  person("Hyunsu", "Cho", role = c("aut"),
+         email = "chohyu01@cs.washington.edu"),
+  person("Kailong", "Chen", role = c("aut")),
+  person("Rory", "Mitchell", role = c("aut")),
+  person("Ignacio", "Cano", role = c("aut")),
+  person("Tianyi", "Zhou", role = c("aut")),
+  person("Mu", "Li", role = c("aut")),
+  person("Junyuan", "Xie", role = c("aut")),
+  person("Min", "Lin", role = c("aut")),
+  person("Yifeng", "Geng", role = c("aut")),
+  person("Yutian", "Li", role = c("aut")),
+  person("XGBoost contributors", role = c("cph"),
+         comment = "base XGBoost implementation")
+  )
 Description: Extreme Gradient Boosting, which is an efficient implementation
     of the gradient boosting framework from Chen & Guestrin (2016) <doi:10.1145/2939672.2939785>.
     This package is its R interface. The package includes efficient linear 
@@ -19,6 +41,7 @@ Description: Extreme Gradient Boosting, which is an efficient implementation
 License: Apache License (== 2.0) | file LICENSE
 URL: https://github.com/dmlc/xgboost
 BugReports: https://github.com/dmlc/xgboost/issues
+NeedsCompilation: yes
 VignetteBuilder: knitr
 Suggests:
     knitr,
@@ -38,3 +61,4 @@ Imports:
     magrittr (>= 1.5),
     stringi (>= 0.5.2)
 RoxygenNote: 6.0.1
+SystemRequirements: GNU make, C++11

R-package/NAMESPACE

@@ -18,6 +18,7 @@ export("xgb.parameters<-")
 export(cb.cv.predict)
 export(cb.early.stop)
 export(cb.evaluation.log)
+export(cb.gblinear.history)
 export(cb.print.evaluation)
 export(cb.reset.parameters)
 export(cb.save.model)
@@ -32,6 +33,7 @@ export(xgb.attributes)
 export(xgb.create.features)
 export(xgb.cv)
 export(xgb.dump)
+export(xgb.gblinear.history)
 export(xgb.ggplot.deepness)
 export(xgb.ggplot.importance)
 export(xgb.importance)
@@ -49,10 +51,11 @@ export(xgboost)
 import(methods)
 importClassesFrom(Matrix,dgCMatrix)
 importClassesFrom(Matrix,dgeMatrix)
-importFrom(Matrix,cBind)
 importFrom(Matrix,colSums)
 importFrom(Matrix,sparse.model.matrix)
+importFrom(Matrix,sparseMatrix)
 importFrom(Matrix,sparseVector)
+importFrom(Matrix,t)
 importFrom(data.table,":=")
 importFrom(data.table,as.data.table)
 importFrom(data.table,data.table)

R-package/R/callbacks.R

@@ -524,6 +524,223 @@ cb.cv.predict <- function(save_models = FALSE) {
 }
 
 
+#' Callback closure for collecting the model coefficients history of a gblinear booster
+#' during its training.
+#'
+#' @param sparse when set to FALSE/TURE, a dense/sparse matrix is used to store the result.
+#'       Sparse format is useful when one expects only a subset of coefficients to be non-zero,
+#'       when using the "thrifty" feature selector with fairly small number of top features
+#'       selected per iteration.
+#'
+#' @details
+#' To keep things fast and simple, gblinear booster does not internally store the history of linear
+#' model coefficients at each boosting iteration. This callback provides a workaround for storing
+#' the coefficients' path, by extracting them after each training iteration.
+#'
+#' Callback function expects the following values to be set in its calling frame:
+#' \code{bst} (or \code{bst_folds}).
+#'
+#' @return
+#' Results are stored in the \code{coefs} element of the closure.
+#' The \code{\link{xgb.gblinear.history}} convenience function provides an easy way to access it.
+#' With \code{xgb.train}, it is either a dense of a sparse matrix.
+#' While with \code{xgb.cv}, it is a list (an element per each fold) of such matrices.
+#'
+#' @seealso
+#' \code{\link{callbacks}}, \code{\link{xgb.gblinear.history}}.
+#'
+#' @examples
+#' #### Binary classification:
+#' #
+#' # In the iris dataset, it is hard to linearly separate Versicolor class from the rest
+#' # without considering the 2nd order interactions:
+#' require(magrittr)
+#' x <- model.matrix(Species ~ .^2, iris)[,-1]
+#' colnames(x)
+#' dtrain <- xgb.DMatrix(scale(x), label = 1*(iris$Species == "versicolor"))
+#' param <- list(booster = "gblinear", objective = "reg:logistic", eval_metric = "auc",
+#'               lambda = 0.0003, alpha = 0.0003, nthread = 2)
+#' # For 'shotgun', which is a default linear updater, using high eta values may result in
+#' # unstable behaviour in some datasets. With this simple dataset, however, the high learning
+#' # rate does not break the convergence, but allows us to illustrate the typical pattern of
+#' # "stochastic explosion" behaviour of this lock-free algorithm at early boosting iterations.
+#' bst <- xgb.train(param, dtrain, list(tr=dtrain), nrounds = 200, eta = 1.,
+#'                  callbacks = list(cb.gblinear.history()))
+#' # Extract the coefficients' path and plot them vs boosting iteration number:
+#' coef_path <- xgb.gblinear.history(bst)
+#' matplot(coef_path, type = 'l')
+#' 
+#' # With the deterministic coordinate descent updater, it is safer to use higher learning rates.
+#' # Will try the classical componentwise boosting which selects a single best feature per round:
+#' bst <- xgb.train(param, dtrain, list(tr=dtrain), nrounds = 200, eta = 0.8,
+#'                  updater = 'coord_descent', feature_selector = 'thrifty', top_k = 1,
+#'                  callbacks = list(cb.gblinear.history()))
+#' xgb.gblinear.history(bst) %>% matplot(type = 'l')
+#' #  Componentwise boosting is known to have similar effect to Lasso regularization.
+#' # Try experimenting with various values of top_k, eta, nrounds,
+#' # as well as different feature_selectors.
+#'
+#' # For xgb.cv:
+#' bst <- xgb.cv(param, dtrain, nfold = 5, nrounds = 100, eta = 0.8,
+#'              callbacks = list(cb.gblinear.history()))
+#' # coefficients in the CV fold #3
+#' xgb.gblinear.history(bst)[[3]] %>% matplot(type = 'l')
+#'
+#' 
+#' #### Multiclass classification:
+#' #
+#' dtrain <- xgb.DMatrix(scale(x), label = as.numeric(iris$Species) - 1)
+#' param <- list(booster = "gblinear", objective = "multi:softprob", num_class = 3,
+#'               lambda = 0.0003, alpha = 0.0003, nthread = 2)
+#' # For the default linear updater 'shotgun' it sometimes is helpful
+#' # to use smaller eta to reduce instability
+#' bst <- xgb.train(param, dtrain, list(tr=dtrain), nrounds = 70, eta = 0.5,
+#'                  callbacks = list(cb.gblinear.history()))
+#' # Will plot the coefficient paths separately for each class:
+#' xgb.gblinear.history(bst, class_index = 0) %>% matplot(type = 'l')
+#' xgb.gblinear.history(bst, class_index = 1) %>% matplot(type = 'l')
+#' xgb.gblinear.history(bst, class_index = 2) %>% matplot(type = 'l')
+#'
+#' # CV:
+#' bst <- xgb.cv(param, dtrain, nfold = 5, nrounds = 70, eta = 0.5,
+#'               callbacks = list(cb.gblinear.history(FALSE)))
+#' # 1st forld of 1st class
+#' xgb.gblinear.history(bst, class_index = 0)[[1]] %>% matplot(type = 'l')
+#'
+#' @export
+cb.gblinear.history <- function(sparse=FALSE) {
+  coefs <- NULL
+
+  init <- function(env) {
+    if (!is.null(env$bst)) { # xgb.train:
+      coef_path <- list()
+    } else if (!is.null(env$bst_folds)) { # xgb.cv:
+      coef_path <- rep(list(), length(env$bst_folds))
+    } else stop("Parent frame has neither 'bst' nor 'bst_folds'")
+  }
+
+  # convert from list to (sparse) matrix
+  list2mat <- function(coef_list) {
+    if (sparse) {
+      coef_mat <- sparseMatrix(x = unlist(lapply(coef_list, slot, "x")),
+                               i = unlist(lapply(coef_list, slot, "i")),
+                               p = c(0, cumsum(sapply(coef_list, function(x) length(x@x)))),
+                               dims = c(length(coef_list[[1]]), length(coef_list)))
+      return(t(coef_mat))
+    } else {
+      return(do.call(rbind, coef_list))
+    }
+  }
+
+  finalizer <- function(env) {
+    if (length(coefs) == 0)
+      return()
+    if (!is.null(env$bst)) { # # xgb.train:
+      coefs <<- list2mat(coefs)
+    } else { # xgb.cv:
+      # first lapply transposes the list
+      coefs <<- lapply(seq_along(coefs[[1]]), function(i) lapply(coefs, "[[", i)) %>%
+                lapply(function(x) list2mat(x))
+    }
+  }
+
+  extract.coef <- function(env) {
+    if (!is.null(env$bst)) { # # xgb.train:
+      cf <- as.numeric(grep('(booster|bias|weigh)', xgb.dump(env$bst), invert = TRUE, value = TRUE))
+      if (sparse) cf <- as(cf, "sparseVector")
+    } else { # xgb.cv:
+      cf <- vector("list", length(env$bst_folds))
+      for (i in seq_along(env$bst_folds)) {
+        dmp <- xgb.dump(xgb.handleToBooster(env$bst_folds[[i]]$bst))
+        cf[[i]] <- as.numeric(grep('(booster|bias|weigh)', dmp, invert = TRUE, value = TRUE))
+        if (sparse) cf[[i]] <- as(cf[[i]], "sparseVector")
+      }
+    }
+    cf
+  }
+
+  callback <- function(env = parent.frame(), finalize = FALSE) {
+    if (is.null(coefs)) init(env)
+    if (finalize) return(finalizer(env))
+    cf <- extract.coef(env)
+    coefs <<- c(coefs, list(cf))
+  }
+
+  attr(callback, 'call') <- match.call()
+  attr(callback, 'name') <- 'cb.gblinear.history'
+  callback
+}
+
+#' Extract gblinear coefficients history.
+#'
+#' A helper function to extract the matrix of linear coefficients' history
+#' from a gblinear model created while using the \code{cb.gblinear.history()}
+#' callback.
+#'
+#' @param model either an \code{xgb.Booster} or a result of \code{xgb.cv()}, trained
+#'        using the \code{cb.gblinear.history()} callback.
+#' @param class_index zero-based class index to extract the coefficients for only that
+#'        specific class in a multinomial multiclass model. When it is NULL, all the
+#'        coeffients are returned. Has no effect in non-multiclass models.
+#'
+#' @return 
+#' For an \code{xgb.train} result, a matrix (either dense or sparse) with the columns
+#' corresponding to iteration's coefficients (in the order as \code{xgb.dump()} would
+#' return) and the rows corresponding to boosting iterations.
+#'
+#' For an \code{xgb.cv} result, a list of such matrices is returned with the elements
+#' corresponding to CV folds.
+#'
+#' @export
+xgb.gblinear.history <- function(model, class_index = NULL) {
+
+  if (!(inherits(model, "xgb.Booster") ||
+        inherits(model, "xgb.cv.synchronous")))
+    stop("model must be an object of either xgb.Booster or xgb.cv.synchronous class")
+  is_cv <- inherits(model, "xgb.cv.synchronous")
+
+  if (is.null(model[["callbacks"]]) || is.null(model$callbacks[["cb.gblinear.history"]]))
+    stop("model must be trained while using the cb.gblinear.history() callback")
+
+  if (!is_cv) {
+    # extract num_class & num_feat from the internal model
+    dmp <- xgb.dump(model)
+    if(length(dmp) < 2 || dmp[2] != "bias:")
+      stop("It does not appear to be a gblinear model")
+    dmp <- dmp[-c(1,2)]
+    n <- which(dmp == 'weight:')
+    if(length(n) != 1)
+      stop("It does not appear to be a gblinear model")
+    num_class <- n - 1
+    num_feat <- (length(dmp) - 4) / num_class
+  } else {
+    # in case of CV, the object is expected to have this info
+    if (model$params$booster != "gblinear")
+      stop("It does not appear to be a gblinear model")
+    num_class <- NVL(model$params$num_class, 1)
+    num_feat <- model$nfeatures
+    if (is.null(num_feat))
+      stop("This xgb.cv result does not have nfeatures info")
+  }
+
+  if (!is.null(class_index) &&
+      num_class > 1 &&
+      (class_index[1] < 0 || class_index[1] >= num_class))
+    stop("class_index has to be within [0,", num_class - 1, "]")
+
+  coef_path <- environment(model$callbacks$cb.gblinear.history)[["coefs"]]
+  if (!is.null(class_index) && num_class > 1) {
+    coef_path <- if (is.list(coef_path)) {
+      lapply(coef_path, 
+             function(x) x[, seq(1 + class_index, by=num_class, length.out=num_feat)])
+    } else {
+      coef_path <- coef_path[, seq(1 + class_index, by=num_class, length.out=num_feat)]
+    }
+  }
+  coef_path
+}
+
+
 #
 # Internal utility functions for callbacks ------------------------------------
 # 

R-package/R/xgb.Booster.R

@@ -37,11 +37,14 @@ xgb.handleToBooster <- function(handle, raw = NULL) {
 # Check whether xgb.Booster.handle is null
 # internal utility function
 is.null.handle <- function(handle) {
+  if (is.null(handle)) return(TRUE)
+
   if (!identical(class(handle), "xgb.Booster.handle"))
     stop("argument type must be xgb.Booster.handle")
 
-  if (is.null(handle) || .Call(XGCheckNullPtr_R, handle))
+  if (.Call(XGCheckNullPtr_R, handle))
     return(TRUE)
+
   return(FALSE)
 }
 
@@ -537,7 +540,7 @@ xgb.ntree <- function(bst) {
 print.xgb.Booster <- function(x, verbose = FALSE, ...) {
   cat('##### xgb.Booster\n')
 
-  valid_handle <- is.null.handle(x$handle)
+  valid_handle <- !is.null.handle(x$handle)
   if (!valid_handle)
     cat("Handle is invalid! Suggest using xgb.Booster.complete\n")
 

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

@@ -83,5 +83,5 @@ xgb.create.features <- function(model, data, ...){
   check.deprecation(...)
   pred_with_leaf <- predict(model, data, predleaf = TRUE)
   cols <- lapply(as.data.frame(pred_with_leaf), factor)
-  cBind(data, sparse.model.matrix( ~ . -1, cols))
+  cbind(data, sparse.model.matrix( ~ . -1, cols))
 }

R-package/R/xgb.cv.R

@@ -34,6 +34,7 @@
 #'   \item \code{rmse} Rooted mean square error
 #'   \item \code{logloss} negative log-likelihood function
 #'   \item \code{auc} Area under curve
+#'   \item \code{aucpr} Area under PR curve
 #'   \item \code{merror} Exact matching error, used to evaluate multi-class classification
 #' }
 #' @param obj customized objective function. Returns gradient and second order 
@@ -82,12 +83,13 @@
 #'   \item \code{params} parameters that were passed to the xgboost library. Note that it does not 
 #'         capture parameters changed by the \code{\link{cb.reset.parameters}} callback.
 #'   \item \code{callbacks} callback functions that were either automatically assigned or 
-#'         explicitely passed.
+#'         explicitly passed.
 #'   \item \code{evaluation_log} evaluation history storead as a \code{data.table} with the
 #'         first column corresponding to iteration number and the rest corresponding to the 
 #'         CV-based evaluation means and standard deviations for the training and test CV-sets.
 #'         It is created by the \code{\link{cb.evaluation.log}} callback.
 #'   \item \code{niter} number of boosting iterations.
+#'   \item \code{nfeatures} number of features in training data.
 #'   \item \code{folds} the list of CV folds' indices - either those passed through the \code{folds} 
 #'         parameter or randomly generated.
 #'   \item \code{best_iteration} iteration number with the best evaluation metric value
@@ -184,6 +186,7 @@ xgb.cv <- function(params=list(), data, nrounds, nfold, label = NULL, missing =
     handle <- xgb.Booster.handle(params, list(dtrain, dtest))
     list(dtrain = dtrain, bst = handle, watchlist = list(train = dtrain, test=dtest), index = folds[[k]])
   })
+  rm(dall)
   # a "basket" to collect some results from callbacks
   basket <- list()
 
@@ -221,6 +224,7 @@ xgb.cv <- function(params=list(), data, nrounds, nfold, label = NULL, missing =
     callbacks = callbacks,
     evaluation_log = evaluation_log,
     niter = end_iteration,
+    nfeatures = ncol(data),
     folds = folds
   )
   ret <- c(ret, basket)

R-package/R/xgb.dump.R

@@ -30,7 +30,8 @@
 #' 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'
-#' xgb.dump(bst, 'xgb.model.dump', 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/R/xgb.train.R

@@ -121,12 +121,13 @@
 #'   \itemize{
 #'      \item \code{rmse} root mean square error. \url{http://en.wikipedia.org/wiki/Root_mean_square_error}
 #'      \item \code{logloss} negative log-likelihood. \url{http://en.wikipedia.org/wiki/Log-likelihood}
-#'      \item \code{mlogloss} multiclass logloss. \url{https://www.kaggle.com/wiki/MultiClassLogLoss/}
+#'      \item \code{mlogloss} multiclass logloss. \url{http://wiki.fast.ai/index.php/Log_Loss}
 #'      \item \code{error} Binary classification error rate. It is calculated as \code{(# wrong cases) / (# all cases)}.
 #'            By default, it uses the 0.5 threshold for predicted values to define negative and positive instances.
 #'            Different threshold (e.g., 0.) could be specified as "error@0."
 #'      \item \code{merror} Multiclass classification error rate. It is calculated as \code{(# wrong cases) / (# all cases)}.
 #'      \item \code{auc} Area under the curve. \url{http://en.wikipedia.org/wiki/Receiver_operating_characteristic#'Area_under_curve} for ranking evaluation.
+#'      \item \code{aucpr} Area under the PR curve. \url{https://en.wikipedia.org/wiki/Precision_and_recall} for ranking evaluation.
 #'      \item \code{ndcg} Normalized Discounted Cumulative Gain (for ranking task). \url{http://en.wikipedia.org/wiki/NDCG}
 #'   }
 #' 
@@ -162,6 +163,7 @@
 #'         (only available with early stopping).
 #'   \item \code{feature_names} names of the training dataset features
 #'         (only when comun names were defined in training data).
+#'   \item \code{nfeatures} number of features in training data.
 #' }
 #' 
 #' @seealso
@@ -351,8 +353,8 @@ xgb.train <- function(params = list(), data, nrounds, watchlist = list(),
     if (inherits(xgb_model, 'xgb.Booster') &&
         !is_update &&
         !is.null(xgb_model$evaluation_log) &&
-        all.equal(colnames(evaluation_log),
+        isTRUE(all.equal(colnames(evaluation_log),
-                  colnames(xgb_model$evaluation_log))) {
+                         colnames(xgb_model$evaluation_log)))) {
       evaluation_log <- rbindlist(list(xgb_model$evaluation_log, evaluation_log))
     }
     bst$evaluation_log <- evaluation_log
@@ -363,6 +365,7 @@ xgb.train <- function(params = list(), data, nrounds, watchlist = list(),
   bst$callbacks <- callbacks
   if (!is.null(colnames(dtrain)))
     bst$feature_names <- colnames(dtrain)
+  bst$nfeatures <- ncol(dtrain)
 
   return(bst)
 }

R-package/R/xgboost.R

@@ -77,10 +77,11 @@ NULL
 
 # Various imports
 #' @importClassesFrom Matrix dgCMatrix dgeMatrix
-#' @importFrom Matrix cBind
 #' @importFrom Matrix colSums
 #' @importFrom Matrix sparse.model.matrix
 #' @importFrom Matrix sparseVector
+#' @importFrom Matrix sparseMatrix
+#' @importFrom Matrix t
 #' @importFrom data.table data.table
 #' @importFrom data.table is.data.table
 #' @importFrom data.table as.data.table

R-package/README.md

@@ -30,4 +30,4 @@ Examples
 Development
 -----------
 
-* See the [R Package section](https://xgboost.readthedocs.io/en/latest/how_to/contribute.html#r-package) of the contributiors guide.
+* See the [R Package section](https://xgboost.readthedocs.io/en/latest/how_to/contribute.html#r-package) of the contributors guide.

R-package/demo/basic_walkthrough.R

@@ -99,7 +99,8 @@ err <- as.numeric(sum(as.integer(pred > 0.5) != label))/length(label)
 print(paste("test-error=", err))
 
 # You can dump the tree you learned using xgb.dump into a text file
-xgb.dump(bst, "dump.raw.txt", with_stats = T)
+dump_path = file.path(tempdir(), 'dump.raw.txt')
+xgb.dump(bst, dump_path, with_stats = T)
 
 # Finally, you can check which features are the most important.
 print("Most important features (look at column Gain):")

R-package/demo/predict_leaf_indices.R

@@ -32,7 +32,7 @@ create.new.tree.features <- function(model, original.features){
     leaf.id <- sort(unique(pred_with_leaf[,i]))
     cols[[i]] <- factor(x = pred_with_leaf[,i], level = leaf.id)
   }
-  cBind(original.features, sparse.model.matrix( ~ . -1, as.data.frame(cols)))
+  cbind(original.features, sparse.model.matrix( ~ . -1, as.data.frame(cols)))
 }
 
 # Convert previous features to one hot encoding

R-package/man/cb.gblinear.history.Rd

@@ -0,0 +1,95 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/callbacks.R
+\name{cb.gblinear.history}
+\alias{cb.gblinear.history}
+\title{Callback closure for collecting the model coefficients history of a gblinear booster
+during its training.}
+\usage{
+cb.gblinear.history(sparse = FALSE)
+}
+\arguments{
+\item{sparse}{when set to FALSE/TURE, a dense/sparse matrix is used to store the result.
+Sparse format is useful when one expects only a subset of coefficients to be non-zero,
+when using the "thrifty" feature selector with fairly small number of top features
+selected per iteration.}
+}
+\value{
+Results are stored in the \code{coefs} element of the closure.
+The \code{\link{xgb.gblinear.history}} convenience function provides an easy way to access it.
+With \code{xgb.train}, it is either a dense of a sparse matrix.
+While with \code{xgb.cv}, it is a list (an element per each fold) of such matrices.
+}
+\description{
+Callback closure for collecting the model coefficients history of a gblinear booster
+during its training.
+}
+\details{
+To keep things fast and simple, gblinear booster does not internally store the history of linear
+model coefficients at each boosting iteration. This callback provides a workaround for storing
+the coefficients' path, by extracting them after each training iteration.
+
+Callback function expects the following values to be set in its calling frame:
+\code{bst} (or \code{bst_folds}).
+}
+\examples{
+#### Binary classification:
+#
+# In the iris dataset, it is hard to linearly separate Versicolor class from the rest
+# without considering the 2nd order interactions:
+require(magrittr)
+x <- model.matrix(Species ~ .^2, iris)[,-1]
+colnames(x)
+dtrain <- xgb.DMatrix(scale(x), label = 1*(iris$Species == "versicolor"))
+param <- list(booster = "gblinear", objective = "reg:logistic", eval_metric = "auc",
+              lambda = 0.0003, alpha = 0.0003, nthread = 2)
+# For 'shotgun', which is a default linear updater, using high eta values may result in
+# unstable behaviour in some datasets. With this simple dataset, however, the high learning
+# rate does not break the convergence, but allows us to illustrate the typical pattern of
+# "stochastic explosion" behaviour of this lock-free algorithm at early boosting iterations.
+bst <- xgb.train(param, dtrain, list(tr=dtrain), nrounds = 200, eta = 1.,
+                 callbacks = list(cb.gblinear.history()))
+# Extract the coefficients' path and plot them vs boosting iteration number:
+coef_path <- xgb.gblinear.history(bst)
+matplot(coef_path, type = 'l')
+
+# With the deterministic coordinate descent updater, it is safer to use higher learning rates.
+# Will try the classical componentwise boosting which selects a single best feature per round:
+bst <- xgb.train(param, dtrain, list(tr=dtrain), nrounds = 200, eta = 0.8,
+                 updater = 'coord_descent', feature_selector = 'thrifty', top_k = 1,
+                 callbacks = list(cb.gblinear.history()))
+xgb.gblinear.history(bst) \%>\% matplot(type = 'l')
+#  Componentwise boosting is known to have similar effect to Lasso regularization.
+# Try experimenting with various values of top_k, eta, nrounds,
+# as well as different feature_selectors.
+
+# For xgb.cv:
+bst <- xgb.cv(param, dtrain, nfold = 5, nrounds = 100, eta = 0.8,
+             callbacks = list(cb.gblinear.history()))
+# coefficients in the CV fold #3
+xgb.gblinear.history(bst)[[3]] \%>\% matplot(type = 'l')
+
+
+#### Multiclass classification:
+#
+dtrain <- xgb.DMatrix(scale(x), label = as.numeric(iris$Species) - 1)
+param <- list(booster = "gblinear", objective = "multi:softprob", num_class = 3,
+              lambda = 0.0003, alpha = 0.0003, nthread = 2)
+# For the default linear updater 'shotgun' it sometimes is helpful
+# to use smaller eta to reduce instability
+bst <- xgb.train(param, dtrain, list(tr=dtrain), nrounds = 70, eta = 0.5,
+                 callbacks = list(cb.gblinear.history()))
+# Will plot the coefficient paths separately for each class:
+xgb.gblinear.history(bst, class_index = 0) \%>\% matplot(type = 'l')
+xgb.gblinear.history(bst, class_index = 1) \%>\% matplot(type = 'l')
+xgb.gblinear.history(bst, class_index = 2) \%>\% matplot(type = 'l')
+
+# CV:
+bst <- xgb.cv(param, dtrain, nfold = 5, nrounds = 70, eta = 0.5,
+              callbacks = list(cb.gblinear.history(FALSE)))
+# 1st forld of 1st class
+xgb.gblinear.history(bst, class_index = 0)[[1]] \%>\% matplot(type = 'l')
+
+}
+\seealso{
+\code{\link{callbacks}}, \code{\link{xgb.gblinear.history}}.
+}

R-package/man/xgb.cv.Rd

@@ -51,6 +51,7 @@ from each CV model. This parameter engages the \code{\link{cb.cv.predict}} callb
   \item \code{rmse} Rooted mean square error
   \item \code{logloss} negative log-likelihood function
   \item \code{auc} Area under curve
+  \item \code{aucpr} Area under PR curve
   \item \code{merror} Exact matching error, used to evaluate multi-class classification
 }}
 
@@ -98,12 +99,13 @@ An object of class \code{xgb.cv.synchronous} with the following elements:
   \item \code{params} parameters that were passed to the xgboost library. Note that it does not 
         capture parameters changed by the \code{\link{cb.reset.parameters}} callback.
   \item \code{callbacks} callback functions that were either automatically assigned or 
-        explicitely passed.
+        explicitly passed.
   \item \code{evaluation_log} evaluation history storead as a \code{data.table} with the
         first column corresponding to iteration number and the rest corresponding to the 
         CV-based evaluation means and standard deviations for the training and test CV-sets.
         It is created by the \code{\link{cb.evaluation.log}} callback.
   \item \code{niter} number of boosting iterations.
+  \item \code{nfeatures} number of features in training data.
   \item \code{folds} the list of CV folds' indices - either those passed through the \code{folds} 
         parameter or randomly generated.
   \item \code{best_iteration} iteration number with the best evaluation metric value

R-package/man/xgb.dump.Rd

@@ -44,7 +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'
-xgb.dump(bst, 'xgb.model.dump', 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.gblinear.history.Rd

@@ -0,0 +1,29 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/callbacks.R
+\name{xgb.gblinear.history}
+\alias{xgb.gblinear.history}
+\title{Extract gblinear coefficients history.}
+\usage{
+xgb.gblinear.history(model, class_index = NULL)
+}
+\arguments{
+\item{model}{either an \code{xgb.Booster} or a result of \code{xgb.cv()}, trained
+using the \code{cb.gblinear.history()} callback.}
+
+\item{class_index}{zero-based class index to extract the coefficients for only that
+specific class in a multinomial multiclass model. When it is NULL, all the
+coeffients are returned. Has no effect in non-multiclass models.}
+}
+\value{
+For an \code{xgb.train} result, a matrix (either dense or sparse) with the columns
+corresponding to iteration's coefficients (in the order as \code{xgb.dump()} would
+return) and the rows corresponding to boosting iterations.
+
+For an \code{xgb.cv} result, a list of such matrices is returned with the elements
+corresponding to CV folds.
+}
+\description{
+A helper function to extract the matrix of linear coefficients' history
+from a gblinear model created while using the \code{cb.gblinear.history()}
+callback.
+}

R-package/man/xgb.train.Rd

@@ -155,6 +155,7 @@ An object of class \code{xgb.Booster} with the following elements:
         (only available with early stopping).
   \item \code{feature_names} names of the training dataset features
         (only when comun names were defined in training data).
+  \item \code{nfeatures} number of features in training data.
 }
 }
 \description{
@@ -179,12 +180,13 @@ The folloiwing is the list of built-in metrics for which Xgboost provides optimi
   \itemize{
      \item \code{rmse} root mean square error. \url{http://en.wikipedia.org/wiki/Root_mean_square_error}
      \item \code{logloss} negative log-likelihood. \url{http://en.wikipedia.org/wiki/Log-likelihood}
-     \item \code{mlogloss} multiclass logloss. \url{https://www.kaggle.com/wiki/MultiClassLogLoss/}
+     \item \code{mlogloss} multiclass logloss. \url{http://wiki.fast.ai/index.php/Log_Loss}
      \item \code{error} Binary classification error rate. It is calculated as \code{(# wrong cases) / (# all cases)}.
            By default, it uses the 0.5 threshold for predicted values to define negative and positive instances.
            Different threshold (e.g., 0.) could be specified as "error@0."
      \item \code{merror} Multiclass classification error rate. It is calculated as \code{(# wrong cases) / (# all cases)}.
      \item \code{auc} Area under the curve. \url{http://en.wikipedia.org/wiki/Receiver_operating_characteristic#'Area_under_curve} for ranking evaluation.
+     \item \code{aucpr} Area under the PR curve. \url{https://en.wikipedia.org/wiki/Precision_and_recall} for ranking evaluation.
      \item \code{ndcg} Normalized Discounted Cumulative Gain (for ranking task). \url{http://en.wikipedia.org/wiki/NDCG}
   }
 

R-package/remove_warning_suppression_pragma.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+# remove all #pragma's that suppress compiler warnings
+set -e
+set -x
+for file in xgboost/src/dmlc-core/include/dmlc/*.h
+do
+  sed -i.bak -e 's/^.*#pragma GCC diagnostic.*$//' -e 's/^.*#pragma clang diagnostic.*$//' -e 's/^.*#pragma warning.*$//' "${file}"
+done
+for file in xgboost/src/dmlc-core/include/dmlc/*.h.bak
+do
+  rm "${file}"
+done
+set +x
+set +e

R-package/src/Makevars.in

@@ -10,6 +10,12 @@ XGB_RFLAGS = -DXGBOOST_STRICT_R_MODE=1 -DDMLC_LOG_BEFORE_THROW=0\
            -DDMLC_LOG_CUSTOMIZE=1 -DXGBOOST_CUSTOMIZE_LOGGER=1\
            -DRABIT_CUSTOMIZE_MSG_ -DRABIT_STRICT_CXX98_
 
+# disable the use of thread_local for 32 bit windows:
+ifeq ($(R_OSTYPE)$(WIN),windows)
+    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)

R-package/src/Makevars.win

@@ -4,7 +4,7 @@ ENABLE_STD_THREAD=0
 # _*_ mode: Makefile; _*_
 
 # This file is only used for windows compilation from github
-# It will be replaced by Makevars in CRAN version
+# It will be replaced with Makevars.in for the CRAN version
 .PHONY: all xgblib
 all: $(SHLIB)
 $(SHLIB): xgblib
@@ -22,6 +22,12 @@ XGB_RFLAGS = -DXGBOOST_STRICT_R_MODE=1 -DDMLC_LOG_BEFORE_THROW=0\
            -DDMLC_LOG_CUSTOMIZE=1 -DXGBOOST_CUSTOMIZE_LOGGER=1\
            -DRABIT_CUSTOMIZE_MSG_ -DRABIT_STRICT_CXX98_
 
+# disable the use of thread_local for 32 bit windows:
+ifeq ($(R_OSTYPE)$(WIN),windows)
+    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)

R-package/src/init.c

@@ -19,10 +19,10 @@ extern SEXP XGBoosterBoostOneIter_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGBoosterCreate_R(SEXP);
 extern SEXP XGBoosterDumpModel_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGBoosterEvalOneIter_R(SEXP, SEXP, SEXP, SEXP);
-extern SEXP XGBoosterGetAttr_R(SEXP, SEXP);
 extern SEXP XGBoosterGetAttrNames_R(SEXP);
-extern SEXP XGBoosterLoadModel_R(SEXP, SEXP);
+extern SEXP XGBoosterGetAttr_R(SEXP, SEXP);
 extern SEXP XGBoosterLoadModelFromRaw_R(SEXP, SEXP);
+extern SEXP XGBoosterLoadModel_R(SEXP, SEXP);
 extern SEXP XGBoosterModelToRaw_R(SEXP);
 extern SEXP XGBoosterPredict_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGBoosterSaveModel_R(SEXP, SEXP);
@@ -45,10 +45,10 @@ static const R_CallMethodDef CallEntries[] = {
   {"XGBoosterCreate_R",           (DL_FUNC) &XGBoosterCreate_R,           1},
   {"XGBoosterDumpModel_R",        (DL_FUNC) &XGBoosterDumpModel_R,        4},
   {"XGBoosterEvalOneIter_R",      (DL_FUNC) &XGBoosterEvalOneIter_R,      4},
-  {"XGBoosterGetAttr_R",          (DL_FUNC) &XGBoosterGetAttr_R,          2},
   {"XGBoosterGetAttrNames_R",     (DL_FUNC) &XGBoosterGetAttrNames_R,     1},
-  {"XGBoosterLoadModel_R",        (DL_FUNC) &XGBoosterLoadModel_R,        2},
+  {"XGBoosterGetAttr_R",          (DL_FUNC) &XGBoosterGetAttr_R,          2},
   {"XGBoosterLoadModelFromRaw_R", (DL_FUNC) &XGBoosterLoadModelFromRaw_R, 2},
+  {"XGBoosterLoadModel_R",        (DL_FUNC) &XGBoosterLoadModel_R,        2},
   {"XGBoosterModelToRaw_R",       (DL_FUNC) &XGBoosterModelToRaw_R,       1},
   {"XGBoosterPredict_R",          (DL_FUNC) &XGBoosterPredict_R,          4},
   {"XGBoosterSaveModel_R",        (DL_FUNC) &XGBoosterSaveModel_R,        2},

R-package/tests/testthat/test_basic.R

@@ -11,6 +11,7 @@ set.seed(1994)
 # disable some tests for Win32
 windows_flag = .Platform$OS.type == "windows" &&
                .Machine$sizeof.pointer != 8
+solaris_flag = (Sys.info()['sysname'] == "SunOS")
 
 test_that("train and predict binary classification", {
   nrounds = 2
@@ -152,20 +153,20 @@ test_that("training continuation works", {
   bst1 <- xgb.train(param, dtrain, nrounds = 2, watchlist, verbose = 0)
   # continue for two more:
   bst2 <- xgb.train(param, dtrain, nrounds = 2, watchlist, verbose = 0, xgb_model = bst1)
-  if (!windows_flag)
+  if (!windows_flag && !solaris_flag)
     expect_equal(bst$raw, bst2$raw)
   expect_false(is.null(bst2$evaluation_log))
   expect_equal(dim(bst2$evaluation_log), c(4, 2))
   expect_equal(bst2$evaluation_log, bst$evaluation_log)
   # test continuing from raw model data
   bst2 <- xgb.train(param, dtrain, nrounds = 2, watchlist, verbose = 0, xgb_model = bst1$raw)
-  if (!windows_flag)
+  if (!windows_flag && !solaris_flag)
     expect_equal(bst$raw, bst2$raw)
   expect_equal(dim(bst2$evaluation_log), c(2, 2))
   # test continuing from a model in file
   xgb.save(bst1, "xgboost.model")
   bst2 <- xgb.train(param, dtrain, nrounds = 2, watchlist, verbose = 0, xgb_model = "xgboost.model")
-  if (!windows_flag)
+  if (!windows_flag && !solaris_flag)
     expect_equal(bst$raw, bst2$raw)
   expect_equal(dim(bst2$evaluation_log), c(2, 2))
 })

R-package/tests/testthat/test_glm.R

@@ -2,18 +2,47 @@ context('Test generalized linear models')
 
 require(xgboost)
 
-test_that("glm works", {
+test_that("gblinear works", {
   data(agaricus.train, package='xgboost')
   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)
-  expect_equal(class(dtrain), "xgb.DMatrix")
+
-  expect_equal(class(dtest), "xgb.DMatrix")
   param <- list(objective = "binary:logistic", booster = "gblinear",
-                nthread = 2, alpha = 0.0001, lambda = 1)
+                nthread = 2, eta = 0.8, alpha = 0.0001, lambda = 0.0001)
   watchlist <- list(eval = dtest, train = dtrain)
-  num_round <- 2
+
-  bst <- xgb.train(param, dtrain, num_round, watchlist)
+  n <- 5         # iterations
+  ERR_UL <- 0.005 # upper limit for the test set error
+  VERB <- 0      # chatterbox switch
+
+  param$updater = 'shotgun'
+  bst <- xgb.train(param, dtrain, n, watchlist, verbose = VERB, feature_selector = 'shuffle')
   ypred <- predict(bst, dtest)
   expect_equal(length(getinfo(dtest, 'label')), 1611)
+  expect_lt(bst$evaluation_log$eval_error[n], ERR_UL)
+
+  bst <- xgb.train(param, dtrain, n, watchlist, verbose = VERB, feature_selector = 'cyclic',
+                   callbacks = list(cb.gblinear.history()))
+  expect_lt(bst$evaluation_log$eval_error[n], ERR_UL)
+  h <- xgb.gblinear.history(bst)
+  expect_equal(dim(h), c(n, ncol(dtrain) + 1))
+  expect_is(h, "matrix")
+
+  param$updater = 'coord_descent'
+  bst <- xgb.train(param, dtrain, n, watchlist, verbose = VERB, feature_selector = 'cyclic')
+  expect_lt(bst$evaluation_log$eval_error[n], ERR_UL)
+
+  bst <- xgb.train(param, dtrain, n, watchlist, verbose = VERB, feature_selector = 'shuffle')
+  expect_lt(bst$evaluation_log$eval_error[n], ERR_UL)
+
+  bst <- xgb.train(param, dtrain, 2, watchlist, verbose = VERB, feature_selector = 'greedy')
+  expect_lt(bst$evaluation_log$eval_error[2], ERR_UL)
+
+  bst <- xgb.train(param, dtrain, n, watchlist, verbose = VERB, feature_selector = 'thrifty',
+                   top_n = 50, callbacks = list(cb.gblinear.history(sparse = TRUE)))
+  expect_lt(bst$evaluation_log$eval_error[n], ERR_UL)
+  h <- xgb.gblinear.history(bst)
+  expect_equal(dim(h), c(n, ncol(dtrain) + 1))
+  expect_s4_class(h, "dgCMatrix")
 })

R-package/tests/testthat/test_helpers.R

@@ -5,6 +5,8 @@ require(data.table)
 require(Matrix)
 require(vcd, quietly = TRUE)
 
+float_tolerance = 5e-6
+
 set.seed(1982)
 data(Arthritis)
 df <- data.table(Arthritis, keep.rownames = F)
@@ -40,9 +42,10 @@ mbst.GLM <- xgboost(data = as.matrix(iris[, -5]), label = mlabel, verbose = 0,
 
 test_that("xgb.dump works", {
   expect_length(xgb.dump(bst.Tree), 200)
-  expect_true(xgb.dump(bst.Tree, 'xgb.model.dump', with_stats = T))
+  dump_file = file.path(tempdir(), 'xgb.model.dump')
-  expect_true(file.exists('xgb.model.dump'))
+  expect_true(xgb.dump(bst.Tree, dump_file, with_stats = T))
-  expect_gt(file.size('xgb.model.dump'), 8000)
+  expect_true(file.exists(dump_file))
+  expect_gt(file.size(dump_file), 8000)
 
   # JSON format
   dmp <- xgb.dump(bst.Tree, dump_format = "json")
@@ -85,7 +88,8 @@ test_that("predict feature contributions works", {
   X <- sparse_matrix
   colnames(X) <- NULL
   expect_error(pred_contr_ <- predict(bst.Tree, X, predcontrib = TRUE), regexp = NA)
-  expect_equal(pred_contr, pred_contr_, check.attributes = FALSE)
+  expect_equal(pred_contr, pred_contr_, check.attributes = FALSE,
+               tolerance = float_tolerance)
 
   # gbtree binary classifier (approximate method)
   expect_error(pred_contr <- predict(bst.Tree, sparse_matrix, predcontrib = TRUE, approxcontrib = TRUE), regexp = NA)
@@ -104,7 +108,8 @@ test_that("predict feature contributions works", {
   coefs <- xgb.dump(bst.GLM)[-c(1,2,4)] %>% as.numeric
   coefs <- c(coefs[-1], coefs[1]) # intercept must be the last
   pred_contr_manual <- sweep(cbind(sparse_matrix, 1), 2, coefs, FUN="*")
-  expect_equal(as.numeric(pred_contr), as.numeric(pred_contr_manual), 1e-5)
+  expect_equal(as.numeric(pred_contr), as.numeric(pred_contr_manual),
+               tolerance = float_tolerance)
 
   # gbtree multiclass
   pred <- predict(mbst.Tree, as.matrix(iris[, -5]), outputmargin = TRUE, reshape = TRUE)
@@ -123,11 +128,12 @@ test_that("predict feature contributions works", {
   coefs_all <- xgb.dump(mbst.GLM)[-c(1,2,6)] %>% as.numeric %>% matrix(ncol = 3, byrow = TRUE)
   for (g in seq_along(pred_contr)) {
     expect_equal(colnames(pred_contr[[g]]), c(colnames(iris[, -5]), "BIAS"))
-    expect_lt(max(abs(rowSums(pred_contr[[g]]) - pred[, g])), 2e-6)
+    expect_lt(max(abs(rowSums(pred_contr[[g]]) - pred[, g])), float_tolerance)
     # manual calculation of linear terms
     coefs <- c(coefs_all[-1, g], coefs_all[1, g]) # intercept needs to be the last
     pred_contr_manual <- sweep(as.matrix(cbind(iris[,-5], 1)), 2, coefs, FUN="*")
-    expect_equal(as.numeric(pred_contr[[g]]), as.numeric(pred_contr_manual), 2e-6)
+    expect_equal(as.numeric(pred_contr[[g]]), as.numeric(pred_contr_manual),
+                 tolerance = float_tolerance)
   }
 })
 
@@ -171,14 +177,16 @@ if (grepl('Windows', Sys.info()[['sysname']]) ||
       # check that lossless conversion works with 17 digits
       # numeric -> character -> numeric
       X <- 10^runif(100, -20, 20)
-      X2X <- as.numeric(format(X, digits = 17))
+      if (capabilities('long.double')) {
-      expect_identical(X, X2X)
+          X2X <- as.numeric(format(X, digits = 17))
+          expect_identical(X, X2X)
+      }
       # retrieved attributes to be the same as written
       for (x in X) {
         xgb.attr(bst.Tree, "x") <- x
-        expect_identical(as.numeric(xgb.attr(bst.Tree, "x")), x)
+        expect_equal(as.numeric(xgb.attr(bst.Tree, "x")), x, tolerance = float_tolerance)
         xgb.attributes(bst.Tree) <- list(a = "A", b = x)
-        expect_identical(as.numeric(xgb.attr(bst.Tree, "b")), x)
+        expect_equal(as.numeric(xgb.attr(bst.Tree, "b")), x, tolerance = float_tolerance)
       }
     })
 }
@@ -187,7 +195,7 @@ test_that("xgb.Booster serializing as R object works", {
   saveRDS(bst.Tree, 'xgb.model.rds')
   bst <- readRDS('xgb.model.rds')
   dtrain <- xgb.DMatrix(sparse_matrix, label = label)
-  expect_equal(predict(bst.Tree, dtrain), predict(bst, dtrain))
+  expect_equal(predict(bst.Tree, dtrain), predict(bst, dtrain), tolerance = float_tolerance)
   expect_equal(xgb.dump(bst.Tree), xgb.dump(bst))
   xgb.save(bst, 'xgb.model')
   nil_ptr <- new("externalptr")
@@ -195,7 +203,7 @@ test_that("xgb.Booster serializing as R object works", {
   expect_true(identical(bst$handle, nil_ptr))
   bst <- xgb.Booster.complete(bst)
   expect_true(!identical(bst$handle, nil_ptr))
-  expect_equal(predict(bst.Tree, dtrain), predict(bst, dtrain))
+  expect_equal(predict(bst.Tree, dtrain), predict(bst, dtrain), tolerance = float_tolerance)
 })
 
 test_that("xgb.model.dt.tree works with and without feature names", {
@@ -233,13 +241,14 @@ test_that("xgb.importance works with and without feature names", {
   expect_output(str(importance.Tree), 'Feature.*\\"Age\\"')
 
   importance.Tree.0 <- xgb.importance(model = bst.Tree)
-  expect_equal(importance.Tree, importance.Tree.0)
+  expect_equal(importance.Tree, importance.Tree.0, tolerance = float_tolerance)
 
   # when model contains no feature names:
   bst.Tree.x <- bst.Tree
   bst.Tree.x$feature_names <- NULL
   importance.Tree.x <- xgb.importance(model = bst.Tree)
-  expect_equal(importance.Tree[, -1, with=FALSE], importance.Tree.x[, -1, with=FALSE])
+  expect_equal(importance.Tree[, -1, with=FALSE], importance.Tree.x[, -1, with=FALSE],
+               tolerance = float_tolerance)
 
   imp2plot <- xgb.plot.importance(importance_matrix = importance.Tree)
   expect_equal(colnames(imp2plot), c("Feature", "Gain", "Cover", "Frequency", "Importance"))

amalgamation/dmlc-minimum0.cc

@@ -7,6 +7,8 @@
 #include "../dmlc-core/src/io/recordio_split.cc"
 #include "../dmlc-core/src/io/input_split_base.cc"
 #include "../dmlc-core/src/io/local_filesys.cc"
+#include "../dmlc-core/src/io/filesys.cc"
+#include "../dmlc-core/src/io/indexed_recordio_split.cc"
 #include "../dmlc-core/src/data.cc"
 #include "../dmlc-core/src/io.cc"
 #include "../dmlc-core/src/recordio.cc"

amalgamation/xgboost-all0.cc

@@ -53,10 +53,16 @@
 #include "../src/tree/updater_histmaker.cc"
 #include "../src/tree/updater_skmaker.cc"
 
+// linear
+#include "../src/linear/linear_updater.cc"
+#include "../src/linear/updater_coordinate.cc"
+#include "../src/linear/updater_shotgun.cc"
+
 // global
 #include "../src/learner.cc"
 #include "../src/logging.cc"
 #include "../src/common/common.cc"
+#include "../src/common/host_device_vector.cc"
 #include "../src/common/hist_util.cc"
 
 // c_api

appveyor.yml

@@ -53,7 +53,7 @@ install:
           Import-Module "$Env:TEMP\appveyor-tool.ps1"
           Bootstrap
           $DEPS = "c('data.table','magrittr','stringi','ggplot2','DiagrammeR','Ckmeans.1d.dp','vcd','testthat','igraph','knitr','rmarkdown')"
-          cmd /c "R.exe -q -e ""install.packages($DEPS, repos='$CRAN', type='win.binary')"" 2>&1"
+          cmd.exe /c "R.exe -q -e ""install.packages($DEPS, repos='$CRAN', type='both')"" 2>&1"
         }
 
 build_script:
@@ -81,7 +81,7 @@ build_script:
     - if /i "%target%" == "rmingw" (
         make Rbuild &&
         ls -l &&
-        R.exe CMD INSTALL --no-multiarch xgboost*.tar.gz
+        R.exe CMD INSTALL xgboost*.tar.gz
       )
     # R package: cmake + VC2015
     - if /i "%target%" == "rmsvc" (
@@ -98,10 +98,9 @@ test_script:
     # 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&&
-        R.exe CMD check xgboost*.tar.gz --no-manual --no-build-vignettes --as-cran --install-args=--build --no-multiarch
+        R.exe CMD check xgboost*.tar.gz --no-manual --no-build-vignettes --as-cran --install-args=--build
       )
     # MSVC R package: run only the unit tests
-    #   TODO: create a binary msvc-built package to keep as an artifact
     - if /i "%target%" == "rmsvc" (
         cd build_rmsvc%ver%\R-package &&
         R.exe -q -e "library(testthat); setwd('tests'); source('testthat.R')"

build.sh

@@ -15,25 +15,21 @@ else
 
     if [[ ! -e ./rabit/Makefile ]]; then
         echo ""
-        echo "Please clone the rabit repository into this directory."
+        echo "Please init the rabit submodule:"
-        echo "Here are the commands:"
+        echo "git submodule update --init --recursive -- rabit"
-        echo "rm -rf rabit"
-        echo "git clone https://github.com/dmlc/rabit.git rabit"
         not_ready=1
     fi
     
     if [[ ! -e ./dmlc-core/Makefile ]]; then
         echo ""
-        echo "Please clone the dmlc-core repository into this directory."
+        echo "Please init the dmlc-core submodule:"
-        echo "Here are the commands:"
+        echo "git submodule update --init --recursive -- dmlc-core"
-        echo "rm -rf dmlc-core"
-        echo "git clone https://github.com/dmlc/dmlc-core.git dmlc-core"
         not_ready=1
     fi
 
     if [[ "${not_ready}" == "1" ]]; then
         echo ""
-        echo "Please fix the errors above and retry the build or reclone the repository with:"
+        echo "Please fix the errors above and retry the build, or reclone the repository with:"
         echo "git clone --recursive https://github.com/dmlc/xgboost.git"
         echo ""
         exit 1

cmake/Utils.cmake

@@ -54,10 +54,25 @@ function(set_default_configuration_release)
 	endif()
 endfunction(set_default_configuration_release)
 
+# Generate nvcc compiler flags given a list of architectures
+# Also generates PTX for the most recent architecture for forwards compatibility
 function(format_gencode_flags flags out)
+  # Set up architecture flags
+  if(NOT flags) 
+    if((CUDA_VERSION_MAJOR EQUAL 9) OR (CUDA_VERSION_MAJOR GREATER 9))
+      set(flags "35;50;52;60;61;70")
+    else()
+      set(flags "35;50;52;60;61")
+    endif()
+  endif()
+  # Generate SASS
   foreach(ver ${flags})
     set(${out} "${${out}}-gencode arch=compute_${ver},code=sm_${ver};")
   endforeach()
+  # Generate PTX for last architecture
+  list(GET flags -1 ver)
+  set(${out} "${${out}}-gencode arch=compute_${ver},code=compute_${ver};")
+  
   set(${out} "${${out}}" PARENT_SCOPE)
 endfunction(format_gencode_flags flags)
 

cmake/modules/FindLibR.cmake

@@ -117,7 +117,7 @@ else()
     # ask R for R_HOME 
     if(LIBR_EXECUTABLE)
       execute_process(
-        COMMAND ${LIBR_EXECUTABLE} "--slave" "--no-save" "-e" "cat(normalizePath(R.home(), winslash='/'))"
+        COMMAND ${LIBR_EXECUTABLE} "--slave" "--no-save" "-e" "cat(normalizePath(R.home(),winslash='/'))"
         OUTPUT_VARIABLE LIBR_HOME)
     endif()
     # if R executable not available, query R_HOME path from registry

demo/gpu_acceleration/README.md

@@ -2,8 +2,6 @@
 
 This demo shows how to train a model on the [forest cover type](https://archive.ics.uci.edu/ml/datasets/covertype) dataset using GPU acceleration. The forest cover type dataset has 581,012 rows and 54 features, making it time consuming to process. We compare the run-time and accuracy of the GPU and CPU histogram algorithms.
 
-This demo requires the [GPU plug-in](https://github.com/dmlc/xgboost/tree/master/plugin/updater_gpu) to be built and installed.
+This demo requires the [GPU plug-in](https://xgboost.readthedocs.io/en/latest/gpu/index.html) to be built and installed.
 
 The dataset is automatically loaded via the sklearn script. 
-
-

demo/guide-python/README.md

@@ -1,7 +1,7 @@
 XGBoost Python Feature Walkthrough
 ==================================
 * [Basic walkthrough of wrappers](basic_walkthrough.py)
-* [Cutomize loss function, and evaluation metric](custom_objective.py)
+* [Customize loss function, and evaluation metric](custom_objective.py)
 * [Boosting from existing prediction](boost_from_prediction.py)
 * [Predicting using first n trees](predict_first_ntree.py)
 * [Generalized Linear Model](generalized_linear_model.py)

demo/guide-python/cross_validation.py

@@ -42,7 +42,7 @@ xgb.cv(param, dtrain, num_round, nfold=5,
        metrics={'auc'}, seed=0, fpreproc=fpreproc)
 
 ###
-# you can also do cross validation with cutomized loss function
+# you can also do cross validation with customized loss function
 # See custom_objective.py
 ##
 print('running cross validation, with cutomsized loss function')

doc/README

@@ -1,7 +1,5 @@
 The documentation of xgboost is generated with recommonmark and sphinx.
 
 You can build it locally by typing "make html" in this folder.
-- clone https://github.com/tqchen/recommonmark to root
-- type make html
 
 Checkout https://recommonmark.readthedocs.org for guide on how to write markdown with extensions used in this doc, such as math formulas and table of content.

doc/_static/xgboost-theme/layout.html

@@ -56,7 +56,7 @@
       };
     </script>
 
-    {% for name in ['jquery.js', 'underscore.js', 'doctools.js', 'searchtools.js'] %}
+    {% for name in ['jquery.js', 'underscore.js', 'doctools.js', 'searchtools-new.js'] %}
     <script type="text/javascript" src="{{ pathto('_static/' + name, 1) }}"></script>
     {% endfor %}
 

doc/_static/xgboost.css

@@ -185,7 +185,7 @@ pre {
 
 .dropdown-menu li  {
     padding: 0px 0px;
-    width: 120px;
+    width: 100%;
 }
 .dropdown-menu li a {
     color: #0079b2;

doc/build.md

@@ -4,7 +4,7 @@ Installation Guide
 This page gives instructions on how to build and install the xgboost package from
 scratch on various systems. It consists of two steps:
 
-1. First build the shared library from the C++ codes (`libxgboost.so` for linux/osx and `libxgboost.dll` for windows).
+1. First build the shared library from the C++ codes (`libxgboost.so` for Linux/OSX and `xgboost.dll` for Windows).
    - Exception: for R-package installation please directly refer to the R package section.
 2. Then install the language packages (e.g. Python Package).
 
@@ -39,7 +39,7 @@ even better to send pull request if you can fix the problem.
 
 Our goal is to build the shared library:
 - On Linux/OSX the target library is `libxgboost.so`
-- On Windows the target library is `libxgboost.dll`
+- On Windows the target library is `xgboost.dll`
 
 The minimal building requirement is
 
@@ -85,12 +85,33 @@ Now, clone the repository
 
 ```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
+
+```config.mk
+export CC = gcc
+export CXX = g++
+```
+
+and replace these two lines into(5 or 6 or 7; depending on your gcc-version)
+
+```config.mk
+export CC = gcc-7
+export CXX = g++-7
+```
+
+To find your gcc version
+
+```bash
+gcc-version
 ```
 
 and build using the following commands
 
 ```bash
-cd xgboost; cp make/config.mk ./config.mk; make -j4
+make -j4
 ```
 head over to `Python Package Installation` for the next steps
 
@@ -111,12 +132,13 @@ After installing [Git for Windows](https://git-for-windows.github.io/), you shou
 All the following steps are in the `Git Bash`.
 
 In MinGW, `make` command comes with the name `mingw32-make`. You can add the following line into the `.bashrc` file.
-
 ```bash
 alias make='mingw32-make'
 ```
+(On 64-bit Windows, you should get [mingw64](https://sourceforge.net/projects/mingw-w64/) instead.) Make sure
+that the path to MinGW is in the system PATH.
 
-To build with MinGW
+To build with MinGW, type:
 
 ```bash
 cp make/mingw64.mk config.mk; make -j4
@@ -130,7 +152,7 @@ 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 libxgboost.dll into python-package\xgboost.
+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.
 
 Other versions of Visual Studio may work but are untested.
 
@@ -169,6 +191,8 @@ If build seems to use only a single process, you might try to append an option l
 
 ### Windows Binaries
 
+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. And you are good to follow the below instructions.
+
 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/)
 
 ### Customized Building

doc/conf.py

@@ -14,7 +14,6 @@
 import sys
 import os, subprocess
 import shlex
-import urllib
 # 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
 # documentation root, use os.path.abspath to make it absolute, like shown here.
@@ -79,6 +78,8 @@ master_doc = 'index'
 # Usually you set "language" from the command line for these cases.
 language = None
 
+autoclass_content = 'both'
+
 # There are two options for replacing |today|: either, you set today to some
 # non-false value, then it is used:
 #today = ''
@@ -164,8 +165,14 @@ def setup(app):
     # Add hook for building doxygen xml when needed
     # no c++ API for now
     # app.connect("builder-inited", generate_doxygen_xml)
-    urllib.urlretrieve('https://code.jquery.com/jquery-2.2.4.min.js',
+
-                       '_static/jquery.js')
+    # urlretrieve got moved in Python 3.x
+    try:
+      from urllib import urlretrieve
+    except ImportError:
+      from urllib.request import urlretrieve
+    urlretrieve('https://code.jquery.com/jquery-2.2.4.min.js',
+                '_static/jquery.js')
     app.add_config_value('recommonmark_config', {
             'url_resolver': lambda url: github_doc_root + url,
             'enable_eval_rst': True,

doc/how_to/external_memory.md

@@ -11,7 +11,7 @@ filename#cacheprefix
 The ```filename``` is the normal path to libsvm file you want to load in, ```cacheprefix``` is a
 path to a cache file that xgboost will use for external memory cache.
 
-The following code was extracted from [../demo/guide-python/external_memory.py](../demo/guide-python/external_memory.py)
+The following code was extracted from [../../demo/guide-python/external_memory.py](../../demo/guide-python/external_memory.py)
 ```python
 dtrain = xgb.DMatrix('../data/agaricus.txt.train#dtrain.cache')
 ```
@@ -28,7 +28,7 @@ Distributed Version
 -------------------
 The external memory mode naturally works on distributed version, you can simply set path like
 ```
-data = "hdfs:///path-to-data/#dtrain.cache"
+data = "hdfs://path-to-data/#dtrain.cache"
 ```
 xgboost will cache the data to the local position. When you run on YARN, the current folder is temporal
 so that you can directly use ```dtrain.cache``` to cache to current folder.

doc/parameter.md

@@ -96,7 +96,7 @@ Parameters for Tree Booster
   - A type of boosting process to run.
   - Choices: {'default', 'update'}
     - 'default': the normal boosting process which creates new trees.
-    - 'update': starts from an existing model and only updates its trees. In each boosting iteration, a tree from the initial model is taken, a specified sequence of updater plugins is run for that tree, and a modified tree is added to the new model. The new model would have either the same or smaller number of trees, depending on the number of boosting iteratons performed. Currently, the following built-in updater plugins could be meaningfully used with this process type: 'refresh', 'prune'. With 'update', one cannot use updater plugins that create new nrees.
+    - 'update': starts from an existing model and only updates its trees. In each boosting iteration, a tree from the initial model is taken, a specified sequence of updater plugins is run for that tree, and a modified tree is added to the new model. The new model would have either the same or smaller number of trees, depending on the number of boosting iteratons performed. Currently, the following built-in updater plugins could be meaningfully used with this process type: 'refresh', 'prune'. With 'update', one cannot use updater plugins that create new trees.
 * grow_policy, string [default='depthwise']
   - Controls a way new nodes are added to the tree.
   - Currently supported only if `tree_method` is set to 'hist'.
@@ -142,11 +142,14 @@ Additional parameters for Dart Booster
 Parameters for Linear Booster
 -----------------------------
 * lambda [default=0, alias: reg_lambda]
-  - L2 regularization term on weights, increase this value will make model more conservative.
+  - L2 regularization term on weights, increase this value will make model more conservative. Normalised to number of training examples.
 * alpha [default=0, alias: reg_alpha]
-  - L1 regularization term on weights, increase this value will make model more conservative.
+  - L1 regularization term on weights, increase this value will make model more conservative. Normalised to number of training examples.
-* lambda_bias [default=0, alias: reg_lambda_bias]
+* updater [default='shotgun']
-  - L2 regularization term on bias (no L1 reg on bias because it is not important)
+	- Linear model algorithm
+      - '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. 
+
 
 Parameters for Tweedie Regression
 ---------------------------------
@@ -165,8 +168,13 @@ Specify the learning task and the corresponding learning objective. The objectiv
   - "reg:logistic" --logistic regression
   - "binary:logistic" --logistic regression for binary classification, output probability
   - "binary:logitraw" --logistic regression for binary classification, output score before logistic transformation
+  - "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)
+  - "survival:cox" --Cox regression for right censored survival time data (negative values are considered right censored).
+    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
@@ -194,6 +202,7 @@ Specify the learning task and the corresponding learning objective. The objectiv
 training repeatedly
   - "poisson-nloglik": negative log-likelihood for Poisson regression
   - "gamma-nloglik": negative log-likelihood for gamma regression
+  - "cox-nloglik": negative partial log-likelihood for Cox proportional hazards regression
   - "gamma-deviance": residual deviance for gamma regression
   - "tweedie-nloglik": negative log-likelihood for Tweedie regression (at a specified value of the tweedie_variance_power parameter)
 * seed [default=0]

doc/python/python_intro.md

@@ -25,7 +25,9 @@ Data Interface
 --------------
 The XGBoost python module is able to load data from:
 - libsvm txt format file
-- Numpy 2D array, and
+- comma-separated values (CSV) file
+- Numpy 2D array
+- Scipy 2D sparse array, and
 - xgboost binary buffer file.
 
 The data is stored in a ```DMatrix``` object.
@@ -35,6 +37,16 @@ The data is stored in a ```DMatrix``` object.
 dtrain = xgb.DMatrix('train.svm.txt')
 dtest = xgb.DMatrix('test.svm.buffer')
 ```
+* To load a CSV file into ```DMatrix```:
+```python
+# label_column specifies the index of the column containing the true label
+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).)
+
 * To load a numpy array into ```DMatrix```:
 ```python
 data = np.random.rand(5, 10)  # 5 entities, each contains 10 features

doc/requirements.txt

@@ -0,0 +1,3 @@
+sphinx==1.5.6
+commonmark==0.5.4
+mock

doc/tutorials/dart.md

@@ -1,9 +1,9 @@
 DART booster
 ============
-[XGBoost](https://github.com/dmlc/xgboost)) mostly combines a huge number of regression trees with a small learning rate.
+[XGBoost](https://github.com/dmlc/xgboost) mostly combines a huge number of regression trees with a small learning rate.
 In this situation, trees added early are significant and trees added late are unimportant.
 
-Rasmi et al. proposed a new method to add dropout techniques from the deep neural net community to boosted trees, and reported better results in some situations.
+Vinayak and Gilad-Bachrach proposed a new method to add dropout techniques from the deep neural net community to boosted trees, and reported better results in some situations.
 
 This is a instruction of new tree booster `dart`.
 

doc/tutorials/monotonic.md

@@ -76,3 +76,15 @@ 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
+sure to set the `tree_method` parameter to one of `'exact'`, `'hist'`, and
+`'gpu_hist'`.
+
+**Note for the `'hist'` tree construction algorithm**.
+If `tree_method` is set to either `'hist'` or `'gpu_hist'`, enabling monotonic
+constraints may produce unnecessarily shallow trees. This is because the
+`'hist'` method reduces the number of candidate splits to be considered at each
+split. Monotonic constraints may wipe out all available split candidates, in
+which case no split is made. To reduce the effect, you may want to increase
+the `max_bin` parameter to consider more split candidates.

include/xgboost/base.h

@@ -81,20 +81,19 @@ namespace xgboost {
  * \brief unsigned integer type used in boost,
  *  used for feature index and row index.
  */
-typedef uint32_t bst_uint;
+using bst_uint = uint32_t;  // NOLINT
-typedef int32_t bst_int;
+using bst_int = int32_t;    // NOLINT
 /*! \brief long integers */
 typedef uint64_t bst_ulong;  // NOLINT(*)
 /*! \brief float type, used for storing statistics */
-typedef float bst_float;
+using bst_float = float;  // NOLINT
-
 
 namespace detail {
 /*! \brief Implementation of gradient statistics pair. Template specialisation
  * may be used to overload different gradients types e.g. low precision, high
  * precision, integer, floating point. */
 template <typename T>
-class bst_gpair_internal {
+class GradientPairInternal {
   /*! \brief gradient statistics */
   T grad_;
   /*! \brief second order gradient statistics */
@@ -104,23 +103,23 @@ class bst_gpair_internal {
   XGBOOST_DEVICE void SetHess(float h) { hess_ = h; }
 
  public:
-  typedef T value_t;
+  using ValueT = T;
 
-  XGBOOST_DEVICE bst_gpair_internal() : grad_(0), hess_(0) {}
+  XGBOOST_DEVICE GradientPairInternal() : grad_(0), hess_(0) {}
 
-  XGBOOST_DEVICE bst_gpair_internal(float grad, float hess) {
+  XGBOOST_DEVICE GradientPairInternal(float grad, float hess) {
     SetGrad(grad);
     SetHess(hess);
   }
 
   // Copy constructor if of same value type
-  XGBOOST_DEVICE bst_gpair_internal(const bst_gpair_internal<T> &g)
+  XGBOOST_DEVICE GradientPairInternal(const GradientPairInternal<T> &g)
-      : grad_(g.grad_), hess_(g.hess_) {}
+      : grad_(g.grad_), hess_(g.hess_) {}  // NOLINT
 
   // Copy constructor if different value type - use getters and setters to
   // perform conversion
   template <typename T2>
-  XGBOOST_DEVICE bst_gpair_internal(const bst_gpair_internal<T2> &g) {
+  XGBOOST_DEVICE explicit GradientPairInternal(const GradientPairInternal<T2> &g) {
     SetGrad(g.GetGrad());
     SetHess(g.GetHess());
   }
@@ -128,85 +127,85 @@ class bst_gpair_internal {
   XGBOOST_DEVICE float GetGrad() const { return grad_; }
   XGBOOST_DEVICE float GetHess() const { return hess_; }
 
-  XGBOOST_DEVICE bst_gpair_internal<T> &operator+=(
+  XGBOOST_DEVICE GradientPairInternal<T> &operator+=(
-      const bst_gpair_internal<T> &rhs) {
+      const GradientPairInternal<T> &rhs) {
     grad_ += rhs.grad_;
     hess_ += rhs.hess_;
     return *this;
   }
 
-  XGBOOST_DEVICE bst_gpair_internal<T> operator+(
+  XGBOOST_DEVICE GradientPairInternal<T> operator+(
-      const bst_gpair_internal<T> &rhs) const {
+      const GradientPairInternal<T> &rhs) const {
-    bst_gpair_internal<T> g;
+    GradientPairInternal<T> g;
     g.grad_ = grad_ + rhs.grad_;
     g.hess_ = hess_ + rhs.hess_;
     return g;
   }
 
-  XGBOOST_DEVICE bst_gpair_internal<T> &operator-=(
+  XGBOOST_DEVICE GradientPairInternal<T> &operator-=(
-      const bst_gpair_internal<T> &rhs) {
+      const GradientPairInternal<T> &rhs) {
     grad_ -= rhs.grad_;
     hess_ -= rhs.hess_;
     return *this;
   }
 
-  XGBOOST_DEVICE bst_gpair_internal<T> operator-(
+  XGBOOST_DEVICE GradientPairInternal<T> operator-(
-      const bst_gpair_internal<T> &rhs) const {
+      const GradientPairInternal<T> &rhs) const {
-    bst_gpair_internal<T> g;
+    GradientPairInternal<T> g;
     g.grad_ = grad_ - rhs.grad_;
     g.hess_ = hess_ - rhs.hess_;
     return g;
   }
 
-  XGBOOST_DEVICE bst_gpair_internal(int value) {
+  XGBOOST_DEVICE explicit GradientPairInternal(int value) {
-    *this = bst_gpair_internal<T>(static_cast<float>(value),
+    *this = GradientPairInternal<T>(static_cast<float>(value),
                                   static_cast<float>(value));
   }
 
   friend std::ostream &operator<<(std::ostream &os,
-                                  const bst_gpair_internal<T> &g) {
+                                  const GradientPairInternal<T> &g) {
     os << g.GetGrad() << "/" << g.GetHess();
     return os;
   }
 };
 
 template<>
-inline XGBOOST_DEVICE float bst_gpair_internal<int64_t>::GetGrad() const {
+inline XGBOOST_DEVICE float GradientPairInternal<int64_t>::GetGrad() const {
   return grad_ * 1e-4f;
 }
 template<>
-inline XGBOOST_DEVICE float bst_gpair_internal<int64_t>::GetHess() const {
+inline XGBOOST_DEVICE float GradientPairInternal<int64_t>::GetHess() const {
   return hess_ * 1e-4f;
 }
 template<>
-inline XGBOOST_DEVICE void bst_gpair_internal<int64_t>::SetGrad(float g) {
+inline XGBOOST_DEVICE void GradientPairInternal<int64_t>::SetGrad(float g) {
   grad_ = static_cast<int64_t>(std::round(g * 1e4));
 }
 template<>
-inline XGBOOST_DEVICE void bst_gpair_internal<int64_t>::SetHess(float h) {
+inline XGBOOST_DEVICE void GradientPairInternal<int64_t>::SetHess(float h) {
   hess_ = static_cast<int64_t>(std::round(h * 1e4));
 }
 
 }  // namespace detail
 
 /*! \brief gradient statistics pair usually needed in gradient boosting */
-typedef detail::bst_gpair_internal<float> bst_gpair;
+using GradientPair = detail::GradientPairInternal<float>;
 
 /*! \brief High precision gradient statistics pair */
-typedef detail::bst_gpair_internal<double> bst_gpair_precise;
+using GradientPairPrecise = detail::GradientPairInternal<double>;
 
 /*! \brief High precision gradient statistics pair with integer backed
  * storage. Operators are associative where floating point versions are not
  * associative. */
-typedef detail::bst_gpair_internal<int64_t> bst_gpair_integer;
+using GradientPairInteger = detail::GradientPairInternal<int64_t>;
 
 /*! \brief small eps gap for minimum split decision. */
-const bst_float rt_eps = 1e-6f;
+const bst_float kRtEps = 1e-6f;
 
 /*! \brief define unsigned long for openmp loop */
-typedef dmlc::omp_ulong omp_ulong;
+using omp_ulong = dmlc::omp_ulong;  // NOLINT
 /*! \brief define unsigned int for openmp loop */
-typedef dmlc::omp_uint bst_omp_uint;
+using bst_omp_uint = dmlc::omp_uint;  // NOLINT
 
 /*!
  * \brief define compatible keywords in g++

include/xgboost/c_api.h

@@ -30,16 +30,16 @@ typedef uint64_t bst_ulong;  // NOLINT(*)
 
 
 /*! \brief handle to DMatrix */
-typedef void *DMatrixHandle;
+typedef void *DMatrixHandle;  // NOLINT(*)
 /*! \brief handle to Booster */
-typedef void *BoosterHandle;
+typedef void *BoosterHandle;  // NOLINT(*)
 /*! \brief handle to a data iterator */
-typedef void *DataIterHandle;
+typedef void *DataIterHandle;  // NOLINT(*)
 /*! \brief handle to a internal data holder. */
-typedef void *DataHolderHandle;
+typedef void *DataHolderHandle;  // NOLINT(*)
 
 /*! \brief Mini batch used in XGBoost Data Iteration */
-typedef struct {
+typedef struct {  // NOLINT(*)
   /*! \brief number of rows in the minibatch */
   size_t size;
   /*! \brief row pointer to the rows in the data */
@@ -66,7 +66,7 @@ typedef struct {
  * \param handle The handle to the callback.
  * \param batch The data content to be set.
  */
-XGB_EXTERN_C typedef int XGBCallbackSetData(
+XGB_EXTERN_C typedef int XGBCallbackSetData(  // NOLINT(*)
     DataHolderHandle handle, XGBoostBatchCSR batch);
 
 /*!
@@ -80,9 +80,8 @@ XGB_EXTERN_C typedef int XGBCallbackSetData(
  * \param set_function_handle The handle to be passed to set function.
  * \return 0 if we are reaching the end and batch is not returned.
  */
-XGB_EXTERN_C typedef int XGBCallbackDataIterNext(
+XGB_EXTERN_C typedef int XGBCallbackDataIterNext(  // NOLINT(*)
-    DataIterHandle data_handle,
+    DataIterHandle data_handle, XGBCallbackSetData *set_function,
-    XGBCallbackSetData* set_function,
     DataHolderHandle set_function_handle);
 
 /*!
@@ -95,7 +94,7 @@ XGB_EXTERN_C typedef int XGBCallbackDataIterNext(
  *  this function is thread safe and can be called by different thread
  * \return const char* error information
  */
-XGB_DLL const char *XGBGetLastError();
+XGB_DLL const char *XGBGetLastError(void);
 
 /*!
  * \brief load a data matrix
@@ -216,11 +215,9 @@ XGB_DLL int XGDMatrixCreateFromMat(const float *data,
  * \param nthread number of threads (up to maximum cores available, if <=0 use all cores)
  * \return 0 when success, -1 when failure happens
  */
-XGB_DLL int XGDMatrixCreateFromMat_omp(const float *data,
+XGB_DLL int XGDMatrixCreateFromMat_omp(const float *data,  // NOLINT
-                                       bst_ulong nrow,
+                                       bst_ulong nrow, bst_ulong ncol,
-                                       bst_ulong ncol,
+                                       float missing, DMatrixHandle *out,
-                                       float missing,
-                                       DMatrixHandle *out,
                                        int nthread);
 /*!
  * \brief create a new dmatrix from sliced content of existing matrix

include/xgboost/data.h

@@ -12,6 +12,7 @@
 #include <string>
 #include <memory>
 #include <vector>
+#include <numeric>
 #include "./base.h"
 
 namespace xgboost {
@@ -29,44 +30,45 @@ enum DataType {
 /*!
  * \brief Meta information about dataset, always sit in memory.
  */
-struct MetaInfo {
+class MetaInfo {
+ public:
   /*! \brief number of rows in the data */
-  uint64_t num_row;
+  uint64_t num_row_{0};
   /*! \brief number of columns in the data */
-  uint64_t num_col;
+  uint64_t num_col_{0};
   /*! \brief number of nonzero entries in the data */
-  uint64_t num_nonzero;
+  uint64_t num_nonzero_{0};
   /*! \brief label of each instance */
-  std::vector<bst_float> labels;
+  std::vector<bst_float> labels_;
   /*!
    * \brief specified root index of each instance,
    *  can be used for multi task setting
    */
-  std::vector<bst_uint> root_index;
+  std::vector<bst_uint> root_index_;
   /*!
    * \brief the index of begin and end of a group
    *  needed when the learning task is ranking.
    */
-  std::vector<bst_uint> group_ptr;
+  std::vector<bst_uint> group_ptr_;
   /*! \brief weights of each instance, optional */
-  std::vector<bst_float> weights;
+  std::vector<bst_float> weights_;
   /*!
    * \brief initialized margins,
    * 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;
+  std::vector<bst_float> base_margin_;
   /*! \brief version flag, used to check version of this info */
   static const int kVersion = 1;
   /*! \brief default constructor */
-  MetaInfo() : num_row(0), num_col(0), num_nonzero(0) {}
+  MetaInfo()  = default;
   /*!
    * \brief Get weight of each instances.
    * \param i Instance index.
    * \return The weight.
    */
   inline bst_float GetWeight(size_t i) const {
-    return weights.size() != 0 ?  weights[i] : 1.0f;
+    return weights_.size() != 0 ?  weights_[i] : 1.0f;
   }
   /*!
    * \brief Get the root index of i-th instance.
@@ -74,7 +76,20 @@ struct MetaInfo {
    * \return The pre-defined root index of i-th instance.
    */
   inline unsigned GetRoot(size_t i) const {
-    return root_index.size() != 0 ? root_index[i] : 0U;
+    return root_index_.size() != 0 ? root_index_[i] : 0U;
+  }
+  /*! \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()) {
+      return label_order_cache_;
+    }
+    label_order_cache_.resize(labels_.size());
+    std::iota(label_order_cache_.begin(), label_order_cache_.end(), 0);
+    const auto l = labels_;
+    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]);});
+
+    return label_order_cache_;
   }
   /*! \brief clear all the information */
   void Clear();
@@ -96,6 +111,10 @@ struct MetaInfo {
    * \param num Number of elements in the source array.
    */
   void SetInfo(const char* key, const void* dptr, DataType dtype, size_t num);
+
+ private:
+  /*! \brief argsort of labels */
+  mutable std::vector<size_t> label_order_cache_;
 };
 
 /*! \brief read-only sparse instance batch in CSR format */
@@ -107,7 +126,7 @@ struct SparseBatch {
     /*! \brief feature value */
     bst_float fvalue;
     /*! \brief default constructor */
-    Entry() {}
+    Entry() = default;
     /*!
      * \brief constructor with index and value
      * \param index The feature or row index.
@@ -123,11 +142,11 @@ struct SparseBatch {
   /*! \brief an instance of sparse vector in the batch */
   struct Inst {
     /*! \brief pointer to the elements*/
-    const Entry *data;
+    const Entry *data{nullptr};
     /*! \brief length of the instance */
-    bst_uint length;
+    bst_uint length{0};
     /*! \brief constructor */
-    Inst() : data(0), length(0) {}
+    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 {
@@ -149,7 +168,7 @@ struct RowBatch : public SparseBatch {
   const Entry *data_ptr;
   /*! \brief get i-th row from the batch */
   inline Inst operator[](size_t i) const {
-    return Inst(data_ptr + ind_ptr[i], static_cast<bst_uint>(ind_ptr[i + 1] - ind_ptr[i]));
+    return {data_ptr + ind_ptr[i], static_cast<bst_uint>(ind_ptr[i + 1] - ind_ptr[i])};
   }
 };
 
@@ -188,16 +207,16 @@ class DataSource : public dmlc::DataIter<RowBatch> {
  * \brief A vector-like structure to represent set of rows.
  * But saves the memory when all rows are in the set (common case in xgb)
  */
-struct RowSet {
+class RowSet {
  public:
   /*! \return i-th row index */
   inline bst_uint operator[](size_t i) const;
   /*! \return the size of the set. */
-  inline size_t size() const;
+  inline size_t Size() const;
   /*! \brief push the index back to the set */
-  inline void push_back(bst_uint i);
+  inline void PushBack(bst_uint i);
   /*! \brief clear the set */
-  inline void clear();
+  inline void Clear();
   /*!
    * \brief save rowset to file.
    * \param fo The file to be saved.
@@ -210,11 +229,11 @@ struct RowSet {
    */
   inline bool Load(dmlc::Stream* fi);
   /*! \brief constructor */
-  RowSet() : size_(0) {}
+  RowSet()  = default;
 
  private:
   /*! \brief The internal data structure of size */
-  uint64_t size_;
+  uint64_t size_{0};
   /*! \brief The internal data structure of row set if not all*/
   std::vector<bst_uint> rows_;
 };
@@ -232,11 +251,11 @@ struct RowSet {
 class DMatrix {
  public:
   /*! \brief default constructor */
-  DMatrix() : cache_learner_ptr_(nullptr) {}
+  DMatrix()  = default;
   /*! \brief meta information of the dataset */
-  virtual MetaInfo& info() = 0;
+  virtual MetaInfo& Info() = 0;
   /*! \brief meta information of the dataset */
-  virtual const MetaInfo& info() const = 0;
+  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.
@@ -256,14 +275,16 @@ class DMatrix {
    * \param subsample subsample ratio when generating 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(const std::vector<bool>& enabled,
                              float subsample,
-                             size_t max_row_perbatch) = 0;
+                             size_t max_row_perbatch, bool sorted) = 0;
   // the following are column meta data, should be able to answer them fast.
   /*! \return whether column access is enabled */
-  virtual bool HaveColAccess() const = 0;
+  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 */
@@ -271,9 +292,9 @@ class DMatrix {
   /*! \brief get column density */
   virtual float GetColDensity(size_t cidx) const = 0;
   /*! \return reference of buffered rowset, in column access */
-  virtual const RowSet& buffered_rowset() const = 0;
+  virtual const RowSet& BufferedRowset() const = 0;
   /*! \brief virtual destructor */
-  virtual ~DMatrix() {}
+  virtual ~DMatrix() = default;
   /*!
    * \brief Save DMatrix to local file.
    *  The saved file only works for non-sharded dataset(single machine training).
@@ -323,7 +344,7 @@ class DMatrix {
   // allow learner class to access this field.
   friend class LearnerImpl;
   /*! \brief public field to back ref cached matrix. */
-  LearnerImpl* cache_learner_ptr_;
+  LearnerImpl* cache_learner_ptr_{nullptr};
 };
 
 // implementation of inline functions
@@ -331,15 +352,15 @@ inline bst_uint RowSet::operator[](size_t i) const {
   return rows_.size() == 0 ? static_cast<bst_uint>(i) : rows_[i];
 }
 
-inline size_t RowSet::size() const {
+inline size_t RowSet::Size() const {
   return size_;
 }
 
-inline void RowSet::clear() {
+inline void RowSet::Clear() {
   rows_.clear(); size_ = 0;
 }
 
-inline void RowSet::push_back(bst_uint i) {
+inline void RowSet::PushBack(bst_uint i) {
   if (rows_.size() == 0) {
     if (i == size_) {
       ++size_; return;

include/xgboost/feature_map.h

@@ -45,7 +45,7 @@ class FeatureMap {
    */
   inline void PushBack(int fid, const char *fname, const char *ftype) {
     CHECK_EQ(fid, static_cast<int>(names_.size()));
-    names_.push_back(std::string(fname));
+    names_.emplace_back(fname);
     types_.push_back(GetType(ftype));
   }
   /*! \brief clear the feature map */
@@ -54,11 +54,11 @@ class FeatureMap {
     types_.clear();
   }
   /*! \return number of known features */
-  inline size_t size() const {
+  inline size_t Size() const {
     return names_.size();
   }
   /*! \return name of specific feature */
-  inline const char* name(size_t idx) const {
+  inline const char* Name(size_t idx) const {
     CHECK_LT(idx,  names_.size()) << "FeatureMap feature index exceed bound";
     return names_[idx].c_str();
   }
@@ -75,7 +75,7 @@ class FeatureMap {
    * \return The translated type.
    */
   inline static Type GetType(const char* tname) {
-    using namespace std;
+    using std::strcmp;
     if (!strcmp("i", tname)) return kIndicator;
     if (!strcmp("q", tname)) return kQuantitive;
     if (!strcmp("int", tname)) return kInteger;

include/xgboost/gbm.h

@@ -18,6 +18,7 @@
 #include "./data.h"
 #include "./objective.h"
 #include "./feature_map.h"
+#include "../../src/common/host_device_vector.h"
 
 namespace xgboost {
 /*!
@@ -26,7 +27,7 @@ namespace xgboost {
 class GradientBooster {
  public:
   /*! \brief virtual destructor */
-  virtual ~GradientBooster() {}
+  virtual ~GradientBooster() = default;
   /*!
    * \brief set configuration from pair iterators.
    * \param begin The beginning iterator.
@@ -68,8 +69,9 @@ class GradientBooster {
    * the booster may change content of gpair
    */
   virtual void DoBoost(DMatrix* p_fmat,
-                       std::vector<bst_gpair>* in_gpair,
+                       HostDeviceVector<GradientPair>* in_gpair,
                        ObjFunction* obj = nullptr) = 0;
+
   /*!
    * \brief generate predictions for given feature matrix
    * \param dmat feature matrix
@@ -78,8 +80,8 @@ class GradientBooster {
    *    we do not limit number of trees, this parameter is only valid for gbtree, but not for gblinear
    */
   virtual void PredictBatch(DMatrix* dmat,
-                       std::vector<bst_float>* out_preds,
+                            HostDeviceVector<bst_float>* out_preds,
-                       unsigned ntree_limit = 0) = 0;
+                            unsigned ntree_limit = 0) = 0;
   /*!
    * \brief online prediction function, predict score for one instance at a time
    *  NOTE: use the batch prediction interface if possible, batch prediction is usually
@@ -116,10 +118,17 @@ class GradientBooster {
    * \param ntree_limit limit the number of trees used in prediction, when it equals 0, this means
    *    we do not limit number of trees
    * \param approximate use a faster (inconsistent) approximation of SHAP values
+   * \param condition condition on the condition_feature (0=no, -1=cond off, 1=cond on).
+   * \param condition_feature feature to condition on (i.e. fix) during calculations
    */
   virtual void PredictContribution(DMatrix* dmat,
                            std::vector<bst_float>* out_contribs,
-                           unsigned ntree_limit = 0, bool approximate = false) = 0;
+                           unsigned ntree_limit = 0, bool approximate = false,
+                           int condition = 0, unsigned condition_feature = 0) = 0;
+
+  virtual void PredictInteractionContributions(DMatrix* dmat,
+                           std::vector<bst_float>* out_contribs,
+                           unsigned ntree_limit, bool approximate) = 0;
 
   /*!
    * \brief dump the model in the requested format

include/xgboost/learner.h

@@ -37,7 +37,7 @@ namespace xgboost {
 class Learner : public rabit::Serializable {
  public:
   /*! \brief virtual destructor */
-  virtual ~Learner() {}
+  ~Learner() override = default;
   /*!
    * \brief set configuration from pair iterators.
    * \param begin The beginning iterator.
@@ -62,12 +62,12 @@ class Learner : public rabit::Serializable {
    * \brief load model from stream
    * \param fi input stream.
    */
-  virtual void Load(dmlc::Stream* fi) = 0;
+  void Load(dmlc::Stream* fi) override = 0;
   /*!
    * \brief save model to stream.
    * \param fo output stream
    */
-  virtual void Save(dmlc::Stream* fo) const = 0;
+  void Save(dmlc::Stream* fo) const override = 0;
   /*!
    * \brief update the model for one iteration
    *  With the specified objective function.
@@ -84,7 +84,7 @@ class Learner : public rabit::Serializable {
    */
   virtual void BoostOneIter(int iter,
                             DMatrix* train,
-                            std::vector<bst_gpair>* in_gpair) = 0;
+                            HostDeviceVector<GradientPair>* in_gpair) = 0;
   /*!
    * \brief evaluate the model for specific iteration using the configured metrics.
    * \param iter iteration number
@@ -105,14 +105,17 @@ class Learner : public rabit::Serializable {
    * \param pred_leaf whether to only predict the leaf index of each tree in a boosted tree predictor
    * \param pred_contribs whether to only predict the feature contributions
    * \param approx_contribs whether to approximate the feature contributions for speed
+   * \param pred_interactions whether to compute the feature pair contributions
    */
   virtual void Predict(DMatrix* data,
                        bool output_margin,
-                       std::vector<bst_float> *out_preds,
+                       HostDeviceVector<bst_float> *out_preds,
                        unsigned ntree_limit = 0,
                        bool pred_leaf = false,
                        bool pred_contribs = false,
-                       bool approx_contribs = false) const = 0;
+                       bool approx_contribs = false,
+                       bool pred_interactions = false) const = 0;
+
   /*!
    * \brief Set additional attribute to the Booster.
    *  The property will be saved along the booster.
@@ -166,7 +169,7 @@ class Learner : public rabit::Serializable {
    */
   inline void Predict(const SparseBatch::Inst &inst,
                       bool output_margin,
-                      std::vector<bst_float> *out_preds,
+                      HostDeviceVector<bst_float> *out_preds,
                       unsigned ntree_limit = 0) const;
   /*!
    * \brief Create a new instance of learner.
@@ -189,9 +192,9 @@ class Learner : public rabit::Serializable {
 // implementation of inline functions.
 inline void Learner::Predict(const SparseBatch::Inst& inst,
                              bool output_margin,
-                             std::vector<bst_float>* out_preds,
+                             HostDeviceVector<bst_float>* out_preds,
                              unsigned ntree_limit) const {
-  gbm_->PredictInstance(inst, out_preds, ntree_limit);
+  gbm_->PredictInstance(inst, &out_preds->HostVector(), ntree_limit);
   if (!output_margin) {
     obj_->PredTransform(out_preds);
   }

include/xgboost/linear_updater.h

@@ -0,0 +1,67 @@
+/*
+ * Copyright 2018 by Contributors
+ */
+#pragma once
+
+#include <dmlc/registry.h>
+#include <xgboost/base.h>
+#include <xgboost/data.h>
+#include <functional>
+#include <string>
+#include <utility>
+#include <vector>
+#include "../../src/gbm/gblinear_model.h"
+#include "../../src/common/host_device_vector.h"
+
+namespace xgboost {
+/*!
+ * \brief interface of linear updater
+ */
+class LinearUpdater {
+ public:
+  /*! \brief virtual destructor */
+  virtual ~LinearUpdater() = default;
+  /*!
+   * \brief Initialize the updater with given arguments.
+   * \param args arguments to the objective function.
+   */
+  virtual void Init(
+      const std::vector<std::pair<std::string, std::string> >& args) = 0;
+
+  /**
+   * \brief Updates linear model given gradients.
+   *
+   * \param in_gpair            The gradient pair statistics of the data.
+   * \param data                Input data matrix.
+   * \param model               Model to be updated.
+   * \param sum_instance_weight The sum instance weights, used to normalise l1/l2 penalty.
+   */
+
+  virtual void Update(HostDeviceVector<GradientPair>* in_gpair, DMatrix* data,
+                      gbm::GBLinearModel* model,
+                      double sum_instance_weight) = 0;
+
+  /*!
+   * \brief Create a linear updater given name
+   * \param name Name of the linear updater.
+   */
+  static LinearUpdater* Create(const std::string& name);
+};
+
+/*!
+ * \brief Registry entry for linear updater.
+ */
+struct LinearUpdaterReg
+    : public dmlc::FunctionRegEntryBase<LinearUpdaterReg,
+                                        std::function<LinearUpdater*()> > {};
+
+/*!
+ * \brief Macro to register linear updater.
+ */
+#define XGBOOST_REGISTER_LINEAR_UPDATER(UniqueId, Name)                        \
+  static DMLC_ATTRIBUTE_UNUSED ::xgboost::LinearUpdaterReg&                    \
+      __make_##LinearUpdaterReg##_##UniqueId##__ =                             \
+          ::dmlc::Registry< ::xgboost::LinearUpdaterReg>::Get()->__REGISTER__( \
+              Name)
+
+}  // namespace xgboost

include/xgboost/logging.h

@@ -21,7 +21,7 @@ class BaseLogger {
     log_stream_ << "[" << dmlc::DateLogger().HumanDate() << "] ";
 #endif
   }
-  std::ostream& stream() { return log_stream_; }
+  std::ostream& stream() { return log_stream_; }  // NOLINT
 
  protected:
   std::ostringstream log_stream_;

include/xgboost/metric.h

@@ -35,7 +35,7 @@ class Metric {
   /*! \return name of metric */
   virtual const char* Name() const = 0;
   /*! \brief virtual destructor */
-  virtual ~Metric() {}
+  virtual ~Metric() = default;
   /*!
    * \brief create a metric according to name.
    * \param name name of the metric.

include/xgboost/objective.h

@@ -14,13 +14,16 @@
 #include <functional>
 #include "./data.h"
 #include "./base.h"
+#include "../../src/common/host_device_vector.h"
+
 
 namespace xgboost {
+
 /*! \brief interface of objective function */
 class ObjFunction {
  public:
   /*! \brief virtual destructor */
-  virtual ~ObjFunction() {}
+  virtual ~ObjFunction() = default;
   /*!
    * \brief set configuration from pair iterators.
    * \param begin The beginning iterator.
@@ -41,10 +44,11 @@ class ObjFunction {
    * \param iteration current iteration number.
    * \param out_gpair output of get gradient, saves gradient and second order gradient in
    */
-  virtual void GetGradient(const std::vector<bst_float>& preds,
+  virtual void GetGradient(HostDeviceVector<bst_float>* preds,
                            const MetaInfo& info,
                            int iteration,
-                           std::vector<bst_gpair>* out_gpair) = 0;
+                           HostDeviceVector<GradientPair>* out_gpair) = 0;
+
   /*! \return the default evaluation metric for the objective */
   virtual const char* DefaultEvalMetric() const = 0;
   // the following functions are optional, most of time default implementation is good enough
@@ -52,13 +56,14 @@ class ObjFunction {
    * \brief transform prediction values, this is only called when Prediction is called
    * \param io_preds prediction values, saves to this vector as well
    */
-  virtual void PredTransform(std::vector<bst_float> *io_preds) {}
+  virtual void PredTransform(HostDeviceVector<bst_float> *io_preds) {}
+
   /*!
    * \brief transform prediction values, this is only called when Eval is called,
    *  usually it redirect to PredTransform
    * \param io_preds prediction values, saves to this vector as well
    */
-  virtual void EvalTransform(std::vector<bst_float> *io_preds) {
+  virtual void EvalTransform(HostDeviceVector<bst_float> *io_preds) {
     this->PredTransform(io_preds);
   }
   /*!

include/xgboost/predictor.h

@@ -13,6 +13,7 @@
 #include <utility>
 #include <vector>
 #include "../../src/gbm/gbtree_model.h"
+#include "../../src/common/host_device_vector.h"
 
 // Forward declarations
 namespace xgboost {
@@ -35,7 +36,7 @@ namespace xgboost {
 
 class Predictor {
  public:
-  virtual ~Predictor() {}
+  virtual ~Predictor() = default;
 
   /**
    * \fn  virtual void Predictor::Init(const std::vector<std::pair<std::string,
@@ -51,10 +52,6 @@ class Predictor {
                     const std::vector<std::shared_ptr<DMatrix>>& cache);
 
   /**
-   * \fn  virtual void Predictor::PredictBatch( DMatrix* dmat,
-   * std::vector<bst_float>* out_preds, const gbm::GBTreeModel &model, int
-   * tree_begin, unsigned ntree_limit = 0) = 0;
-   *
    * \brief Generate batch predictions for a given feature matrix. May use
    * cached predictions if available instead of calculating from scratch.
    *
@@ -66,7 +63,7 @@ class Predictor {
    * limit trees.
    */
 
-  virtual void PredictBatch(DMatrix* dmat, std::vector<bst_float>* out_preds,
+  virtual void PredictBatch(DMatrix* dmat, HostDeviceVector<bst_float>* out_preds,
                             const gbm::GBTreeModel& model, int tree_begin,
                             unsigned ntree_limit = 0) = 0;
 
@@ -140,14 +137,24 @@ class Predictor {
    * a vector of length (nfeats + 1) * num_output_group * nsample, arranged in
    * that order.
    *
-   * \param [in,out]  dmat          The input feature matrix.
+   * \param [in,out]  dmat               The input feature matrix.
-   * \param [in,out]  out_contribs  The output feature contribs.
+   * \param [in,out]  out_contribs       The output feature contribs.
-   * \param           model         Model to make predictions from.
+   * \param           model              Model to make predictions from.
-   * \param           ntree_limit   (Optional) The ntree limit.
+   * \param           ntree_limit        (Optional) The ntree limit.
-   * \param           approximate   Use fast approximate algorithm.
+   * \param           approximate        Use fast approximate algorithm.
+   * \param           condition          Condition on the condition_feature (0=no, -1=cond off, 1=cond on).
+   * \param           condition_feature  Feature to condition on (i.e. fix) during calculations.
    */
 
   virtual void PredictContribution(DMatrix* dmat,
+                                   std::vector<bst_float>* out_contribs,
+                                   const gbm::GBTreeModel& model,
+                                   unsigned ntree_limit = 0,
+                                   bool approximate = false,
+                                   int condition = 0,
+                                   unsigned condition_feature = 0) = 0;
+
+  virtual void PredictInteractionContributions(DMatrix* dmat,
                                    std::vector<bst_float>* out_contribs,
                                    const gbm::GBTreeModel& model,
                                    unsigned ntree_limit = 0,
@@ -163,41 +170,14 @@ class Predictor {
   static Predictor* Create(std::string name);
 
  protected:
-  /**
-   * \fn  bool PredictFromCache(DMatrix* dmat, std::vector<bst_float>*
-   * out_preds, const gbm::GBTreeModel& model, unsigned ntree_limit = 0)
-   *
-   * \brief Attempt to predict from cache.
-   *
-   * \return  True if it succeeds, false if it fails.
-   */
-  bool PredictFromCache(DMatrix* dmat, std::vector<bst_float>* out_preds,
-                        const gbm::GBTreeModel& model,
-                        unsigned ntree_limit = 0);
-
-  /**
-   * \fn void Predictor::InitOutPredictions(const MetaInfo& info,
-   * std::vector<bst_float>* out_preds, const gbm::GBTreeModel& model) const;
-   *
-   * \brief  Init out predictions according to base margin.
-   *
-   * \param          info      Dmatrix info possibly containing base margin.
-   * \param [in,out] out_preds The out preds.
-   * \param          model     The model.
-   */
-  void InitOutPredictions(const MetaInfo& info,
-                          std::vector<bst_float>* out_preds,
-                          const gbm::GBTreeModel& model) const;
-
   /**
    * \struct  PredictionCacheEntry
    *
    * \brief Contains pointer to input matrix and associated cached predictions.
    */
-
   struct PredictionCacheEntry {
     std::shared_ptr<DMatrix> data;
-    std::vector<bst_float> predictions;
+    HostDeviceVector<bst_float> predictions;
   };
 
   /**

include/xgboost/tree_model.h

@@ -71,70 +71,70 @@ template<typename TSplitCond, typename TNodeStat>
 class TreeModel {
  public:
   /*! \brief data type to indicate split condition */
-  typedef TNodeStat  NodeStat;
+  using NodeStat = TNodeStat;
   /*! \brief auxiliary statistics of node to help tree building */
-  typedef TSplitCond SplitCond;
+  using SplitCond = TSplitCond;
   /*! \brief tree node */
   class Node {
    public:
-    Node() : sindex_(0) {
+    Node()  {
       // assert compact alignment
       static_assert(sizeof(Node) == 4 * sizeof(int) + sizeof(Info),
                     "Node: 64 bit align");
     }
     /*! \brief index of left child */
-    inline int cleft() const {
+    inline int LeftChild() const {
       return this->cleft_;
     }
     /*! \brief index of right child */
-    inline int cright() const {
+    inline int RightChild() const {
       return this->cright_;
     }
     /*! \brief index of default child when feature is missing */
-    inline int cdefault() const {
+    inline int DefaultChild() const {
-      return this->default_left() ? this->cleft() : this->cright();
+      return this->DefaultLeft() ? this->LeftChild() : this->RightChild();
     }
     /*! \brief feature index of split condition */
-    inline unsigned split_index() const {
+    inline unsigned SplitIndex() const {
       return sindex_ & ((1U << 31) - 1U);
     }
     /*! \brief when feature is unknown, whether goes to left child */
-    inline bool default_left() const {
+    inline bool DefaultLeft() const {
       return (sindex_ >> 31) != 0;
     }
     /*! \brief whether current node is leaf node */
-    inline bool is_leaf() const {
+    inline bool IsLeaf() const {
       return cleft_ == -1;
     }
     /*! \return get leaf value of leaf node */
-    inline bst_float leaf_value() const {
+    inline bst_float LeafValue() const {
       return (this->info_).leaf_value;
     }
     /*! \return get split condition of the node */
-    inline TSplitCond split_cond() const {
+    inline TSplitCond SplitCond() const {
       return (this->info_).split_cond;
     }
     /*! \brief get parent of the node */
-    inline int parent() const {
+    inline int Parent() const {
       return parent_ & ((1U << 31) - 1);
     }
     /*! \brief whether current node is left child */
-    inline bool is_left_child() const {
+    inline bool IsLeftChild() const {
       return (parent_ & (1U << 31)) != 0;
     }
     /*! \brief whether this node is deleted */
-    inline bool is_deleted() const {
+    inline bool IsDeleted() const {
       return sindex_ == std::numeric_limits<unsigned>::max();
     }
     /*! \brief whether current node is root */
-    inline bool is_root() const {
+    inline bool IsRoot() const {
       return parent_ == -1;
     }
     /*!
      * \brief set the right child
      * \param nid node id to right child
      */
-    inline void set_right_child(int nid) {
+    inline void SetRightChild(int nid) {
       this->cright_ = nid;
     }
     /*!
@@ -143,7 +143,7 @@ class TreeModel {
      * \param split_cond  split condition
      * \param default_left the default direction when feature is unknown
      */
-    inline void set_split(unsigned split_index, TSplitCond split_cond,
+    inline void SetSplit(unsigned split_index, TSplitCond split_cond,
                           bool default_left = false) {
       if (default_left) split_index |= (1U << 31);
       this->sindex_ = split_index;
@@ -155,13 +155,13 @@ class TreeModel {
      * \param right right index, could be used to store
      *        additional information
      */
-    inline void set_leaf(bst_float value, int right = -1) {
+    inline void SetLeaf(bst_float value, int right = -1) {
       (this->info_).leaf_value = value;
       this->cleft_ = -1;
       this->cright_ = right;
     }
     /*! \brief mark that this node is deleted */
-    inline void mark_delete() {
+    inline void MarkDelete() {
       this->sindex_ = std::numeric_limits<unsigned>::max();
     }
 
@@ -181,11 +181,11 @@ class TreeModel {
     // pointer to left, right
     int cleft_, cright_;
     // split feature index, left split or right split depends on the highest bit
-    unsigned sindex_;
+    unsigned sindex_{0};
     // extra info
     Info info_;
     // set parent
-    inline void set_parent(int pidx, bool is_left_child = true) {
+    inline void SetParent(int pidx, bool is_left_child = true) {
       if (is_left_child) pidx |= (1U << 31);
       this->parent_ = pidx;
     }
@@ -193,35 +193,35 @@ class TreeModel {
 
  protected:
   // vector of nodes
-  std::vector<Node> nodes;
+  std::vector<Node> nodes_;
   // free node space, used during training process
-  std::vector<int>  deleted_nodes;
+  std::vector<int>  deleted_nodes_;
   // stats of nodes
-  std::vector<TNodeStat> stats;
+  std::vector<TNodeStat> stats_;
   // leaf vector, that is used to store additional information
-  std::vector<bst_float> leaf_vector;
+  std::vector<bst_float> leaf_vector_;
   // allocate a new node,
   // !!!!!! NOTE: may cause BUG here, nodes.resize
   inline int AllocNode() {
     if (param.num_deleted != 0) {
-      int nd = deleted_nodes.back();
+      int nd = deleted_nodes_.back();
-      deleted_nodes.pop_back();
+      deleted_nodes_.pop_back();
       --param.num_deleted;
       return nd;
     }
     int nd = param.num_nodes++;
     CHECK_LT(param.num_nodes, std::numeric_limits<int>::max())
         << "number of nodes in the tree exceed 2^31";
-    nodes.resize(param.num_nodes);
+    nodes_.resize(param.num_nodes);
-    stats.resize(param.num_nodes);
+    stats_.resize(param.num_nodes);
-    leaf_vector.resize(param.num_nodes * param.size_leaf_vector);
+    leaf_vector_.resize(param.num_nodes * param.size_leaf_vector);
     return nd;
   }
   // delete a tree node, keep the parent field to allow trace back
   inline void DeleteNode(int nid) {
     CHECK_GE(nid, param.num_roots);
-    deleted_nodes.push_back(nid);
+    deleted_nodes_.push_back(nid);
-    nodes[nid].mark_delete();
+    nodes_[nid].MarkDelete();
     ++param.num_deleted;
   }
 
@@ -232,11 +232,11 @@ class TreeModel {
    * \param value new leaf value
    */
   inline void ChangeToLeaf(int rid, bst_float value) {
-    CHECK(nodes[nodes[rid].cleft() ].is_leaf());
+    CHECK(nodes_[nodes_[rid].LeftChild() ].IsLeaf());
-    CHECK(nodes[nodes[rid].cright()].is_leaf());
+    CHECK(nodes_[nodes_[rid].RightChild()].IsLeaf());
-    this->DeleteNode(nodes[rid].cleft());
+    this->DeleteNode(nodes_[rid].LeftChild());
-    this->DeleteNode(nodes[rid].cright());
+    this->DeleteNode(nodes_[rid].RightChild());
-    nodes[rid].set_leaf(value);
+    nodes_[rid].SetLeaf(value);
   }
   /*!
    * \brief collapse a non leaf node to a leaf node, delete its children
@@ -244,12 +244,12 @@ class TreeModel {
    * \param value new leaf value
    */
   inline void CollapseToLeaf(int rid, bst_float value) {
-    if (nodes[rid].is_leaf()) return;
+    if (nodes_[rid].IsLeaf()) return;
-    if (!nodes[nodes[rid].cleft() ].is_leaf()) {
+    if (!nodes_[nodes_[rid].LeftChild() ].IsLeaf()) {
-      CollapseToLeaf(nodes[rid].cleft(), 0.0f);
+      CollapseToLeaf(nodes_[rid].LeftChild(), 0.0f);
     }
-    if (!nodes[nodes[rid].cright() ].is_leaf()) {
+    if (!nodes_[nodes_[rid].RightChild() ].IsLeaf()) {
-      CollapseToLeaf(nodes[rid].cright(), 0.0f);
+      CollapseToLeaf(nodes_[rid].RightChild(), 0.0f);
     }
     this->ChangeToLeaf(rid, value);
   }
@@ -262,47 +262,47 @@ class TreeModel {
     param.num_nodes = 1;
     param.num_roots = 1;
     param.num_deleted = 0;
-    nodes.resize(1);
+    nodes_.resize(1);
   }
   /*! \brief get node given nid */
   inline Node& operator[](int nid) {
-    return nodes[nid];
+    return nodes_[nid];
   }
   /*! \brief get node given nid */
   inline const Node& operator[](int nid) const {
-    return nodes[nid];
+    return nodes_[nid];
   }
 
   /*! \brief get const reference to nodes */
-  inline const std::vector<Node>& GetNodes() const { return nodes; }
+  inline const std::vector<Node>& GetNodes() const { return nodes_; }
 
   /*! \brief get node statistics given nid */
-  inline NodeStat& stat(int nid) {
+  inline NodeStat& Stat(int nid) {
-    return stats[nid];
+    return stats_[nid];
   }
   /*! \brief get node statistics given nid */
-  inline const NodeStat& stat(int nid) const {
+  inline const NodeStat& Stat(int nid) const {
-    return stats[nid];
+    return stats_[nid];
   }
   /*! \brief get leaf vector given nid */
-  inline bst_float* leafvec(int nid) {
+  inline bst_float* Leafvec(int nid) {
-    if (leaf_vector.size() == 0) return nullptr;
+    if (leaf_vector_.size() == 0) return nullptr;
-    return& leaf_vector[nid * param.size_leaf_vector];
+    return& leaf_vector_[nid * param.size_leaf_vector];
   }
   /*! \brief get leaf vector given nid */
-  inline const bst_float* leafvec(int nid) const {
+  inline const bst_float* Leafvec(int nid) const {
-    if (leaf_vector.size() == 0) return nullptr;
+    if (leaf_vector_.size() == 0) return nullptr;
-    return& leaf_vector[nid * param.size_leaf_vector];
+    return& leaf_vector_[nid * param.size_leaf_vector];
   }
   /*! \brief initialize the model */
   inline void InitModel() {
     param.num_nodes = param.num_roots;
-    nodes.resize(param.num_nodes);
+    nodes_.resize(param.num_nodes);
-    stats.resize(param.num_nodes);
+    stats_.resize(param.num_nodes);
-    leaf_vector.resize(param.num_nodes * param.size_leaf_vector, 0.0f);
+    leaf_vector_.resize(param.num_nodes * param.size_leaf_vector, 0.0f);
     for (int i = 0; i < param.num_nodes; i ++) {
-      nodes[i].set_leaf(0.0f);
+      nodes_[i].SetLeaf(0.0f);
-      nodes[i].set_parent(-1);
+      nodes_[i].SetParent(-1);
     }
   }
   /*!
@@ -311,35 +311,35 @@ class TreeModel {
    */
   inline void Load(dmlc::Stream* fi) {
     CHECK_EQ(fi->Read(&param, sizeof(TreeParam)), sizeof(TreeParam));
-    nodes.resize(param.num_nodes);
+    nodes_.resize(param.num_nodes);
-    stats.resize(param.num_nodes);
+    stats_.resize(param.num_nodes);
     CHECK_NE(param.num_nodes, 0);
-    CHECK_EQ(fi->Read(dmlc::BeginPtr(nodes), sizeof(Node) * nodes.size()),
+    CHECK_EQ(fi->Read(dmlc::BeginPtr(nodes_), sizeof(Node) * nodes_.size()),
-             sizeof(Node) * nodes.size());
+             sizeof(Node) * nodes_.size());
-    CHECK_EQ(fi->Read(dmlc::BeginPtr(stats), sizeof(NodeStat) * stats.size()),
+    CHECK_EQ(fi->Read(dmlc::BeginPtr(stats_), sizeof(NodeStat) * stats_.size()),
-             sizeof(NodeStat) * stats.size());
+             sizeof(NodeStat) * stats_.size());
     if (param.size_leaf_vector != 0) {
-      CHECK(fi->Read(&leaf_vector));
+      CHECK(fi->Read(&leaf_vector_));
     }
     // chg deleted nodes
-    deleted_nodes.resize(0);
+    deleted_nodes_.resize(0);
     for (int i = param.num_roots; i < param.num_nodes; ++i) {
-      if (nodes[i].is_deleted()) deleted_nodes.push_back(i);
+      if (nodes_[i].IsDeleted()) deleted_nodes_.push_back(i);
     }
-    CHECK_EQ(static_cast<int>(deleted_nodes.size()), param.num_deleted);
+    CHECK_EQ(static_cast<int>(deleted_nodes_.size()), param.num_deleted);
   }
   /*!
    * \brief save model to stream
    * \param fo output stream
    */
   inline void Save(dmlc::Stream* fo) const {
-    CHECK_EQ(param.num_nodes, static_cast<int>(nodes.size()));
+    CHECK_EQ(param.num_nodes, static_cast<int>(nodes_.size()));
-    CHECK_EQ(param.num_nodes, static_cast<int>(stats.size()));
+    CHECK_EQ(param.num_nodes, static_cast<int>(stats_.size()));
     fo->Write(&param, sizeof(TreeParam));
     CHECK_NE(param.num_nodes, 0);
-    fo->Write(dmlc::BeginPtr(nodes), sizeof(Node) * nodes.size());
+    fo->Write(dmlc::BeginPtr(nodes_), sizeof(Node) * nodes_.size());
-    fo->Write(dmlc::BeginPtr(stats), sizeof(NodeStat) * nodes.size());
+    fo->Write(dmlc::BeginPtr(stats_), sizeof(NodeStat) * nodes_.size());
-    if (param.size_leaf_vector != 0) fo->Write(leaf_vector);
+    if (param.size_leaf_vector != 0) fo->Write(leaf_vector_);
   }
   /*!
    * \brief add child nodes to node
@@ -348,10 +348,10 @@ class TreeModel {
   inline void AddChilds(int nid) {
     int pleft  = this->AllocNode();
     int pright = this->AllocNode();
-    nodes[nid].cleft_  = pleft;
+    nodes_[nid].cleft_  = pleft;
-    nodes[nid].cright_ = pright;
+    nodes_[nid].cright_ = pright;
-    nodes[nodes[nid].cleft() ].set_parent(nid, true);
+    nodes_[nodes_[nid].LeftChild() ].SetParent(nid, true);
-    nodes[nodes[nid].cright()].set_parent(nid, false);
+    nodes_[nodes_[nid].RightChild()].SetParent(nid, false);
   }
   /*!
    * \brief only add a right child to a leaf node
@@ -359,8 +359,8 @@ class TreeModel {
    */
   inline void AddRightChild(int nid) {
     int pright = this->AllocNode();
-    nodes[nid].right  = pright;
+    nodes_[nid].right  = pright;
-    nodes[nodes[nid].right].set_parent(nid, false);
+    nodes_[nodes_[nid].right].SetParent(nid, false);
   }
   /*!
    * \brief get current depth
@@ -369,9 +369,9 @@ class TreeModel {
    */
   inline int GetDepth(int nid, bool pass_rchild = false) const {
     int depth = 0;
-    while (!nodes[nid].is_root()) {
+    while (!nodes_[nid].IsRoot()) {
-      if (!pass_rchild || nodes[nid].is_left_child()) ++depth;
+      if (!pass_rchild || nodes_[nid].IsLeftChild()) ++depth;
-      nid = nodes[nid].parent();
+      nid = nodes_[nid].Parent();
     }
     return depth;
   }
@@ -380,9 +380,9 @@ class TreeModel {
    * \param nid node id
    */
   inline int MaxDepth(int nid) const {
-    if (nodes[nid].is_leaf()) return 0;
+    if (nodes_[nid].IsLeaf()) return 0;
-    return std::max(MaxDepth(nodes[nid].cleft())+1,
+    return std::max(MaxDepth(nodes_[nid].LeftChild())+1,
-                     MaxDepth(nodes[nid].cright())+1);
+                     MaxDepth(nodes_[nid].RightChild())+1);
   }
   /*!
    * \brief get maximum depth
@@ -395,7 +395,7 @@ class TreeModel {
     return maxd;
   }
   /*! \brief number of extra nodes besides the root */
-  inline int num_extra_nodes() const {
+  inline int NumExtraNodes() const {
     return param.num_nodes - param.num_roots - param.num_deleted;
   }
 };
@@ -421,7 +421,7 @@ struct PathElement {
   bst_float zero_fraction;
   bst_float one_fraction;
   bst_float pweight;
-  PathElement() {}
+  PathElement() = default;
   PathElement(int i, bst_float z, bst_float o, bst_float w) :
     feature_index(i), zero_fraction(z), one_fraction(o), pweight(w) {}
 };
@@ -457,19 +457,19 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat> {
      * \brief returns the size of the feature vector
      * \return the size of the feature vector
      */
-    inline size_t size() const;
+    inline size_t Size() const;
     /*!
      * \brief get ith value
      * \param i feature index.
      * \return the i-th feature value
      */
-    inline bst_float fvalue(size_t i) const;
+    inline bst_float Fvalue(size_t i) const;
     /*!
      * \brief check whether i-th entry is missing
      * \param i feature index.
      * \return whether i-th value is missing.
      */
-    inline bool is_missing(size_t i) const;
+    inline bool IsMissing(size_t i) const;
 
    private:
     /*!
@@ -480,7 +480,7 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat> {
       bst_float fvalue;
       int flag;
     };
-    std::vector<Entry> data;
+    std::vector<Entry> data_;
   };
   /*!
    * \brief get the leaf index
@@ -501,13 +501,33 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat> {
    * \param feat dense feature vector, if the feature is missing the field is set to NaN
    * \param root_id starting root index of the instance
    * \param out_contribs output vector to hold the contributions
+   * \param condition fix one feature to either off (-1) on (1) or not fixed (0 default)
+   * \param condition_feature the index of the feature to fix
    */
   inline void CalculateContributions(const RegTree::FVec& feat, unsigned root_id,
-                                     bst_float *out_contribs) const;
+                                     bst_float *out_contribs,
+                                     int condition = 0,
+                                     unsigned condition_feature = 0) const;
+  /*!
+   * \brief Recursive function that computes the feature attributions for a single tree.
+   * \param feat dense feature vector, if the feature is missing the field is set to NaN
+   * \param phi dense output vector of feature attributions
+   * \param node_index the index of the current node in the tree
+   * \param unique_depth how many unique features are above the current node in the tree
+   * \param parent_unique_path a vector of statistics about our current path through the tree
+   * \param parent_zero_fraction what fraction of the parent path weight is coming as 0 (integrated)
+   * \param parent_one_fraction what fraction of the parent path weight is coming as 1 (fixed)
+   * \param parent_feature_index what feature the parent node used to split
+   * \param condition fix one feature to either off (-1) on (1) or not fixed (0 default)
+   * \param condition_feature the index of the feature to fix
+   * \param condition_fraction what fraction of the current weight matches our conditioning feature
+   */
   inline void TreeShap(const RegTree::FVec& feat, bst_float *phi,
                        unsigned node_index, unsigned unique_depth,
                        PathElement *parent_unique_path, bst_float parent_zero_fraction,
-                       bst_float parent_one_fraction, int parent_feature_index) const;
+                       bst_float parent_one_fraction, int parent_feature_index,
+                       int condition, unsigned condition_feature,
+                       bst_float condition_fraction) const;
 
   /*!
    * \brief calculate the approximate feature contributions for the given root
@@ -542,63 +562,63 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat> {
  private:
   inline bst_float FillNodeMeanValue(int nid);
 
-  std::vector<bst_float> node_mean_values;
+  std::vector<bst_float> node_mean_values_;
 };
 
 // implementations of inline functions
 // do not need to read if only use the model
 inline void RegTree::FVec::Init(size_t size) {
   Entry e; e.flag = -1;
-  data.resize(size);
+  data_.resize(size);
-  std::fill(data.begin(), data.end(), e);
+  std::fill(data_.begin(), data_.end(), e);
 }
 
 inline void RegTree::FVec::Fill(const RowBatch::Inst& inst) {
   for (bst_uint i = 0; i < inst.length; ++i) {
-    if (inst[i].index >= data.size()) continue;
+    if (inst[i].index >= data_.size()) continue;
-    data[inst[i].index].fvalue = inst[i].fvalue;
+    data_[inst[i].index].fvalue = inst[i].fvalue;
   }
 }
 
 inline void RegTree::FVec::Drop(const RowBatch::Inst& inst) {
   for (bst_uint i = 0; i < inst.length; ++i) {
-    if (inst[i].index >= data.size()) continue;
+    if (inst[i].index >= data_.size()) continue;
-    data[inst[i].index].flag = -1;
+    data_[inst[i].index].flag = -1;
   }
 }
 
-inline size_t RegTree::FVec::size() const {
+inline size_t RegTree::FVec::Size() const {
-  return data.size();
+  return data_.size();
 }
 
-inline bst_float RegTree::FVec::fvalue(size_t i) const {
+inline bst_float RegTree::FVec::Fvalue(size_t i) const {
-  return data[i].fvalue;
+  return data_[i].fvalue;
 }
 
-inline bool RegTree::FVec::is_missing(size_t i) const {
+inline bool RegTree::FVec::IsMissing(size_t i) const {
-  return data[i].flag == -1;
+  return data_[i].flag == -1;
 }
 
 inline int RegTree::GetLeafIndex(const RegTree::FVec& feat, unsigned root_id) const {
-  int pid = static_cast<int>(root_id);
+  auto pid = static_cast<int>(root_id);
-  while (!(*this)[pid].is_leaf()) {
+  while (!(*this)[pid].IsLeaf()) {
-    unsigned split_index = (*this)[pid].split_index();
+    unsigned split_index = (*this)[pid].SplitIndex();
-    pid = this->GetNext(pid, feat.fvalue(split_index), feat.is_missing(split_index));
+    pid = this->GetNext(pid, feat.Fvalue(split_index), feat.IsMissing(split_index));
   }
   return pid;
 }
 
 inline bst_float RegTree::Predict(const RegTree::FVec& feat, unsigned root_id) const {
   int pid = this->GetLeafIndex(feat, root_id);
-  return (*this)[pid].leaf_value();
+  return (*this)[pid].LeafValue();
 }
 
 inline void RegTree::FillNodeMeanValues() {
   size_t num_nodes = this->param.num_nodes;
-  if (this->node_mean_values.size() == num_nodes) {
+  if (this->node_mean_values_.size() == num_nodes) {
     return;
   }
-  this->node_mean_values.resize(num_nodes);
+  this->node_mean_values_.resize(num_nodes);
   for (int root_id = 0; root_id < param.num_roots; ++root_id) {
     this->FillNodeMeanValue(root_id);
   }
@@ -607,40 +627,39 @@ inline void RegTree::FillNodeMeanValues() {
 inline bst_float RegTree::FillNodeMeanValue(int nid) {
   bst_float result;
   auto& node = (*this)[nid];
-  if (node.is_leaf()) {
+  if (node.IsLeaf()) {
-    result = node.leaf_value();
+    result = node.LeafValue();
   } else {
-    result  = this->FillNodeMeanValue(node.cleft()) * this->stat(node.cleft()).sum_hess;
+    result  = this->FillNodeMeanValue(node.LeftChild()) * this->Stat(node.LeftChild()).sum_hess;
-    result += this->FillNodeMeanValue(node.cright()) * this->stat(node.cright()).sum_hess;
+    result += this->FillNodeMeanValue(node.RightChild()) * this->Stat(node.RightChild()).sum_hess;
-    result /= this->stat(nid).sum_hess;
+    result /= this->Stat(nid).sum_hess;
   }
-  this->node_mean_values[nid] = result;
+  this->node_mean_values_[nid] = result;
   return result;
 }
 
 inline void RegTree::CalculateContributionsApprox(const RegTree::FVec& feat, unsigned root_id,
                                                   bst_float *out_contribs) const {
-  CHECK_GT(this->node_mean_values.size(), 0U);
+  CHECK_GT(this->node_mean_values_.size(), 0U);
   // this follows the idea of http://blog.datadive.net/interpreting-random-forests/
-  bst_float node_value;
+  unsigned split_index = 0;
-  unsigned split_index;
+  auto pid = static_cast<int>(root_id);
-  int pid = static_cast<int>(root_id);
   // update bias value
-  node_value = this->node_mean_values[pid];
+  bst_float node_value = this->node_mean_values_[pid];
-  out_contribs[feat.size()] += node_value;
+  out_contribs[feat.Size()] += node_value;
-  if ((*this)[pid].is_leaf()) {
+  if ((*this)[pid].IsLeaf()) {
     // nothing to do anymore
     return;
   }
-  while (!(*this)[pid].is_leaf()) {
+  while (!(*this)[pid].IsLeaf()) {
-    split_index = (*this)[pid].split_index();
+    split_index = (*this)[pid].SplitIndex();
-    pid = this->GetNext(pid, feat.fvalue(split_index), feat.is_missing(split_index));
+    pid = this->GetNext(pid, feat.Fvalue(split_index), feat.IsMissing(split_index));
-    bst_float new_value = this->node_mean_values[pid];
+    bst_float new_value = this->node_mean_values_[pid];
     // update feature weight
     out_contribs[split_index] += new_value - node_value;
     node_value = new_value;
   }
-  bst_float leaf_value = (*this)[pid].leaf_value();
+  bst_float leaf_value = (*this)[pid].LeafValue();
   // update leaf feature weight
   out_contribs[split_index] += leaf_value - node_value;
 }
@@ -700,7 +719,7 @@ inline bst_float UnwoundPathSum(const PathElement *unique_path, unsigned unique_
                             / static_cast<bst_float>((i + 1) * one_fraction);
       total += tmp;
       next_one_portion = unique_path[i].pweight - tmp * zero_fraction * ((unique_depth - i)
-                         / static_cast<bst_float>(unique_depth+1));
+                         / static_cast<bst_float>(unique_depth + 1));
     } else {
       total += (unique_path[i].pweight / zero_fraction) / ((unique_depth - i)
                / static_cast<bst_float>(unique_depth + 1));
@@ -713,41 +732,49 @@ inline bst_float UnwoundPathSum(const PathElement *unique_path, unsigned unique_
 inline void RegTree::TreeShap(const RegTree::FVec& feat, bst_float *phi,
                               unsigned node_index, unsigned unique_depth,
                               PathElement *parent_unique_path, bst_float parent_zero_fraction,
-                              bst_float parent_one_fraction, int parent_feature_index) const {
+                              bst_float parent_one_fraction, int parent_feature_index,
+                              int condition, unsigned condition_feature,
+                              bst_float condition_fraction) const {
   const auto node = (*this)[node_index];
 
+  // stop if we have no weight coming down to us
+  if (condition_fraction == 0) return;
+
   // extend the unique path
-  PathElement *unique_path = parent_unique_path + unique_depth;
+  PathElement *unique_path = parent_unique_path + unique_depth + 1;
-  if (unique_depth > 0) std::copy(parent_unique_path,
+  std::copy(parent_unique_path, parent_unique_path + unique_depth + 1, unique_path);
-                                  parent_unique_path + unique_depth, unique_path);
+
-  ExtendPath(unique_path, unique_depth, parent_zero_fraction,
+  if (condition == 0 || condition_feature != static_cast<unsigned>(parent_feature_index)) {
-             parent_one_fraction, parent_feature_index);
+    ExtendPath(unique_path, unique_depth, parent_zero_fraction,
-  const unsigned split_index = node.split_index();
+               parent_one_fraction, parent_feature_index);
+  }
+  const unsigned split_index = node.SplitIndex();
 
   // leaf node
-  if (node.is_leaf()) {
+  if (node.IsLeaf()) {
     for (unsigned i = 1; i <= unique_depth; ++i) {
       const bst_float w = UnwoundPathSum(unique_path, unique_depth, i);
       const PathElement &el = unique_path[i];
-      phi[el.feature_index] += w * (el.one_fraction - el.zero_fraction) * node.leaf_value();
+      phi[el.feature_index] += w * (el.one_fraction - el.zero_fraction)
+                                 * node.LeafValue() * condition_fraction;
     }
 
   // internal node
   } else {
     // find which branch is "hot" (meaning x would follow it)
     unsigned hot_index = 0;
-    if (feat.is_missing(split_index)) {
+    if (feat.IsMissing(split_index)) {
-      hot_index = node.cdefault();
+      hot_index = node.DefaultChild();
-    } else if (feat.fvalue(split_index) < node.split_cond()) {
+    } else if (feat.Fvalue(split_index) < node.SplitCond()) {
-      hot_index = node.cleft();
+      hot_index = node.LeftChild();
     } else {
-      hot_index = node.cright();
+      hot_index = node.RightChild();
     }
-    const unsigned cold_index = (static_cast<int>(hot_index) == node.cleft() ?
+    const unsigned cold_index = (static_cast<int>(hot_index) == node.LeftChild() ?
-                                 node.cright() : node.cleft());
+                                 node.RightChild() : node.LeftChild());
-    const bst_float w = this->stat(node_index).sum_hess;
+    const bst_float w = this->Stat(node_index).sum_hess;
-    const bst_float hot_zero_fraction = this->stat(hot_index).sum_hess / w;
+    const bst_float hot_zero_fraction = this->Stat(hot_index).sum_hess / w;
-    const bst_float cold_zero_fraction = this->stat(cold_index).sum_hess / w;
+    const bst_float cold_zero_fraction = this->Stat(cold_index).sum_hess / w;
     bst_float incoming_zero_fraction = 1;
     bst_float incoming_one_fraction = 1;
 
@@ -764,47 +791,57 @@ inline void RegTree::TreeShap(const RegTree::FVec& feat, bst_float *phi,
       unique_depth -= 1;
     }
 
+    // divide up the condition_fraction among the recursive calls
+    bst_float hot_condition_fraction = condition_fraction;
+    bst_float cold_condition_fraction = condition_fraction;
+    if (condition > 0 && split_index == condition_feature) {
+      cold_condition_fraction = 0;
+      unique_depth -= 1;
+    } else if (condition < 0 && split_index == condition_feature) {
+      hot_condition_fraction *= hot_zero_fraction;
+      cold_condition_fraction *= cold_zero_fraction;
+      unique_depth -= 1;
+    }
+
     TreeShap(feat, phi, hot_index, unique_depth + 1, unique_path,
-             hot_zero_fraction*incoming_zero_fraction, incoming_one_fraction, split_index);
+             hot_zero_fraction * incoming_zero_fraction, incoming_one_fraction,
+             split_index, condition, condition_feature, hot_condition_fraction);
 
     TreeShap(feat, phi, cold_index, unique_depth + 1, unique_path,
-             cold_zero_fraction*incoming_zero_fraction, 0, split_index);
+             cold_zero_fraction * incoming_zero_fraction, 0,
+             split_index, condition, condition_feature, cold_condition_fraction);
   }
 }
 
 inline void RegTree::CalculateContributions(const RegTree::FVec& feat, unsigned root_id,
-                                            bst_float *out_contribs) const {
+                                            bst_float *out_contribs,
+                                            int condition,
+                                            unsigned condition_feature) const {
   // find the expected value of the tree's predictions
-  bst_float base_value = 0.0f;
+  if (condition == 0) {
-  bst_float total_cover = 0.0f;
+    bst_float node_value = this->node_mean_values_[static_cast<int>(root_id)];
-  for (int i = 0; i < (*this).param.num_nodes; ++i) {
+    out_contribs[feat.Size()] += node_value;
-    const auto node = (*this)[i];
-    if (node.is_leaf()) {
-      const auto cover = this->stat(i).sum_hess;
-      base_value += cover * node.leaf_value();
-      total_cover += cover;
-    }
   }
-  out_contribs[feat.size()] += base_value / total_cover;
 
   // Preallocate space for the unique path data
-  const int maxd = this->MaxDepth(root_id) + 1;
+  const int maxd = this->MaxDepth(root_id) + 2;
-  PathElement *unique_path_data = new PathElement[(maxd * (maxd + 1)) / 2];
+  auto *unique_path_data = new PathElement[(maxd * (maxd + 1)) / 2];
 
-  TreeShap(feat, out_contribs, root_id, 0, unique_path_data, 1, 1, -1);
+  TreeShap(feat, out_contribs, root_id, 0, unique_path_data,
+           1, 1, -1, condition, condition_feature, 1);
   delete[] unique_path_data;
 }
 
 /*! \brief get next position of the tree given current pid */
 inline int RegTree::GetNext(int pid, bst_float fvalue, bool is_unknown) const {
-  bst_float split_value = (*this)[pid].split_cond();
+  bst_float split_value = (*this)[pid].SplitCond();
   if (is_unknown) {
-    return (*this)[pid].cdefault();
+    return (*this)[pid].DefaultChild();
   } else {
     if (fvalue < split_value) {
-      return (*this)[pid].cleft();
+      return (*this)[pid].LeftChild();
     } else {
-      return (*this)[pid].cright();
+      return (*this)[pid].RightChild();
     }
   }
 }

include/xgboost/tree_updater.h

@@ -16,6 +16,7 @@
 #include "./base.h"
 #include "./data.h"
 #include "./tree_model.h"
+#include "../../src/common/host_device_vector.h"
 
 namespace xgboost {
 /*!
@@ -24,7 +25,7 @@ namespace xgboost {
 class TreeUpdater {
  public:
   /*! \brief virtual destructor */
-  virtual ~TreeUpdater() {}
+  virtual ~TreeUpdater() = default;
   /*!
    * \brief Initialize the updater with given arguments.
    * \param args arguments to the objective function.
@@ -39,7 +40,7 @@ class TreeUpdater {
    *         but maybe different random seeds, usually one tree is passed in at a time,
    *         there can be multiple trees when we train random forest style model
    */
-  virtual void Update(const std::vector<bst_gpair>& gpair,
+  virtual void Update(HostDeviceVector<GradientPair>* gpair,
                       DMatrix* data,
                       const std::vector<RegTree*>& trees) = 0;
 
@@ -54,9 +55,10 @@ class TreeUpdater {
    *         updated by the time this function returns.
    */
   virtual bool UpdatePredictionCache(const DMatrix* data,
-                                     std::vector<bst_float>* out_preds) {
+                                     HostDeviceVector<bst_float>* out_preds) {
     return false;
   }
+
   /*!
    * \brief Create a tree updater given name
    * \param name Name of the tree updater.

jvm-packages/README.md

@@ -16,7 +16,58 @@ Apache Flink and Apache Spark.
 
 You can find more about XGBoost on [Documentation](https://xgboost.readthedocs.org/en/latest/jvm/index.html) and [Resource Page](../demo/README.md).
 
+## Add Maven Dependency
+
+XGBoost4J, XGBoost4J-Spark, etc. in maven repository is compiled with g++-4.8.5  
+
+### Access SNAPSHOT version
+
+You need to add github as repo:
+
+<b>maven</b>:
+
+```xml
+<repository>
+  <id>GitHub Repo</id>
+  <name>GitHub Repo</name>
+  <url>https://raw.githubusercontent.com/CodingCat/xgboost/maven-repo/</url>
+</repository>
+```
+
+<b>sbt</b>:
+ 
+```sbt 
+resolvers += "GitHub Repo" at "https://raw.githubusercontent.com/CodingCat/xgboost/maven-repo/"
+```
+
+the add dependency as following:
+
+<b>maven</b> 
+
+```
+<dependency>
+    <groupId>ml.dmlc</groupId>
+    <artifactId>xgboost4j</artifactId>
+    <version>latest_version_num</version>
+</dependency>
+``` 
+ 
+<b>sbt</b> 
+```sbt
+ "ml.dmlc" % "xgboost4j" % "latest_version_num"
+``` 
+
+if you want to use `xgboost4j-spark`, you just need to replace xgboost4j with `xgboost4j-spark`
+
 ## Examples
 
 Full code examples for Scala, Java, Apache Spark, and Apache Flink can
 be found in the [examples package](https://github.com/dmlc/xgboost/tree/master/jvm-packages/xgboost4j-example).
+
+**NOTE on LIBSVM Format**: 
+
+* Use *1-based* ascending indexes for the LIBSVM format in distributed training mode
+
+    * Spark does the internal conversion, and does not accept formats that are 0-based
+
+* Whereas, use *0-based* indexes format when predicting in normal mode - for instance, while using the saved model in the Python package

jvm-packages/dev/build-docker.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+set -x
+
+sudo docker run --rm -m 4g -e JAVA_OPTS='-Xmx6g' --attach stdin --attach stdout --attach stderr --volume `pwd`/../:/xgboost codingcat/xgbrelease:latest /xgboost/jvm-packages/dev/build.sh

jvm-packages/dev/build.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+
+set -x
+
+export JAVA_HOME=/usr/lib/jvm/java-1.8.0
+export MAVEN_OPTS="-Xmx3000m"
+export CMAKE_CXX_COMPILER=/opt/rh/devtoolset-2/root/usr/bin/gcc
+export CXX=/opt/rh/devtoolset-2/root/usr/bin/g++
+export CC=/opt/rh/devtoolset-2/root/usr/bin/gcc
+
+export PATH=$CXX:$CC:/opt/rh/python27/root/usr/bin/python:$PATH
+
+scl enable devtoolset-2 bash
+scl enable python27 bash
+
+rm /usr/bin/python
+ln -s /opt/rh/python27/root/usr/bin/python /usr/bin/python
+
+# build xgboost
+cd /xgboost/jvm-packages;mvn package
+

jvm-packages/dev/change_version.sh

@@ -0,0 +1,44 @@
+#!/bin/sh
+
+#!/bin/bash
+
+#
+# Licensed to the Apache Software Foundation (ASF) under one or more
+# contributor license agreements.  See the NOTICE file distributed with
+# this work for additional information regarding copyright ownership.
+# The ASF licenses this file to You 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.
+#
+# (Yizhi) This is mainly inspired by the script in apache/spark.
+# I did some modificaiton to get it with our project.
+# (Nan) Modified from MxNet
+
+set -e
+
+if [[ ($# -ne 2) || ( $1 == "--help") ||  $1 == "-h" ]]; then
+  echo "Usage: $(basename $0) [-h|--help] <from_version> <to_version>" 1>&2
+  exit 1
+fi
+
+FROM_VERSION=$1
+TO_VERSION=$2
+
+sed_i() {
+  perl -p -000 -e "$1" "$2" > "$2.tmp" && mv "$2.tmp" "$2"
+}
+   
+export -f sed_i
+ 
+BASEDIR=$(dirname $0)/..
+find "$BASEDIR" -name 'pom.xml' -not -path '*target*' -print \
+  -exec bash -c \
+  "sed_i 's/(<artifactId>(xgboost-jvm|xgboost4j.*)<\/artifactId>\s+<version)>'$FROM_VERSION'(<\/version>)/\1>'$TO_VERSION'\3/g' {}" \;

jvm-packages/pom.xml

@@ -6,16 +6,15 @@
 
     <groupId>ml.dmlc</groupId>
     <artifactId>xgboost-jvm</artifactId>
-    <version>0.7</version>
+    <version>0.72-SNAPSHOT</version>
     <packaging>pom</packaging>
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <project.reporting.outputEncoding>UTF-8</project.reporting.outputEncoding>
         <maven.compiler.source>1.7</maven.compiler.source>
         <maven.compiler.target>1.7</maven.compiler.target>
-        <maven.version>3.3.9</maven.version>
         <flink.version>0.10.2</flink.version>
-        <spark.version>2.1.0</spark.version>
+        <spark.version>2.3.0</spark.version>
         <scala.version>2.11.8</scala.version>
         <scala.binary.version>2.11</scala.binary.version>
     </properties>
@@ -34,16 +33,82 @@
     </modules>
     <profiles>
         <profile>
-            <id>spark-2.x</id>
+            <id>assembly</id>
-            <activation>
+            <build>
-                <activeByDefault>true</activeByDefault>
+                <plugins>
-            </activation>
+                    <plugin>
-            <!--<properties>-->
+                        <groupId>org.apache.maven.plugins</groupId>
-            <!--<flink.version>0.10.2</flink.version> -->
+                        <artifactId>maven-assembly-plugin</artifactId>
-            <!--<spark.version>2.0.1</spark.version>-->
+                        <version>2.6</version>
-            <!--<scala.version>2.11.8</scala.version>-->
+                        <configuration>
-            <!--<scala.binary.version>2.11</scala.binary.version>-->
+                            <descriptorRefs>
-            <!--</properties>-->
+                                <descriptorRef>jar-with-dependencies</descriptorRef>
+                            </descriptorRefs>
+                            <skipAssembly>true</skipAssembly>
+                        </configuration>
+                        <executions>
+                            <execution>
+                                <id>make-assembly</id>
+                                <phase>package</phase>
+                                <goals>
+                                    <goal>single</goal>
+                                </goals>
+                            </execution>
+                        </executions>
+                    </plugin>
+                </plugins>
+            </build>
+        </profile>
+        <profile>
+            <id>release-to-github</id>
+            <distributionManagement>
+                <repository>
+                    <id>github.repo</id>
+                    <name>Temporary Staging Repository</name>
+                    <url>file://${project.build.directory}/mvn-repo</url>
+                </repository>
+            </distributionManagement>
+            <properties>
+                <github.global.server>github</github.global.server>
+            </properties>
+            <build>
+                <plugins>
+                    <plugin>
+                        <groupId>com.github.github</groupId>
+                        <artifactId>site-maven-plugin</artifactId>
+                        <version>0.12</version>
+                        <configuration>
+                            <message>Maven artifacts for ${project.version}</message>
+                            <noJekyll>true</noJekyll>
+                            <outputDirectory>${project.build.directory}/mvn-repo</outputDirectory>
+                            <branch>refs/heads/maven-repo</branch>
+                            <excludes>
+                                <exclude>*-with-dependencies.jar</exclude>
+                            </excludes>
+                            <repositoryName>xgboost</repositoryName>
+                            <repositoryOwner>CodingCat</repositoryOwner>
+                            <merge>true</merge>
+                        </configuration>
+                        <executions>
+                            <!-- run site-maven-plugin's 'site' target as part of the build's normal 'deploy' phase -->
+                            <execution>
+                                <goals>
+                                    <goal>site</goal>
+                                </goals>
+                                <phase>deploy</phase>
+                            </execution>
+                        </executions>
+                    </plugin>
+                    <plugin>
+                        <groupId>org.apache.maven.plugins</groupId>
+                        <artifactId>maven-deploy-plugin</artifactId>
+                        <version>2.8.2</version>
+                        <configuration>
+                            <altDeploymentRepository>internal.repo::default::file://${project.build.directory}/mvn-repo</altDeploymentRepository>
+                        </configuration>
+                    </plugin>
+                </plugins>
+            </build>
         </profile>
     </profiles>
     <build>
@@ -173,27 +238,6 @@
                     </execution>
                 </executions>
             </plugin>
-
-            <plugin>
-                <groupId>org.apache.maven.plugins</groupId>
-                <artifactId>maven-assembly-plugin</artifactId>
-                <version>2.6</version>
-                <configuration>
-                    <descriptorRefs>
-                        <descriptorRef>jar-with-dependencies</descriptorRef>
-                    </descriptorRefs>
-                    <skipAssembly>true</skipAssembly>
-                </configuration>
-                <executions>
-                    <execution>
-                        <id>make-assembly</id>
-                        <phase>package</phase>
-                        <goals>
-                            <goal>single</goal>
-                        </goals>
-                    </execution>
-                </executions>
-            </plugin>
             <plugin>
                 <groupId>org.apache.maven.plugins</groupId>
                 <artifactId>maven-surefire-plugin</artifactId>

jvm-packages/xgboost4j-example/README.md

@@ -23,7 +23,7 @@ XGBoost4J Code Examples
 * [External Memory](src/main/scala/ml/dmlc/xgboost4j/scala/example/ExternalMemory.scala)
 
 ## Spark API
-* [Distributed Training with Spark](src/main/scala/ml/dmlc/xgboost4j/scala/example/spark/DistTrainWithSpark.scala)
+* [Distributed Training with Spark](src/main/scala/ml/dmlc/xgboost4j/scala/example/spark/SparkWithDataFrame.scala)
 
 ## Flink API
 * [Distributed Training with Flink](src/main/scala/ml/dmlc/xgboost4j/scala/example/flink/DistTrainWithFlink.scala)

jvm-packages/xgboost4j-example/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>0.7</version>
+        <version>0.72-SNAPSHOT</version>
     </parent>
     <artifactId>xgboost4j-example</artifactId>
-    <version>0.7</version>
+    <version>0.72-SNAPSHOT</version>
     <packaging>jar</packaging>
     <build>
         <plugins>
@@ -26,7 +26,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-spark</artifactId>
-            <version>0.7</version>
+            <version>0.72-SNAPSHOT</version>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>
@@ -37,7 +37,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-flink</artifactId>
-            <version>0.7</version>
+            <version>0.72-SNAPSHOT</version>
         </dependency>
         <dependency>
             <groupId>org.apache.commons</groupId>

jvm-packages/xgboost4j-flink/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>0.7</version>
+        <version>0.72-SNAPSHOT</version>
     </parent>
     <artifactId>xgboost4j-flink</artifactId>
-    <version>0.7</version>
+    <version>0.72-SNAPSHOT</version>
     <build>
         <plugins>
             <plugin>
@@ -26,7 +26,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j</artifactId>
-            <version>0.7</version>
+            <version>0.72-SNAPSHOT</version>
         </dependency>
         <dependency>
             <groupId>org.apache.commons</groupId>

jvm-packages/xgboost4j-spark/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>0.7</version>
+        <version>0.72-SNAPSHOT</version>
     </parent>
     <artifactId>xgboost4j-spark</artifactId>
     <build>
@@ -24,7 +24,19 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j</artifactId>
-            <version>0.7</version>
+            <version>0.72-SNAPSHOT</version>
+        </dependency>
+        <dependency>
+            <groupId>org.apache.spark</groupId>
+            <artifactId>spark-core_${scala.binary.version}</artifactId>
+            <version>${spark.version}</version>
+            <scope>provided</scope>
+        </dependency>
+        <dependency>
+            <groupId>org.apache.spark</groupId>
+            <artifactId>spark-sql_${scala.binary.version}</artifactId>
+            <version>${spark.version}</version>
+            <scope>provided</scope>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>

jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/CheckpointManager.scala

@@ -22,12 +22,16 @@ import org.apache.hadoop.fs.{FileSystem, Path}
 import org.apache.spark.SparkContext
 
 /**
-  * A class which allows user to save checkpoint boosters every a few rounds. If a previous job
+  * A class which allows user to save checkpoints every a few rounds. If a previous job fails,
-  * fails, the job can restart training from a saved booster instead of from scratch. This class
+  * the job can restart training from a saved checkpoints instead of from scratch. This class
   * provides interface and helper methods for the checkpoint functionality.
   *
+  * NOTE: This checkpoint is different from Rabit checkpoint. Rabit checkpoint is a native-level
+  * checkpoint stored in executor memory. This is a checkpoint which Spark driver store on HDFS
+  * for every a few iterations.
+  *
   * @param sc the sparkContext object
-  * @param checkpointPath the hdfs path to store checkpoint boosters
+  * @param checkpointPath the hdfs path to store checkpoints
   */
 private[spark] class CheckpointManager(sc: SparkContext, checkpointPath: String) {
   private val logger = LogFactory.getLog("XGBoostSpark")
@@ -49,11 +53,11 @@ private[spark] class CheckpointManager(sc: SparkContext, checkpointPath: String)
   }
 
   /**
-    * Load existing checkpoint with the highest version.
+    * Load existing checkpoint with the highest version as a Booster object
     *
     * @return the booster with the highest version, null if no checkpoints available.
     */
-  private[spark] def loadBooster: Booster = {
+  private[spark] def loadCheckpointAsBooster: Booster = {
     val versions = getExistingVersions
     if (versions.nonEmpty) {
       val version = versions.max
@@ -68,16 +72,16 @@ private[spark] class CheckpointManager(sc: SparkContext, checkpointPath: String)
   }
 
   /**
-    * Clean up all previous models and save a new model
+    * Clean up all previous checkpoints and save a new checkpoint
     *
-    * @param model the xgboost model to save
+    * @param checkpoint the checkpoint to save as an XGBoostModel
     */
-  private[spark] def updateModel(model: XGBoostModel): Unit = {
+  private[spark] def updateCheckpoint(checkpoint: XGBoostModel): Unit = {
     val fs = FileSystem.get(sc.hadoopConfiguration)
     val prevModelPaths = getExistingVersions.map(version => new Path(getPath(version)))
-    val fullPath = getPath(model.version)
+    val fullPath = getPath(checkpoint.version)
-    logger.info(s"Saving checkpoint model with version ${model.version} to $fullPath")
+    logger.info(s"Saving checkpoint model with version ${checkpoint.version} to $fullPath")
-    model.saveModelAsHadoopFile(fullPath)(sc)
+    checkpoint.saveModelAsHadoopFile(fullPath)(sc)
     prevModelPaths.foreach(path => fs.delete(path, true))
   }
 
@@ -95,22 +99,22 @@ private[spark] class CheckpointManager(sc: SparkContext, checkpointPath: String)
   }
 
   /**
-    * Calculate a list of checkpoint rounds to save checkpoints based on the savingFreq and
+    * Calculate a list of checkpoint rounds to save checkpoints based on the checkpointInterval
-    * total number of rounds for the training. Concretely, the saving rounds start with
+    * and total number of rounds for the training. Concretely, the checkpoint rounds start with
-    * prevRounds + savingFreq, and increase by savingFreq in each step until it reaches total
+    * prevRounds + checkpointInterval, and increase by checkpointInterval in each step until it
-    * number of rounds. If savingFreq is 0, the checkpoint will be disabled and the method
+    * reaches total number of rounds. If checkpointInterval is 0, the checkpoint will be disabled
-    * returns Seq(round)
+    * and the method returns Seq(round)
     *
-    * @param savingFreq the increase on rounds during each step of training
+    * @param checkpointInterval Period (in iterations) between checkpoints.
     * @param round the total number of rounds for the training
     * @return a seq of integers, each represent the index of round to save the checkpoints
     */
-  private[spark] def getSavingRounds(savingFreq: Int, round: Int): Seq[Int] = {
+  private[spark] def getCheckpointRounds(checkpointInterval: Int, round: Int): Seq[Int] = {
-    if (checkpointPath.nonEmpty && savingFreq > 0) {
+    if (checkpointPath.nonEmpty && checkpointInterval > 0) {
       val prevRounds = getExistingVersions.map(_ / 2)
-      val firstSavingRound = (0 +: prevRounds).max + savingFreq
+      val firstCheckpointRound = (0 +: prevRounds).max + checkpointInterval
-      (firstSavingRound until round by savingFreq) :+ round
+      (firstCheckpointRound until round by checkpointInterval) :+ round
-    } else if (savingFreq <= 0) {
+    } else if (checkpointInterval <= 0) {
       Seq(round)
     } else {
       throw new IllegalArgumentException("parameters \"checkpoint_path\" should also be set.")
@@ -128,12 +132,12 @@ object CheckpointManager {
         " an instance of String.")
     }
 
-    val savingFreq: Int = params.get("saving_frequency") match {
+    val checkpointInterval: Int = params.get("checkpoint_interval") match {
       case None => 0
       case Some(freq: Int) => freq
-      case _ => throw new IllegalArgumentException("parameter \"saving_frequency\" must be" +
+      case _ => throw new IllegalArgumentException("parameter \"checkpoint_interval\" must be" +
         " an instance of Int.")
     }
-    (checkpointPath, savingFreq)
+    (checkpointPath, checkpointInterval)
   }
 }

jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/XGBoost.scala

@@ -17,6 +17,7 @@
 package ml.dmlc.xgboost4j.scala.spark
 
 import java.io.File
+import java.nio.file.Files
 
 import scala.collection.mutable
 import scala.util.Random
@@ -24,6 +25,7 @@ import ml.dmlc.xgboost4j.java.{IRabitTracker, Rabit, XGBoostError, RabitTracker
 import ml.dmlc.xgboost4j.scala.rabit.RabitTracker
 import ml.dmlc.xgboost4j.scala.{XGBoost => SXGBoost, _}
 import ml.dmlc.xgboost4j.{LabeledPoint => XGBLabeledPoint}
+import org.apache.commons.io.FileUtils
 import org.apache.commons.logging.LogFactory
 import org.apache.hadoop.fs.{FSDataInputStream, Path}
 import org.apache.spark.rdd.RDD
@@ -120,11 +122,8 @@ object XGBoost extends Serializable {
       }
       val taskId = TaskContext.getPartitionId().toString
       val cacheDirName = if (useExternalMemory) {
-        val dir = new File(s"${TaskContext.get().stageId()}-cache-$taskId")
+        val dir = Files.createTempDirectory(s"${TaskContext.get().stageId()}-cache-$taskId")
-        if (!(dir.exists() || dir.mkdirs())) {
+        Some(dir.toAbsolutePath.toString)
-          throw new XGBoostError(s"failed to create cache directory: $dir")
-        }
-        Some(dir.toString)
       } else {
         None
       }
@@ -325,23 +324,24 @@ object XGBoost extends Serializable {
       case _ => throw new IllegalArgumentException("parameter \"timeout_request_workers\" must be" +
         " an instance of Long.")
     }
-    val (checkpointPath, savingFeq) = CheckpointManager.extractParams(params)
+    val (checkpointPath, checkpointInterval) = CheckpointManager.extractParams(params)
     val partitionedData = repartitionForTraining(trainingData, nWorkers)
 
     val sc = trainingData.sparkContext
     val checkpointManager = new CheckpointManager(sc, checkpointPath)
     checkpointManager.cleanUpHigherVersions(round)
 
-    var prevBooster = checkpointManager.loadBooster
+    var prevBooster = checkpointManager.loadCheckpointAsBooster
     // Train for every ${savingRound} rounds and save the partially completed booster
-    checkpointManager.getSavingRounds(savingFeq, round).map {
+    checkpointManager.getCheckpointRounds(checkpointInterval, round).map {
-      savingRound: Int =>
+      checkpointRound: Int =>
         val tracker = startTracker(nWorkers, trackerConf)
         try {
           val parallelismTracker = new SparkParallelismTracker(sc, timeoutRequestWorkers, nWorkers)
           val overriddenParams = overrideParamsAccordingToTaskCPUs(params, sc)
           val boostersAndMetrics = buildDistributedBoosters(partitionedData, overriddenParams,
-            tracker.getWorkerEnvs, savingRound, obj, eval, useExternalMemory, missing, prevBooster)
+            tracker.getWorkerEnvs, checkpointRound, obj, eval, useExternalMemory, missing,
+            prevBooster)
           val sparkJobThread = new Thread() {
             override def run() {
               // force the job
@@ -359,9 +359,9 @@ object XGBoost extends Serializable {
             model.asInstanceOf[XGBoostClassificationModel].numOfClasses =
               params.getOrElse("num_class", "2").toString.toInt
           }
-          if (savingRound < round) {
+          if (checkpointRound < round) {
             prevBooster = model.booster
-            checkpointManager.updateModel(model)
+            checkpointManager.updateCheckpoint(model)
           }
           model
       } finally {
@@ -480,11 +480,7 @@ private class Watches private(
   def delete(): Unit = {
     toMap.values.foreach(_.delete())
     cacheDirName.foreach { name =>
-      for (cacheFile <- new File(name).listFiles()) {
+      FileUtils.deleteDirectory(new File(name))
-        if (!cacheFile.delete()) {
-          throw new IllegalStateException(s"failed to delete $cacheFile")
-        }
-      }
     }
   }
 

jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostModel.scala

@@ -169,12 +169,12 @@ abstract class XGBoostModel(protected var _booster: Booster)
   def predict(testSet: RDD[MLDenseVector], missingValue: Float): RDD[Array[Float]] = {
     val broadcastBooster = testSet.sparkContext.broadcast(_booster)
     testSet.mapPartitions { testSamples =>
-      val sampleArray = testSamples.toList
+      val sampleArray = testSamples.toArray
-      val numRows = sampleArray.size
+      val numRows = sampleArray.length
-      val numColumns = sampleArray.head.size
       if (numRows == 0) {
         Iterator()
       } else {
+        val numColumns = sampleArray.head.size
         val rabitEnv = Map("DMLC_TASK_ID" -> TaskContext.getPartitionId().toString)
         Rabit.init(rabitEnv.asJava)
         // translate to required format

jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/params/GeneralParams.scala

@@ -71,7 +71,7 @@ trait GeneralParams extends Params {
   val missing = new FloatParam(this, "missing", "the value treated as missing")
 
   /**
-    * the interval to check whether total numCores is no smaller than nWorkers. default: 30 minutes
+    * the maximum time to wait for the job requesting new workers. default: 30 minutes
     */
   val timeoutRequestWorkers = new LongParam(this, "timeout_request_workers", "the maximum time to" +
     " request new Workers if numCores are insufficient. The timeout will be disabled if this" +
@@ -81,16 +81,19 @@ trait GeneralParams extends Params {
     * The hdfs folder to load and save checkpoint boosters. default: `empty_string`
     */
   val checkpointPath = new Param[String](this, "checkpoint_path", "the hdfs folder to load and " +
-    "save checkpoints. The job will try to load the existing booster as the starting point for " +
+    "save checkpoints. If there are existing checkpoints in checkpoint_path. The job will load " +
-    "training. If saving_frequency is also set, the job will save a checkpoint every a few rounds.")
+    "the checkpoint with highest version as the starting point for training. If " +
+    "checkpoint_interval is also set, the job will save a checkpoint every a few rounds.")
 
   /**
-    * The frequency to save checkpoint boosters. default: 0
+    * Param for set checkpoint interval (>= 1) or disable checkpoint (-1). E.g. 10 means that
+    * the trained model will get checkpointed every 10 iterations. Note: `checkpoint_path` must
+    * also be set if the checkpoint interval is greater than 0.
     */
-  val savingFrequency = new IntParam(this, "saving_frequency", "if checkpoint_path is also set," +
+  val checkpointInterval: IntParam = new IntParam(this, "checkpointInterval", "set checkpoint " +
-    " the job will save checkpoints at this frequency. If the job fails and gets restarted with" +
+    "interval (>= 1) or disable checkpoint (-1). E.g. 10 means that the trained model will get " +
-    " same setting, it will load the existing booster instead of training from scratch." +
+    "checkpointed every 10 iterations. Note: `checkpoint_path` must also be set if the checkpoint" +
-    " Checkpoint will be disabled if set to 0.")
+    " interval is greater than 0.", (interval: Int) => interval == -1 || interval >= 1)
 
   /**
     * Rabit tracker configurations. The parameter must be provided as an instance of the
@@ -128,6 +131,6 @@ trait GeneralParams extends Params {
     useExternalMemory -> false, silent -> 0,
     customObj -> null, customEval -> null, missing -> Float.NaN,
     trackerConf -> TrackerConf(), seed -> 0, timeoutRequestWorkers -> 30 * 60 * 1000L,
-    checkpointPath -> "", savingFrequency -> 0
+    checkpointPath -> "", checkpointInterval -> -1
   )
 }

jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/params/LearningTaskParams.scala

@@ -45,7 +45,8 @@ trait LearningTaskParams extends Params {
   /**
    * evaluation metrics for validation data, a default metric will be assigned according to
    * objective(rmse for regression, and error for classification, mean average precision for
-   * ranking). options: rmse, mae, logloss, error, merror, mlogloss, auc, ndcg, map, gamma-deviance
+   * ranking). options: rmse, mae, logloss, error, merror, mlogloss, auc, aucpr, ndcg, map,
+   * gamma-deviance
    */
   val evalMetric = new Param[String](this, "eval_metric", "evaluation metrics for validation" +
     " data, a default metric will be assigned according to objective (rmse for regression, and" +
@@ -97,5 +98,5 @@ private[spark] object LearningTaskParams {
     "reg:gamma")
 
   val supportedEvalMetrics = HashSet("rmse", "mae", "logloss", "error", "merror", "mlogloss",
-    "auc", "ndcg", "map", "gamma-deviance")
+    "auc", "aucpr", "ndcg", "map", "gamma-deviance")
 }

jvm-packages/xgboost4j-spark/src/main/scala/org/apache/spark/SparkParallelismTracker.scala

@@ -76,11 +76,12 @@ class SparkParallelismTracker(
   }
 
   private[this] def safeExecute[T](body: => T): T = {
-    sc.listenerBus.listeners.add(0, new TaskFailedListener)
+    val listener = new TaskFailedListener;
+    sc.addSparkListener(listener)
     try {
       body
     } finally {
-      sc.listenerBus.listeners.remove(0)
+      sc.listenerBus.removeListener(listener)
     }
   }
 

jvm-packages/xgboost4j-spark/src/test/scala/ml/dmlc/xgboost4j/scala/spark/CheckpointManagerSuite.scala

@@ -38,30 +38,30 @@ class CheckpointManagerSuite extends FunSuite  with BeforeAndAfterAll {
     val trainingRDD = sc.parallelize(Classification.train).map(_.asML).cache()
     val paramMap = Map("eta" -> "1", "max_depth" -> "2", "silent" -> "1",
       "objective" -> "binary:logistic")
-    (XGBoost.trainWithRDD(trainingRDD, paramMap, round = 2, sc.defaultParallelism),
+    (XGBoost.trainWithRDD(trainingRDD, paramMap, round = 2, nWorkers = sc.defaultParallelism),
-      XGBoost.trainWithRDD(trainingRDD, paramMap, round = 4, sc.defaultParallelism))
+      XGBoost.trainWithRDD(trainingRDD, paramMap, round = 4, nWorkers = sc.defaultParallelism))
   }
 
   test("test update/load models") {
     val tmpPath = Files.createTempDirectory("test").toAbsolutePath.toString
     val manager = new CheckpointManager(sc, tmpPath)
-    manager.updateModel(model4)
+    manager.updateCheckpoint(model4)
     var files = FileSystem.get(sc.hadoopConfiguration).listStatus(new Path(tmpPath))
     assert(files.length == 1)
     assert(files.head.getPath.getName == "4.model")
-    assert(manager.loadBooster.booster.getVersion == 4)
+    assert(manager.loadCheckpointAsBooster.booster.getVersion == 4)
 
-    manager.updateModel(model8)
+    manager.updateCheckpoint(model8)
     files = FileSystem.get(sc.hadoopConfiguration).listStatus(new Path(tmpPath))
     assert(files.length == 1)
     assert(files.head.getPath.getName == "8.model")
-    assert(manager.loadBooster.booster.getVersion == 8)
+    assert(manager.loadCheckpointAsBooster.booster.getVersion == 8)
   }
 
   test("test cleanUpHigherVersions") {
     val tmpPath = Files.createTempDirectory("test").toAbsolutePath.toString
     val manager = new CheckpointManager(sc, tmpPath)
-    manager.updateModel(model8)
+    manager.updateCheckpoint(model8)
     manager.cleanUpHigherVersions(round = 8)
     assert(new File(s"$tmpPath/8.model").exists())
 
@@ -69,12 +69,12 @@ class CheckpointManagerSuite extends FunSuite  with BeforeAndAfterAll {
     assert(!new File(s"$tmpPath/8.model").exists())
   }
 
-  test("test saving rounds") {
+  test("test checkpoint rounds") {
     val tmpPath = Files.createTempDirectory("test").toAbsolutePath.toString
     val manager = new CheckpointManager(sc, tmpPath)
-    assertResult(Seq(7))(manager.getSavingRounds(savingFreq = 0, round = 7))
+    assertResult(Seq(7))(manager.getCheckpointRounds(checkpointInterval = 0, round = 7))
-    assertResult(Seq(2, 4, 6, 7))(manager.getSavingRounds(savingFreq = 2, round = 7))
+    assertResult(Seq(2, 4, 6, 7))(manager.getCheckpointRounds(checkpointInterval = 2, round = 7))
-    manager.updateModel(model4)
+    manager.updateCheckpoint(model4)
-    assertResult(Seq(4, 6, 7))(manager.getSavingRounds(2, 7))
+    assertResult(Seq(4, 6, 7))(manager.getCheckpointRounds(2, 7))
   }
 }

jvm-packages/xgboost4j-spark/src/test/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostGeneralSuite.scala

@@ -338,7 +338,7 @@ class XGBoostGeneralSuite extends FunSuite with PerTest {
     }
   }
 
-  test("training with saving checkpoint boosters") {
+  test("training with checkpoint boosters") {
     import DataUtils._
     val eval = new EvalError()
     val trainingRDD = sc.parallelize(Classification.train).map(_.asML)
@@ -347,7 +347,7 @@ class XGBoostGeneralSuite extends FunSuite with PerTest {
     val tmpPath = Files.createTempDirectory("model1").toAbsolutePath.toString
     val paramMap = List("eta" -> "1", "max_depth" -> 2, "silent" -> "1",
       "objective" -> "binary:logistic", "checkpoint_path" -> tmpPath,
-      "saving_frequency" -> 2).toMap
+      "checkpoint_interval" -> 2).toMap
     val prevModel = XGBoost.trainWithRDD(trainingRDD, paramMap, round = 5,
       nWorkers = numWorkers)
     def error(model: XGBoostModel): Float = eval.eval(

jvm-packages/xgboost4j/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>0.7</version>
+        <version>0.72-SNAPSHOT</version>
     </parent>
     <artifactId>xgboost4j</artifactId>
-    <version>0.7</version>
+    <version>0.72-SNAPSHOT</version>
     <packaging>jar</packaging>
 
     <dependencies>

jvm-packages/xgboost4j/src/main/scala/ml/dmlc/xgboost4j/scala/Booster.scala

@@ -16,8 +16,6 @@
 
 package ml.dmlc.xgboost4j.scala
 
-import java.io.IOException
-
 import com.esotericsoftware.kryo.io.{Output, Input}
 import com.esotericsoftware.kryo.{Kryo, KryoSerializable}
 import ml.dmlc.xgboost4j.java.{Booster => JBooster}
@@ -25,6 +23,12 @@ import ml.dmlc.xgboost4j.java.XGBoostError
 import scala.collection.JavaConverters._
 import scala.collection.mutable
 
+/**
+  * Booster for xgboost, this is a model API that support interactive build of a XGBoost Model
+  *
+  * DEVELOPER WARNING: A Java Booster must not be shared by more than one Scala Booster
+  * @param booster the java booster object.
+  */
 class Booster private[xgboost4j](private[xgboost4j] var booster: JBooster)
   extends Serializable  with KryoSerializable {
 

jvm-packages/xgboost4j/src/main/scala/ml/dmlc/xgboost4j/scala/XGBoost.scala

@@ -66,7 +66,12 @@ object XGBoost {
       // we have to filter null value for customized obj and eval
       params.filter(_._2 != null).mapValues(_.toString.asInstanceOf[AnyRef]).asJava,
       round, jWatches, metrics, obj, eval, earlyStoppingRound, jBooster)
-    new Booster(xgboostInJava)
+    if (booster == null) {
+      new Booster(xgboostInJava)
+    } else {
+      // Avoid creating a new SBooster with the same JBooster
+      booster
+    }
   }
 
   /**

jvm-packages/xgboost4j/src/test/scala/ml/dmlc/xgboost4j/scala/ScalaBoosterImplSuite.scala

@@ -198,4 +198,16 @@ class ScalaBoosterImplSuite extends FunSuite {
     trainBoosterWithFastHisto(trainMat, Map("training" -> trainMat),
       round = 10, paramMap, 0.85f)
   }
+
+  test("test training from existing model in scala") {
+    val trainMat = new DMatrix("../../demo/data/agaricus.txt.train")
+    val paramMap = List("max_depth" -> "0", "silent" -> "0",
+      "objective" -> "binary:logistic", "tree_method" -> "hist",
+      "grow_policy" -> "depthwise", "max_depth" -> "2", "max_bin" -> "2",
+      "eval_metric" -> "auc").toMap
+
+    val prevBooster = XGBoost.train(trainMat, paramMap, round = 2)
+    val nextBooster = XGBoost.train(trainMat, paramMap, round = 4, booster = prevBooster)
+    assert(prevBooster == nextBooster)
+  }
 }

plugin/dense_parser/dense_libsvm.cc

@@ -69,7 +69,7 @@ class DensifyParser : public dmlc::Parser<IndexType> {
   std::vector<xgboost::bst_float> dense_value_;
 };
 
-template<typename IndexType>
+template<typename IndexType, typename DType = real_t>
 Parser<IndexType> *
 CreateDenseLibSVMParser(const std::string& path,
                         const std::map<std::string, std::string>& args,
@@ -82,5 +82,6 @@ CreateDenseLibSVMParser(const std::string& path,
 }
 }  // namespace data
 
-DMLC_REGISTER_DATA_PARSER(uint32_t, dense_libsvm, data::CreateDenseLibSVMParser<uint32_t>);
+DMLC_REGISTER_DATA_PARSER(uint32_t, real_t, dense_libsvm,
+  data::CreateDenseLibSVMParser<uint32_t __DMLC_COMMA real_t>);
 }  // namespace dmlc

plugin/example/custom_obj.cc

@@ -33,30 +33,32 @@ class MyLogistic : public ObjFunction {
   void Configure(const std::vector<std::pair<std::string, std::string> >& args) override {
     param_.InitAllowUnknown(args);
   }
-  void GetGradient(const std::vector<bst_float> &preds,
+  void GetGradient(HostDeviceVector<bst_float> *preds,
                    const MetaInfo &info,
                    int iter,
-                   std::vector<bst_gpair> *out_gpair) override {
+                   HostDeviceVector<GradientPair> *out_gpair) override {
-    out_gpair->resize(preds.size());
+    out_gpair->Resize(preds->Size());
-    for (size_t i = 0; i < preds.size(); ++i) {
+    std::vector<bst_float>& preds_h = preds->HostVector();
+    std::vector<GradientPair>& out_gpair_h = out_gpair->HostVector();
+    for (size_t i = 0; i < preds_h.size(); ++i) {
       bst_float w = info.GetWeight(i);
       // scale the negative examples!
-      if (info.labels[i] == 0.0f) w *= param_.scale_neg_weight;
+      if (info.labels_[i] == 0.0f) w *= param_.scale_neg_weight;
       // logistic transformation
-      bst_float p = 1.0f / (1.0f + std::exp(-preds[i]));
+      bst_float p = 1.0f / (1.0f + std::exp(-preds_h[i]));
       // this is the gradient
-      bst_float grad = (p - info.labels[i]) * w;
+      bst_float grad = (p - info.labels_[i]) * w;
       // this is the second order gradient
       bst_float hess = p * (1.0f - p) * w;
-      out_gpair->at(i) = bst_gpair(grad, hess);
+      out_gpair_h.at(i) = GradientPair(grad, hess);
     }
   }
   const char* DefaultEvalMetric() const override {
     return "error";
   }
-  void PredTransform(std::vector<bst_float> *io_preds) override {
+  void PredTransform(HostDeviceVector<bst_float> *io_preds) override {
     // transform margin value to probability.
-    std::vector<bst_float> &preds = *io_preds;
+    std::vector<bst_float> &preds = io_preds->HostVector();
     for (size_t i = 0; i < preds.size(); ++i) {
       preds[i] = 1.0f / (1.0f + std::exp(-preds[i]));
     }

python-package/README.rst

@@ -26,6 +26,11 @@ Please install ``gcc@5`` from `Homebrew <https://brew.sh/>`_::
 
     brew install gcc@5
 
+After installing ``gcc@5``, set it as your compiler::
+
+    export CC = gcc-5
+    export CXX = g++-5
+
 Linux
 -----
 

python-package/build_trouble_shooting.md

@@ -10,7 +10,7 @@ Linux platform (also Mac OS X in general)
 ------------
 **Trouble 0**: I see error messages like this when install from github using `python setup.py install`.
 
-    XGBoostLibraryNotFound: Cannot find XGBoost Libarary in the candidate path, did you install compilers and run build.sh in root path?
+    XGBoostLibraryNotFound: Cannot find XGBoost Library in the candidate path, did you install compilers and run build.sh in root path?
     List of candidates:
     /home/dmlc/anaconda/lib/python2.7/site-packages/xgboost-0.4-py2.7.egg/xgboost/libxgboostwrapper.so
     /home/dmlc/anaconda/lib/python2.7/site-packages/xgboost-0.4-py2.7.egg/xgboost/../../wrapper/libxgboostwrapper.so

python-package/setup_pip.py

@@ -12,7 +12,10 @@ sys.path.insert(0, '.')
 # please don't use this file for installing from github
 
 if os.name != 'nt':     # if not windows, compile and install
-    os.system('sh ./xgboost/build-python.sh')
+    # if not windows, compile and install
+    if len(sys.argv) < 2 or sys.argv[1] != 'sdist':
+        # do not build for sdist
+        os.system('sh ./xgboost/build-python.sh')
 else:
     print('Windows users please use github installation.')
     sys.exit()
@@ -30,16 +33,14 @@ class BinaryDistribution(Distribution):
 # We can not import `xgboost.libpath` in setup.py directly since xgboost/__init__.py
 # import `xgboost.core` and finally will import `numpy` and `scipy` which are setup
 # `install_requires`. That's why we're using `exec` here.
-libpath_py = os.path.join(CURRENT_DIR, 'xgboost/libpath.py')
+# do not import libpath for sdist
-libpath = {'__file__': libpath_py}
+if len(sys.argv) < 2 or sys.argv[1] != 'sdist':
-exec(compile(open(libpath_py, "rb").read(), libpath_py, 'exec'), libpath, libpath)
+    libpath_py = os.path.join(CURRENT_DIR, 'xgboost/libpath.py')
+    libpath = {'__file__': libpath_py}
+    exec(compile(open(libpath_py, "rb").read(), libpath_py, 'exec'), libpath, libpath)
 
-LIB_PATH = libpath['find_lib_path']()
+    LIB_PATH = libpath['find_lib_path']()
 
-# to deploy to pip, please use
-# make pythonpack
-# python setup.py register sdist upload
-# and be sure to test it firstly using "python setup.py register sdist upload -r pypitest"
 setup(name='xgboost',
       version=open(os.path.join(CURRENT_DIR, 'xgboost/VERSION')).read().strip(),
       description='XGBoost Python Package',

python-package/xgboost/VERSION

@@ -1 +1 @@
-0.7
+0.72

python-package/xgboost/build-python.sh

@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/bin/sh
 # This is a simple script to make xgboost in MAC and Linux for python wrapper only
 # Basically, it first try to make with OpenMP, if fails, disable OpenMP and make it again.
 # This will automatically make xgboost for MAC users who don't have OpenMP support.
@@ -9,22 +9,44 @@
 # note: this script is build for python package only, and it might have some filename
 #       conflict with build.sh which is for everything. 
 
+set -e
+set -x
 
 #pushd xgboost
 oldpath=`pwd`
 cd ./xgboost/
+
+if echo "${OSTYPE}" | grep -q "darwin"; then
+  LIB_XGBOOST=libxgboost.dylib
+  # Use OpenMP-capable compiler if possible
+  if which g++-5; then
+    export CC=gcc-5
+    export CXX=g++-5
+  elif which g++-7; then
+    export CC=gcc-7
+    export CXX=g++-7
+  elif which clang++; then
+    export CC=clang
+    export CXX=clang++
+  fi
+else
+  LIB_XGBOOST=libxgboost.so
+fi
+
 #remove the pre-compiled .so and trigger the system's on-the-fly compiling
 make clean
-if make lib/libxgboost.so -j4; then
+if make lib/${LIB_XGBOOST} -j4; then
     echo "Successfully build multi-thread xgboost"
 else
     echo "-----------------------------"
     echo "Building multi-thread xgboost failed"
     echo "Start to build single-thread xgboost"
     make clean
-    make lib/libxgboost.so -j4 USE_OPENMP=0
+    make lib/${LIB_XGBOOST} -j4 USE_OPENMP=0
     echo "Successfully build single-thread xgboost"
     echo "If you want multi-threaded version"
     echo "See additional instructions in doc/build.md"
 fi
 cd $oldpath
+
+set +x

python-package/xgboost/callback.py

@@ -50,7 +50,7 @@ def print_evaluation(period=1, show_stdv=True):
     """
     def callback(env):
         """internal function"""
-        if env.rank != 0 or len(env.evaluation_result_list) == 0 or period is False:
+        if env.rank != 0 or len(env.evaluation_result_list) == 0 or period is False or period == 0:
             return
         i = env.iteration
         if (i % period == 0 or i + 1 == env.begin_iteration or i + 1 == env.end_iteration):

python-package/xgboost/core.py

@@ -235,8 +235,6 @@ class DMatrix(object):
                  feature_names=None, feature_types=None,
                  nthread=None):
         """
-        Data matrix used in XGBoost.
-
         Parameters
         ----------
         data : string/numpy array/scipy.sparse/pd.DataFrame
@@ -628,9 +626,8 @@ class DMatrix(object):
         feature_names : list or None
         """
         if self._feature_names is None:
-            return ['f{0}'.format(i) for i in range(self.num_col())]
+            self._feature_names = ['f{0}'.format(i) for i in range(self.num_col())]
-        else:
+        return self._feature_names
-            return self._feature_names
 
     @property
     def feature_types(self):
@@ -706,7 +703,7 @@ class DMatrix(object):
 
 
 class Booster(object):
-    """"A Booster of of XGBoost.
+    """A Booster of of XGBoost.
 
     Booster is the model of xgboost, that contains low level routines for
     training, prediction and evaluation.
@@ -716,8 +713,7 @@ class Booster(object):
 
     def __init__(self, params=None, cache=(), model_file=None):
         # pylint: disable=invalid-name
-        """Initialize the Booster.
+        """
-
         Parameters
         ----------
         params : dict
@@ -992,7 +988,8 @@ class Booster(object):
         return self.eval_set([(data, name)], iteration)
 
     def predict(self, data, output_margin=False, ntree_limit=0, pred_leaf=False,
-                pred_contribs=False, approx_contribs=False):
+                pred_contribs=False, approx_contribs=False, pred_interactions=False,
+                validate_features=True):
         """
         Predict with data.
 
@@ -1019,14 +1016,25 @@ class Booster(object):
             in both tree 1 and tree 0.
 
         pred_contribs : bool
-            When this option is on, the output will be a matrix of (nsample, nfeats+1)
+            When this is True the output will be a matrix of size (nsample, nfeats + 1)
             with each record indicating the feature contributions (SHAP values) for that
-            prediction. The sum of all feature contributions is equal to the prediction.
+            prediction. The sum of all feature contributions is equal to the raw untransformed
-            Note that the bias is added as the final column, on top of the regular features.
+            margin value of the prediction. Note the final column is the bias term.
 
         approx_contribs : bool
             Approximate the contributions of each feature
 
+        pred_interactions : bool
+            When this is True the output will be a matrix of size (nsample, nfeats + 1, nfeats + 1)
+            indicating the SHAP interaction values for each pair of features. The sum of each
+            row (or column) of the interaction values equals the corresponding SHAP value (from
+            pred_contribs), and the sum of the entire matrix equals the raw untransformed margin
+            value of the prediction. Note the last row and column correspond to the bias term.
+
+        validate_features : bool
+            When this is True, validate that the Booster's and data's feature_names are identical.
+            Otherwise, it is assumed that the feature_names are the same.
+
         Returns
         -------
         prediction : numpy array
@@ -1040,8 +1048,11 @@ class Booster(object):
             option_mask |= 0x04
         if approx_contribs:
             option_mask |= 0x08
+        if pred_interactions:
+            option_mask |= 0x10
 
-        self._validate_features(data)
+        if validate_features:
+            self._validate_features(data)
 
         length = c_bst_ulong()
         preds = ctypes.POINTER(ctypes.c_float)()
@@ -1055,8 +1066,22 @@ class Booster(object):
             preds = preds.astype(np.int32)
         nrow = data.num_row()
         if preds.size != nrow and preds.size % nrow == 0:
-            ncol = int(preds.size / nrow)
+            chunk_size = int(preds.size / nrow)
-            preds = preds.reshape(nrow, ncol)
+
+            if pred_interactions:
+                ngroup = int(chunk_size / ((data.num_col() + 1) * (data.num_col() + 1)))
+                if ngroup == 1:
+                    preds = preds.reshape(nrow, data.num_col() + 1, data.num_col() + 1)
+                else:
+                    preds = preds.reshape(nrow, ngroup, data.num_col() + 1, data.num_col() + 1)
+            elif pred_contribs:
+                ngroup = int(chunk_size / (data.num_col() + 1))
+                if ngroup == 1:
+                    preds = preds.reshape(nrow, data.num_col() + 1)
+                else:
+                    preds = preds.reshape(nrow, ngroup, data.num_col() + 1)
+            else:
+                preds = preds.reshape(nrow, chunk_size)
         return preds
 
     def save_model(self, fname):

python-package/xgboost/libpath.py

@@ -34,7 +34,7 @@ def find_lib_path():
             # hack for pip installation when copy all parent source directory here
             dll_path.append(os.path.join(curr_path, './windows/Release/'))
         dll_path = [os.path.join(p, 'xgboost.dll') for p in dll_path]
-    elif sys.platform.startswith('linux'):
+    elif sys.platform.startswith('linux') or sys.platform.startswith('freebsd'):
         dll_path = [os.path.join(p, 'libxgboost.so') for p in dll_path]
     elif sys.platform == 'darwin':
         dll_path = [os.path.join(p, 'libxgboost.dylib') for p in dll_path]

python-package/xgboost/sklearn.py

@@ -215,7 +215,8 @@ class XGBModel(XGBModelBase):
         return xgb_params
 
     def fit(self, X, y, sample_weight=None, eval_set=None, eval_metric=None,
-            early_stopping_rounds=None, verbose=True, xgb_model=None):
+            early_stopping_rounds=None, verbose=True, xgb_model=None,
+            sample_weight_eval_set=None):
         # pylint: disable=missing-docstring,invalid-name,attribute-defined-outside-init
         """
         Fit the gradient boosting model
@@ -231,6 +232,9 @@ class XGBModel(XGBModelBase):
         eval_set : list, optional
             A list of (X, y) tuple pairs to use as a validation set for
             early-stopping
+        sample_weight_eval_set : list, optional
+            A list of the form [L_1, L_2, ..., L_n], where each L_i is a list of
+            instance weights on the i-th validation set.
         eval_metric : str, callable, optional
             If a str, should be a built-in evaluation metric to use. See
             doc/parameter.md. If callable, a custom evaluation metric. The call
@@ -263,9 +267,14 @@ class XGBModel(XGBModelBase):
             trainDmatrix = DMatrix(X, label=y, missing=self.missing, nthread=self.n_jobs)
 
         evals_result = {}
+
         if eval_set is not None:
-            evals = list(DMatrix(x[0], label=x[1], missing=self.missing,
+            if sample_weight_eval_set is None:
-                                 nthread=self.n_jobs) for x in eval_set)
+                sample_weight_eval_set = [None] * len(eval_set)
+            evals = list(
+                DMatrix(eval_set[i][0], label=eval_set[i][1], missing=self.missing,
+                        weight=sample_weight_eval_set[i], nthread=self.n_jobs)
+                for i in range(len(eval_set)))
             evals = list(zip(evals, ["validation_{}".format(i) for i in
                                      range(len(evals))]))
         else:
@@ -408,7 +417,8 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
                                             random_state, seed, missing, **kwargs)
 
     def fit(self, X, y, sample_weight=None, eval_set=None, eval_metric=None,
-            early_stopping_rounds=None, verbose=True, xgb_model=None):
+            early_stopping_rounds=None, verbose=True, xgb_model=None,
+            sample_weight_eval_set=None):
         # pylint: disable = attribute-defined-outside-init,arguments-differ
         """
         Fit gradient boosting classifier
@@ -424,6 +434,9 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
         eval_set : list, optional
             A list of (X, y) pairs to use as a validation set for
             early-stopping
+        sample_weight_eval_set : list, optional
+            A list of the form [L_1, L_2, ..., L_n], where each L_i is a list of
+            instance weights on the i-th validation set.
         eval_metric : str, callable, optional
             If a str, should be a built-in evaluation metric to use. See
             doc/parameter.md. If callable, a custom evaluation metric. The call
@@ -478,11 +491,13 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
         training_labels = self._le.transform(y)
 
         if eval_set is not None:
-            # TODO: use sample_weight if given?
+            if sample_weight_eval_set is None:
+                sample_weight_eval_set = [None] * len(eval_set)
             evals = list(
-                DMatrix(x[0], label=self._le.transform(x[1]),
+                DMatrix(eval_set[i][0], label=self._le.transform(eval_set[i][1]),
-                        missing=self.missing, nthread=self.n_jobs)
+                        missing=self.missing, weight=sample_weight_eval_set[i],
-                for x in eval_set
+                        nthread=self.n_jobs)
+                for i in range(len(eval_set))
             )
             nevals = len(evals)
             eval_names = ["validation_{}".format(i) for i in range(nevals)]
@@ -520,6 +535,24 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
         return self
 
     def predict(self, data, output_margin=False, ntree_limit=0):
+        """
+        Predict with `data`.
+        NOTE: This function is not thread safe.
+              For each booster object, predict can only be called from one thread.
+              If you want to run prediction using multiple thread, call xgb.copy() to make copies
+              of model object and then call predict
+        Parameters
+        ----------
+        data : DMatrix
+            The dmatrix storing the input.
+        output_margin : bool
+            Whether to output the raw untransformed margin value.
+        ntree_limit : int
+            Limit number of trees in the prediction; defaults to 0 (use all trees).
+        Returns
+        -------
+        prediction : numpy array
+        """
         test_dmatrix = DMatrix(data, missing=self.missing, nthread=self.n_jobs)
         class_probs = self.get_booster().predict(test_dmatrix,
                                                  output_margin=output_margin,
@@ -531,10 +564,26 @@ class XGBClassifier(XGBModel, XGBClassifierBase):
             column_indexes[class_probs > 0.5] = 1
         return self._le.inverse_transform(column_indexes)
 
-    def predict_proba(self, data, output_margin=False, ntree_limit=0):
+    def predict_proba(self, data, ntree_limit=0):
+        """
+        Predict the probability of each `data` example being of a given class.
+        NOTE: This function is not thread safe.
+              For each booster object, predict can only be called from one thread.
+              If you want to run prediction using multiple thread, call xgb.copy() to make copies
+              of model object and then call predict
+        Parameters
+        ----------
+        data : DMatrix
+            The dmatrix storing the input.
+        ntree_limit : int
+            Limit number of trees in the prediction; defaults to 0 (use all trees).
+        Returns
+        -------
+        prediction : numpy array
+            a numpy array with the probability of each data example being of a given class.
+        """
         test_dmatrix = DMatrix(data, missing=self.missing, nthread=self.n_jobs)
         class_probs = self.get_booster().predict(test_dmatrix,
-                                                 output_margin=output_margin,
                                                  ntree_limit=ntree_limit)
         if self.objective == "multi:softprob":
             return class_probs