Change version to 1.5.0. (#7258)

Co-authored-by: Philip Hyunsu Cho <chohyu01@cs.washington.edu>

.github/workflows/jvm_tests.yml

@@ -0,0 +1,74 @@
+name: XGBoost-JVM-Tests
+
+on: [push, pull_request]
+
+jobs:
+  test-with-jvm:
+    name: Test JVM on OS ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [windows-latest, ubuntu-latest]
+
+    steps:
+    - uses: actions/checkout@v2
+      with:
+        submodules: 'true'
+
+    - uses: actions/setup-python@v2
+      with:
+        python-version: '3.8'
+        architecture: 'x64'
+
+    - uses: actions/setup-java@v1
+      with:
+        java-version: 1.8
+
+    - name: Install Python packages
+      run: |
+        python -m pip install wheel setuptools
+        python -m pip install awscli
+
+    - name: Cache Maven packages
+      uses: actions/cache@v2
+      with:
+        path: ~/.m2
+        key: ${{ runner.os }}-m2-${{ hashFiles('./jvm-packages/pom.xml') }}
+        restore-keys: ${{ runner.os }}-m2
+
+    - name: Test XGBoost4J
+      run: |
+        cd jvm-packages
+        mvn test -B -pl :xgboost4j_2.12
+
+    - name: Extract branch name
+      shell: bash
+      run: echo "##[set-output name=branch;]$(echo ${GITHUB_REF#refs/heads/})"
+      id: extract_branch
+      if: |
+        (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
+        matrix.os == 'windows-latest'
+
+    - name: Publish artifact xgboost4j.dll to S3
+      run: |
+        cd lib/
+        Rename-Item -Path xgboost4j.dll -NewName xgboost4j_${{ github.sha }}.dll
+        dir
+        python -m awscli s3 cp xgboost4j_${{ github.sha }}.dll s3://xgboost-nightly-builds/${{ steps.extract_branch.outputs.branch }}/ --acl public-read
+      if: |
+        (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
+        matrix.os == 'windows-latest'
+      env:
+        AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID_IAM_S3_UPLOADER }}
+        AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY_IAM_S3_UPLOADER }}
+
+
+    - name: Test XGBoost4J-Spark
+      run: |
+        rm -rfv build/
+        cd jvm-packages
+        mvn -B test
+      if: matrix.os == 'ubuntu-latest'  # Distributed training doesn't work on Windows
+      env:
+        RABIT_MOCK: ON

.github/workflows/main.yml

@@ -21,17 +21,21 @@ jobs:
         submodules: 'true'
     - name: Install system packages
       run: |
-        brew install lz4 ninja libomp
+        # Use libomp 11.1.0: https://github.com/dmlc/xgboost/issues/7039
+        wget https://raw.githubusercontent.com/Homebrew/homebrew-core/679923b4eb48a8dc7ecc1f05d06063cd79b3fc00/Formula/libomp.rb -O $(find $(brew --repository) -name libomp.rb)
+        brew install ninja libomp
+        brew pin libomp
     - name: Build gtest binary
       run: |
         mkdir build
         cd build
-        cmake .. -DGOOGLE_TEST=ON -DUSE_OPENMP=ON -DUSE_DMLC_GTEST=ON -DPLUGIN_LZ4=ON -DPLUGIN_DENSE_PARSER=ON -GNinja
+        cmake .. -DGOOGLE_TEST=ON -DUSE_OPENMP=ON -DUSE_DMLC_GTEST=ON -DPLUGIN_DENSE_PARSER=ON -GNinja
         ninja -v
     - name: Run gtest binary
       run: |
         cd build
-        ctest --exclude-regex AllTestsInDMLCUnitTests --extra-verbose
+        ./testxgboost
+        ctest -R TestXGBoostCLI --extra-verbose
 
   gtest-cpu-nonomp:
     name: Test Google C++ unittest (CPU Non-OMP)
@@ -59,45 +63,6 @@ jobs:
         cd build
         ctest --extra-verbose
 
-  python-sdist-test:
-    name: Test installing XGBoost Python source package
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os: [ubuntu-latest, macos-10.15, windows-latest]
-        python-version: ["3.8"]
-    steps:
-    - uses: actions/checkout@v2
-      with:
-        submodules: 'true'
-    - name: Install osx system dependencies
-      if: matrix.os == 'macos-10.15'
-      run: |
-        brew install ninja libomp
-    - name: Install Ubuntu system dependencies
-      if: matrix.os == 'ubuntu-latest'
-      run: |
-        sudo apt-get install -y --no-install-recommends ninja-build
-    - uses: conda-incubator/setup-miniconda@v2
-      with:
-        auto-update-conda: true
-        python-version: ${{ matrix.python-version }}
-        activate-environment: test
-    - name: Display Conda env
-      shell: bash -l {0}
-      run: |
-        conda info
-        conda list
-    - name: Build and install XGBoost
-      shell: bash -l {0}
-      run: |
-        cd python-package
-        python --version
-        python setup.py sdist
-        pip install -v ./dist/xgboost-*.tar.gz
-        cd ..
-        python -c 'import xgboost'
-
   c-api-demo:
     name: Test installing XGBoost lib + building the C API demo
     runs-on: ${{ matrix.os }}
@@ -123,93 +88,47 @@ jobs:
       run: |
         conda info
         conda list
-    - name: Build and install XGBoost
+    - name: Build and install XGBoost static library
       shell: bash -l {0}
       run: |
         mkdir build
         cd build
         cmake .. -DBUILD_STATIC_LIB=ON -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX -GNinja
         ninja -v install
-    - name: Build and run C API demo
+        cd -
+    - name: Build and run C API demo with static
       shell: bash -l {0}
       run: |
+        pushd .
         cd demo/c-api/
         mkdir build
         cd build
         cmake .. -GNinja -DCMAKE_PREFIX_PATH=$CONDA_PREFIX
         ninja -v
+        ctest
         cd ..
-        ./build/api-demo
-
-  test-with-jvm:
-    name: Test JVM on OS ${{ matrix.os }}
-    runs-on: ${{ matrix.os }}
-    strategy:
-      fail-fast: false
-      matrix:
-        os: [windows-latest, ubuntu-latest]
-
-    steps:
-    - uses: actions/checkout@v2
-      with:
-        submodules: 'true'
-
-    - uses: actions/setup-python@v2
-      with:
-        python-version: '3.8'
-        architecture: 'x64'
-
-    - uses: actions/setup-java@v1
-      with:
-        java-version: 1.8
-
-    - name: Install Python packages
+        rm -rf ./build
+        popd
+    - name: Build and install XGBoost shared library
+      shell: bash -l {0}
       run: |
-        python -m pip install wheel setuptools
-        python -m pip install awscli
-
-    - name: Cache Maven packages
-      uses: actions/cache@v2
-      with:
-        path: ~/.m2
-        key: ${{ runner.os }}-m2-${{ hashFiles('./jvm-packages/pom.xml') }}
-        restore-keys: ${{ runner.os }}-m2
-
-    - name: Test XGBoost4J
+        cd build
+        cmake .. -DBUILD_STATIC_LIB=OFF -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX -GNinja
+        ninja -v install
+        cd -
+    - name: Build and run C API demo with shared
+      shell: bash -l {0}
       run: |
-        cd jvm-packages
-        mvn test -B -pl :xgboost4j_2.12
-
-    - name: Extract branch name
-      shell: bash
-      run: echo "##[set-output name=branch;]$(echo ${GITHUB_REF#refs/heads/})"
-      id: extract_branch
-      if: |
-        (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
-        matrix.os == 'windows-latest'
-
-    - name: Publish artifact xgboost4j.dll to S3
-      run: |
-        cd lib/
-        Rename-Item -Path xgboost4j.dll -NewName xgboost4j_${{ github.sha }}.dll
-        dir
-        python -m awscli s3 cp xgboost4j_${{ github.sha }}.dll s3://xgboost-nightly-builds/${{ steps.extract_branch.outputs.branch }}/ --acl public-read
-      if: |
-        (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
-        matrix.os == 'windows-latest'
-      env:
-        AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID_IAM_S3_UPLOADER }}
-        AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY_IAM_S3_UPLOADER }}
-
-
-    - name: Test XGBoost4J-Spark
-      run: |
-        rm -rfv build/
-        cd jvm-packages
-        mvn -B test
-      if: matrix.os == 'ubuntu-latest'  # Distributed training doesn't work on Windows
-      env:
-        RABIT_MOCK: ON
+        pushd .
+        cd demo/c-api/
+        mkdir build
+        cd build
+        cmake .. -GNinja -DCMAKE_PREFIX_PATH=$CONDA_PREFIX
+        ninja -v
+        ctest
+        popd
+        ./tests/ci_build/verify_link.sh ./demo/c-api/build/basic/api-demo
+        ./tests/ci_build/verify_link.sh ./demo/c-api/build/external-memory/external-memory-demo
 
   lint:
     runs-on: ubuntu-latest
@@ -243,7 +162,7 @@ jobs:
         architecture: 'x64'
     - name: Install Python packages
       run: |
-        python -m pip install wheel setuptools mypy dask[complete] distributed
+        python -m pip install wheel setuptools mypy pandas dask[complete] distributed
     - name: Run mypy
       run: |
         make mypy
@@ -279,7 +198,7 @@ jobs:
       run: |
         cd build/
         tar cvjf ${{ steps.extract_branch.outputs.branch }}.tar.bz2 doc_doxygen/
-        python -m awscli s3 cp ./${{ steps.extract_branch.outputs.branch }}.tar.bz2 s3://xgboost-docs/ --acl public-read
+        python -m awscli s3 cp ./${{ steps.extract_branch.outputs.branch }}.tar.bz2 s3://xgboost-docs/doxygen/ --acl public-read
       if: github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')
       env:
         AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID_IAM_S3_UPLOADER }}

.github/workflows/python_tests.yml

@@ -0,0 +1,94 @@
+name: XGBoost-Python-Tests
+
+on: [push, pull_request]
+
+jobs:
+  python-sdist-test:
+    runs-on: ${{ matrix.os }}
+    name: Test installing XGBoost Python source package on ${{ matrix.os }}
+    strategy:
+      matrix:
+        os: [ubuntu-latest, macos-10.15, windows-latest]
+        python-version: ["3.8"]
+    steps:
+    - uses: actions/checkout@v2
+      with:
+        submodules: 'true'
+    - name: Install osx system dependencies
+      if: matrix.os == 'macos-10.15'
+      run: |
+        # Use libomp 11.1.0: https://github.com/dmlc/xgboost/issues/7039
+        wget https://raw.githubusercontent.com/Homebrew/homebrew-core/679923b4eb48a8dc7ecc1f05d06063cd79b3fc00/Formula/libomp.rb -O $(find $(brew --repository) -name libomp.rb)
+        brew install ninja libomp
+        brew pin libomp
+    - name: Install Ubuntu system dependencies
+      if: matrix.os == 'ubuntu-latest'
+      run: |
+        sudo apt-get install -y --no-install-recommends ninja-build
+    - uses: conda-incubator/setup-miniconda@v2
+      with:
+        auto-update-conda: true
+        python-version: ${{ matrix.python-version }}
+        activate-environment: test
+    - name: Display Conda env
+      shell: bash -l {0}
+      run: |
+        conda info
+        conda list
+    - name: Build and install XGBoost
+      shell: bash -l {0}
+      run: |
+        cd python-package
+        python --version
+        python setup.py sdist
+        pip install -v ./dist/xgboost-*.tar.gz
+        cd ..
+        python -c 'import xgboost'
+
+  python-tests:
+    name: Test XGBoost Python package on ${{ matrix.config.os }}
+    runs-on: ${{ matrix.config.os }}
+    strategy:
+      matrix:
+        config:
+          - {os: windows-2016, compiler: 'msvc', python-version: '3.8'}
+
+    steps:
+    - uses: actions/checkout@v2
+      with:
+        submodules: 'true'
+
+    - uses: conda-incubator/setup-miniconda@v2
+      with:
+        auto-update-conda: true
+        python-version: ${{ matrix.config.python-version }}
+        activate-environment: win64_test
+        environment-file: tests/ci_build/conda_env/win64_cpu_test.yml
+
+    - name: Display Conda env
+      shell: bash -l {0}
+      run: |
+        conda info
+        conda list
+
+    - name: Build XGBoost with msvc
+      shell: bash -l {0}
+      if: matrix.config.compiler == 'msvc'
+      run: |
+        mkdir build_msvc
+        cd build_msvc
+        cmake .. -G"Visual Studio 15 2017" -DCMAKE_CONFIGURATION_TYPES="Release" -A x64 -DGOOGLE_TEST=ON  -DUSE_DMLC_GTEST=ON
+        cmake --build . --config Release --parallel $(nproc)
+
+    - name: Install Python package
+      shell: bash -l {0}
+      run: |
+        cd python-package
+        python --version
+        python setup.py bdist_wheel --universal
+        pip install ./dist/*.whl
+
+    - name: Test Python package
+      shell: bash -l {0}
+      run: |
+        pytest -s -v ./tests/python

.github/workflows/r_nold.yml

@@ -8,7 +8,7 @@ on:
     types: [created]
 
 env:
-  R_PACKAGES: c('XML', 'igraph', 'data.table', 'magrittr', 'ggplot2', 'DiagrammeR', 'Ckmeans.1d.dp', 'vcd', 'testthat', 'lintr', 'knitr', 'rmarkdown', 'e1071', 'cplm', 'devtools', 'float', 'titanic')
+  R_PACKAGES: c('XML', 'igraph', 'data.table', 'ggplot2', 'DiagrammeR', 'Ckmeans.1d.dp', 'vcd', 'testthat', 'lintr', 'knitr', 'rmarkdown', 'e1071', 'cplm', 'devtools', 'float', 'titanic')
 
 jobs:
   test-R-noLD:

.github/workflows/r_tests.yml

@@ -3,7 +3,8 @@ name: XGBoost-R-Tests
 on: [push, pull_request]
 
 env:
-  R_PACKAGES: c('XML', 'igraph', 'data.table', 'magrittr', 'ggplot2', 'DiagrammeR', 'Ckmeans.1d.dp', 'vcd', 'testthat', 'lintr', 'knitr', 'rmarkdown', 'e1071', 'cplm', 'devtools', 'float', 'titanic')
+  R_PACKAGES: c('XML', 'igraph', 'data.table', 'ggplot2', 'DiagrammeR', 'Ckmeans.1d.dp', 'vcd', 'testthat', 'lintr', 'knitr', 'rmarkdown', 'e1071', 'cplm', 'devtools', 'float', 'titanic')
+  GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
 
 jobs:
   lintr:
@@ -26,6 +27,13 @@ jobs:
       with:
         r-version: ${{ matrix.config.r }}
 
+    - name: Cache R packages
+      uses: actions/cache@v2
+      with:
+        path: ${{ env.R_LIBS_USER }}
+        key: ${{ runner.os }}-r-${{ matrix.config.r }}-2-${{ hashFiles('R-package/DESCRIPTION') }}
+        restore-keys: ${{ runner.os }}-r-${{ matrix.config.r }}-2-${{ hashFiles('R-package/DESCRIPTION') }}
+
     - name: Install dependencies
       shell: Rscript {0}
       run: |
@@ -62,6 +70,13 @@ jobs:
       with:
         r-version: ${{ matrix.config.r }}
 
+    - name: Cache R packages
+      uses: actions/cache@v2
+      with:
+        path: ${{ env.R_LIBS_USER }}
+        key: ${{ runner.os }}-r-${{ matrix.config.r }}-2-${{ hashFiles('R-package/DESCRIPTION') }}
+        restore-keys: ${{ runner.os }}-r-${{ matrix.config.r }}-2-${{ hashFiles('R-package/DESCRIPTION') }}
+
     - name: Install dependencies
       shell: Rscript {0}
       run: |
@@ -76,7 +91,7 @@ jobs:
 
     - name: Test R
       run: |
-        python tests/ci_build/test_r_package.py --compiler='${{ matrix.config.compiler }}' --build-tool='${{ matrix.config.build }}'
+        python tests/ci_build/test_r_package.py --compiler="${{ matrix.config.compiler }}" --build-tool="${{ matrix.config.build }}"
 
   test-R-CRAN:
     runs-on: ubuntu-latest
@@ -100,7 +115,14 @@ jobs:
 
     - name: Install system packages
       run: |
-        sudo apt-get update && sudo apt-get install libcurl4-openssl-dev libssl-dev libssh2-1-dev libgit2-dev
+        sudo apt-get update && sudo apt-get install libcurl4-openssl-dev libssl-dev libssh2-1-dev libgit2-dev pandoc pandoc-citeproc
+
+    - name: Cache R packages
+      uses: actions/cache@v2
+      with:
+        path: ${{ env.R_LIBS_USER }}
+        key: ${{ runner.os }}-r-${{ matrix.config.r }}-2-${{ hashFiles('R-package/DESCRIPTION') }}
+        restore-keys: ${{ runner.os }}-r-${{ matrix.config.r }}-2-${{ hashFiles('R-package/DESCRIPTION') }}
 
     - name: Install dependencies
       shell: Rscript {0}

.travis.yml

@@ -4,8 +4,9 @@ dist: bionic
 
 env:
   global:
-    - secure: "PR16i9F8QtNwn99C5NDp8nptAS+97xwDtXEJJfEiEVhxPaaRkOp0MPWhogCaK0Eclxk1TqkgWbdXFknwGycX620AzZWa/A1K3gAs+GrpzqhnPMuoBJ0Z9qxXTbSJvCyvMbYwVrjaxc/zWqdMU8waWz8A7iqKGKs/SqbQ3rO6v7c="
-    - secure: "dAGAjBokqm/0nVoLMofQni/fWIBcYSmdq4XvCBX1ZAMDsWnuOfz/4XCY6h2lEI1rVHZQ+UdZkc9PioOHGPZh5BnvE49/xVVWr9c4/61lrDOlkD01ZjSAeoV0fAZq+93V/wPl4QV+MM+Sem9hNNzFSbN5VsQLAiWCSapWsLdKzqA="
+    - secure: "lqkL5SCM/CBwgVb1GWoOngpojsa0zCSGcvF0O3/45rBT1EpNYtQ4LRJ1+XcHi126vdfGoim/8i7AQhn5eOgmZI8yAPBeoUZ5zSrejD3RUpXr2rXocsvRRP25Z4mIuAGHD9VAHtvTdhBZRVV818W02pYduSzAeaY61q/lU3xmWsE="
+    - secure: "mzms6X8uvdhRWxkPBMwx+mDl3d+V1kUpZa7UgjT+dr4rvZMzvKtjKp/O0JZZVogdgZjUZf444B98/7AvWdSkGdkfz2QdmhWmXzNPfNuHtmfCYMdijsgFIGLuD3GviFL/rBiM2vgn32T3QqFiEJiC5StparnnXimPTc9TpXQRq5c="
+
 
 jobs:
   include:
@@ -17,23 +18,17 @@ jobs:
       arch: amd64
       osx_image: xcode10.2
       env: TASK=java_test
+    - os: linux
+      arch: s390x
+      env: TASK=s390x_test
 
 # dependent brew packages
+# the dependencies from homebrew is installed manually from setup script due to outdated image from travis.
 addons:
   homebrew:
-    packages:
-      - cmake
-      - libomp
-      - graphviz
-      - openssl
-      - libgit2
-      - lz4
-      - wget
-      - r
-    update: true
+    update: false
   apt:
     packages:
-      - snapd
       - unzip
 
 before_install:

CMakeLists.txt

@@ -1,5 +1,5 @@
-cmake_minimum_required(VERSION 3.13)
-project(xgboost LANGUAGES CXX C VERSION 1.4.0)
+cmake_minimum_required(VERSION 3.14 FATAL_ERROR)
+project(xgboost LANGUAGES CXX C VERSION 1.5.0)
 include(cmake/Utils.cmake)
 list(APPEND CMAKE_MODULE_PATH "${xgboost_SOURCE_DIR}/cmake/modules")
 cmake_policy(SET CMP0022 NEW)
@@ -49,6 +49,7 @@ option(HIDE_CXX_SYMBOLS "Build shared library and hide all C++ symbols" OFF)
 option(USE_CUDA  "Build with GPU acceleration" OFF)
 option(USE_NCCL  "Build with NCCL to enable distributed GPU support." OFF)
 option(BUILD_WITH_SHARED_NCCL "Build with shared NCCL library." OFF)
+option(BUILD_WITH_CUDA_CUB "Build with cub in CUDA installation" OFF)
 set(GPU_COMPUTE_VER "" CACHE STRING
   "Semicolon separated list of compute versions to be built against, e.g. '35;61'")
 ## Copied From dmlc
@@ -62,7 +63,6 @@ set(ENABLED_SANITIZERS "address" "leak" CACHE STRING
   "Semicolon separated list of sanitizer names. E.g 'address;leak'. Supported sanitizers are
 address, leak, undefined and thread.")
 ## Plugins
-option(PLUGIN_LZ4 "Build lz4 plugin" OFF)
 option(PLUGIN_DENSE_PARSER "Build dense parser plugin" OFF)
 option(PLUGIN_RMM "Build with RAPIDS Memory Manager (RMM)" OFF)
 ## TODO: 1. Add check if DPC++ compiler is used for building
@@ -92,6 +92,9 @@ endif (R_LIB AND GOOGLE_TEST)
 if (USE_AVX)
   message(SEND_ERROR  "The option 'USE_AVX' is deprecated as experimental AVX features have been removed from XGBoost.")
 endif (USE_AVX)
+if (PLUGIN_LZ4)
+  message(SEND_ERROR  "The option 'PLUGIN_LZ4' is removed from XGBoost.")
+endif (PLUGIN_LZ4)
 if (PLUGIN_RMM AND NOT (USE_CUDA))
   message(SEND_ERROR "`PLUGIN_RMM` must be enabled with `USE_CUDA` flag.")
 endif (PLUGIN_RMM AND NOT (USE_CUDA))
@@ -109,6 +112,9 @@ endif (ENABLE_ALL_WARNINGS)
 if (BUILD_STATIC_LIB AND (R_LIB OR JVM_BINDINGS))
   message(SEND_ERROR "Cannot build a static library libxgboost.a when R or JVM packages are enabled.")
 endif (BUILD_STATIC_LIB AND (R_LIB OR JVM_BINDINGS))
+if (PLUGIN_RMM AND (NOT BUILD_WITH_CUDA_CUB))
+  message(SEND_ERROR "Cannot build with RMM using cub submodule.")
+endif (PLUGIN_RMM AND (NOT BUILD_WITH_CUDA_CUB))
 
 #-- Sanitizer
 if (USE_SANITIZER)
@@ -117,14 +123,14 @@ if (USE_SANITIZER)
 endif (USE_SANITIZER)
 
 if (USE_CUDA)
-  SET(USE_OPENMP ON CACHE BOOL "CUDA requires OpenMP" FORCE)
+  set(USE_OPENMP ON CACHE BOOL "CUDA requires OpenMP" FORCE)
   # `export CXX=' is ignored by CMake CUDA.
   set(CMAKE_CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER})
   message(STATUS "Configured CUDA host compiler: ${CMAKE_CUDA_HOST_COMPILER}")
 
   enable_language(CUDA)
-  if (${CMAKE_CUDA_COMPILER_VERSION} VERSION_LESS 10.0)
-    message(FATAL_ERROR "CUDA version must be at least 10.0!")
+  if (${CMAKE_CUDA_COMPILER_VERSION} VERSION_LESS 10.1)
+    message(FATAL_ERROR "CUDA version must be at least 10.1!")
   endif()
   set(GEN_CODE "")
   format_gencode_flags("${GPU_COMPUTE_VER}" GEN_CODE)
@@ -148,27 +154,26 @@ if (USE_OPENMP)
   find_package(OpenMP REQUIRED)
 endif (USE_OPENMP)
 
+if (USE_NCCL)
+  find_package(Nccl REQUIRED)
+endif (USE_NCCL)
+
 # dmlc-core
 msvc_use_static_runtime()
 add_subdirectory(${xgboost_SOURCE_DIR}/dmlc-core)
-set_target_properties(dmlc PROPERTIES
-  CXX_STANDARD 14
-  CXX_STANDARD_REQUIRED ON
-  POSITION_INDEPENDENT_CODE ON)
+
 if (MSVC)
-  target_compile_options(dmlc PRIVATE
-                         -D_CRT_SECURE_NO_WARNINGS -D_CRT_SECURE_NO_DEPRECATE)
   if (TARGET dmlc_unit_tests)
     target_compile_options(dmlc_unit_tests PRIVATE
                            -D_CRT_SECURE_NO_WARNINGS -D_CRT_SECURE_NO_DEPRECATE)
   endif (TARGET dmlc_unit_tests)
 endif (MSVC)
-if (ENABLE_ALL_WARNINGS)
-  target_compile_options(dmlc PRIVATE -Wall -Wextra)
-endif (ENABLE_ALL_WARNINGS)
 
 # rabit
 add_subdirectory(rabit)
+if (RABIT_BUILD_MPI)
+  find_package(MPI REQUIRED)
+endif (RABIT_BUILD_MPI)
 
 # core xgboost
 add_subdirectory(${xgboost_SOURCE_DIR}/src)
@@ -179,6 +184,11 @@ if (R_LIB)
   add_subdirectory(${xgboost_SOURCE_DIR}/R-package)
 endif (R_LIB)
 
+# This creates its own shared library `xgboost4j'.
+if (JVM_BINDINGS)
+  add_subdirectory(${xgboost_SOURCE_DIR}/jvm-packages)
+endif (JVM_BINDINGS)
+
 # Plugin
 add_subdirectory(${xgboost_SOURCE_DIR}/plugin)
 
@@ -189,48 +199,37 @@ else (BUILD_STATIC_LIB)
   add_library(xgboost SHARED)
 endif (BUILD_STATIC_LIB)
 target_link_libraries(xgboost PRIVATE objxgboost)
-
-if (USE_CUDA)
-  xgboost_set_cuda_flags(xgboost)
-endif (USE_CUDA)
-
-#-- Hide all C++ symbols
-if (HIDE_CXX_SYMBOLS)
-  foreach(target objxgboost xgboost dmlc)
-    set_target_properties(${target} PROPERTIES CXX_VISIBILITY_PRESET hidden)
-  endforeach()
-endif (HIDE_CXX_SYMBOLS)
-
 target_include_directories(xgboost
   INTERFACE
   $<INSTALL_INTERFACE:$<INSTALL_PREFIX>/include>
   $<BUILD_INTERFACE:${CMAKE_CURRENT_LIST_DIR}/include>)
-
-# This creates its own shared library `xgboost4j'.
-if (JVM_BINDINGS)
-  add_subdirectory(${xgboost_SOURCE_DIR}/jvm-packages)
-endif (JVM_BINDINGS)
 #-- End shared library
 
 #-- CLI for xgboost
 add_executable(runxgboost ${xgboost_SOURCE_DIR}/src/cli_main.cc)
 target_link_libraries(runxgboost PRIVATE objxgboost)
-if (USE_NVTX)
-  enable_nvtx(runxgboost)
-endif (USE_NVTX)
-
 target_include_directories(runxgboost
   PRIVATE
   ${xgboost_SOURCE_DIR}/include
   ${xgboost_SOURCE_DIR}/dmlc-core/include
-  ${xgboost_SOURCE_DIR}/rabit/include)
-set_target_properties(
-  runxgboost PROPERTIES
-  OUTPUT_NAME xgboost
-  CXX_STANDARD 14
-  CXX_STANDARD_REQUIRED ON)
+  ${xgboost_SOURCE_DIR}/rabit/include
+)
+set_target_properties(runxgboost PROPERTIES OUTPUT_NAME xgboost)
 #-- End CLI for xgboost
 
+# Common setup for all targets
+foreach(target xgboost objxgboost dmlc runxgboost)
+  xgboost_target_properties(${target})
+  xgboost_target_link_libraries(${target})
+  xgboost_target_defs(${target})
+endforeach()
+
+if (JVM_BINDINGS)
+  xgboost_target_properties(xgboost4j)
+  xgboost_target_link_libraries(xgboost4j)
+  xgboost_target_defs(xgboost4j)
+endif (JVM_BINDINGS)
+
 set_output_directory(runxgboost ${xgboost_SOURCE_DIR})
 set_output_directory(xgboost ${xgboost_SOURCE_DIR}/lib)
 # Ensure these two targets do not build simultaneously, as they produce outputs with conflicting names
@@ -255,6 +254,8 @@ if (BUILD_C_DOC)
   run_doxygen()
 endif (BUILD_C_DOC)
 
+include(CPack)
+
 include(GNUInstallDirs)
 # Install all headers.  Please note that currently the C++ headers does not form an "API".
 install(DIRECTORY ${xgboost_SOURCE_DIR}/include/xgboost
@@ -303,12 +304,18 @@ install(
 if (GOOGLE_TEST)
   enable_testing()
   # Unittests.
+  add_executable(testxgboost)
+  target_link_libraries(testxgboost PRIVATE objxgboost)
+  xgboost_target_properties(testxgboost)
+  xgboost_target_link_libraries(testxgboost)
+  xgboost_target_defs(testxgboost)
+
   add_subdirectory(${xgboost_SOURCE_DIR}/tests/cpp)
+
   add_test(
     NAME TestXGBoostLib
     COMMAND testxgboost
     WORKING_DIRECTORY ${xgboost_BINARY_DIR})
-
   # CLI tests
   configure_file(
     ${xgboost_SOURCE_DIR}/tests/cli/machine.conf.in

CONTRIBUTORS.md

@@ -43,7 +43,7 @@ Committers are people who have made substantial contribution to the project and
 
 Become a Committer
 ------------------
-XGBoost is a opensource project and we are actively looking for new committers who are willing to help maintaining and lead the project.
+XGBoost is a open source project and we are actively looking for new committers who are willing to help maintaining and lead the project.
 Committers comes from contributors who:
 * Made substantial contribution to the project.
 * Willing to spent time on maintaining and lead the project.
@@ -59,7 +59,7 @@ List of Contributors
 * [Skipper Seabold](https://github.com/jseabold)
   - Skipper is the major contributor to the scikit-learn module of XGBoost.
 * [Zygmunt Zając](https://github.com/zygmuntz)
-  - Zygmunt is the master behind the early stopping feature frequently used by kagglers.
+  - Zygmunt is the master behind the early stopping feature frequently used by Kagglers.
 * [Ajinkya Kale](https://github.com/ajkl)
 * [Boliang Chen](https://github.com/cblsjtu)
 * [Yangqing Men](https://github.com/yanqingmen)
@@ -91,7 +91,7 @@ List of Contributors
 * [Henry Gouk](https://github.com/henrygouk)
 * [Pierre de Sahb](https://github.com/pdesahb)
 * [liuliang01](https://github.com/liuliang01)
-  - liuliang01 added support for the qid column for LibSVM input format. This makes ranking task easier in distributed setting.
+  - liuliang01 added support for the qid column for LIBSVM input format. This makes ranking task easier in distributed setting.
 * [Andrew Thia](https://github.com/BlueTea88)
   - Andrew Thia implemented feature interaction constraints
 * [Wei Tian](https://github.com/weitian)

Jenkinsfile

@@ -7,7 +7,7 @@
 dockerRun = 'tests/ci_build/ci_build.sh'
 
 // Which CUDA version to use when building reference distribution wheel
-ref_cuda_ver = '10.0'
+ref_cuda_ver = '10.1'
 
 import groovy.transform.Field
 
@@ -58,14 +58,13 @@ pipeline {
             'build-cpu': { BuildCPU() },
             'build-cpu-arm64': { BuildCPUARM64() },
             'build-cpu-rabit-mock': { BuildCPUMock() },
-            // Build reference, distribution-ready Python wheel with CUDA 10.0
-            // using CentOS 6 image
-            'build-gpu-cuda10.0': { BuildCUDA(cuda_version: '10.0') },
-            // The build-gpu-* builds below use Ubuntu image
+            // Build reference, distribution-ready Python wheel with CUDA 10.1
+            // using CentOS 7 image
             'build-gpu-cuda10.1': { BuildCUDA(cuda_version: '10.1') },
-            'build-gpu-cuda10.2': { BuildCUDA(cuda_version: '10.2', build_rmm: true) },
-            'build-gpu-cuda11.0': { BuildCUDA(cuda_version: '11.0') },
-            'build-jvm-packages-gpu-cuda10.0': { BuildJVMPackagesWithCUDA(spark_version: '3.0.0', cuda_version: '10.0') },
+            // The build-gpu-* builds below use Ubuntu image
+            'build-gpu-cuda11.0': { BuildCUDA(cuda_version: '11.0', build_rmm: true) },
+            'build-gpu-rpkg': { BuildRPackageWithCUDA(cuda_version: '10.1') },
+            'build-jvm-packages-gpu-cuda10.1': { BuildJVMPackagesWithCUDA(spark_version: '3.0.0', cuda_version: '11.0') },
             'build-jvm-packages': { BuildJVMPackages(spark_version: '3.0.0') },
             'build-jvm-doc': { BuildJVMDoc() }
           ])
@@ -80,12 +79,10 @@ pipeline {
             'test-python-cpu': { TestPythonCPU() },
             'test-python-cpu-arm64': { TestPythonCPUARM64() },
             // artifact_cuda_version doesn't apply to RMM tests; RMM tests will always match CUDA version between artifact and host env
-            'test-python-gpu-cuda10.2': { TestPythonGPU(artifact_cuda_version: '10.0', host_cuda_version: '10.2', test_rmm: true) },
-            'test-python-gpu-cuda11.0-cross': { TestPythonGPU(artifact_cuda_version: '10.0', host_cuda_version: '11.0') },
+            'test-python-gpu-cuda11.0-cross': { TestPythonGPU(artifact_cuda_version: '10.1', host_cuda_version: '11.0', test_rmm: true) },
             'test-python-gpu-cuda11.0': { TestPythonGPU(artifact_cuda_version: '11.0', host_cuda_version: '11.0') },
-            'test-python-mgpu-cuda10.2': { TestPythonGPU(artifact_cuda_version: '10.0', host_cuda_version: '10.2', multi_gpu: true, test_rmm: true) },
-            'test-cpp-gpu-cuda10.2': { TestCppGPU(artifact_cuda_version: '10.2', host_cuda_version: '10.2', test_rmm: true) },
-            'test-cpp-gpu-cuda11.0': { TestCppGPU(artifact_cuda_version: '11.0', host_cuda_version: '11.0') },
+            'test-python-mgpu-cuda11.0': { TestPythonGPU(artifact_cuda_version: '10.1', host_cuda_version: '11.0', multi_gpu: true, test_rmm: true) },
+            'test-cpp-gpu-cuda11.0': { TestCppGPU(artifact_cuda_version: '11.0', host_cuda_version: '11.0', test_rmm: true) },
             'test-jvm-jdk8': { CrossTestJVMwithJDK(jdk_version: '8', spark_version: '3.0.0') },
             'test-jvm-jdk11': { CrossTestJVMwithJDK(jdk_version: '11') },
             'test-jvm-jdk12': { CrossTestJVMwithJDK(jdk_version: '12') }
@@ -122,7 +119,7 @@ def checkoutSrcs() {
 }
 
 def GetCUDABuildContainerType(cuda_version) {
-  return (cuda_version == ref_cuda_ver) ? 'gpu_build_centos6' : 'gpu_build'
+  return (cuda_version == ref_cuda_ver) ? 'gpu_build_centos7' : 'gpu_build'
 }
 
 def ClangTidy() {
@@ -150,7 +147,7 @@ def BuildCPU() {
       # This step is not necessary, but here we include it, to ensure that DMLC_CORE_USE_CMAKE flag is correctly propagated
       # We want to make sure that we use the configured header build/dmlc/build_config.h instead of include/dmlc/build_config_default.h.
       # See discussion at https://github.com/dmlc/xgboost/issues/5510
-    ${dockerRun} ${container_type} ${docker_binary} tests/ci_build/build_via_cmake.sh -DPLUGIN_LZ4=ON -DPLUGIN_DENSE_PARSER=ON
+    ${dockerRun} ${container_type} ${docker_binary} tests/ci_build/build_via_cmake.sh -DPLUGIN_DENSE_PARSER=ON
     ${dockerRun} ${container_type} ${docker_binary} bash -c "cd build && ctest --extra-verbose"
     """
     // Sanitizer test
@@ -178,10 +175,10 @@ def BuildCPUARM64() {
     ${dockerRun} ${container_type} ${docker_binary} bash -c "cd build && ctest --extra-verbose"
     ${dockerRun} ${container_type} ${docker_binary} bash -c "cd python-package && rm -rf dist/* && python setup.py bdist_wheel --universal"
     ${dockerRun} ${container_type} ${docker_binary} python tests/ci_build/rename_whl.py python-package/dist/*.whl ${commit_id} ${wheel_tag}
-    ${dockerRun} ${container_type} ${docker_binary} auditwheel repair --plat ${wheel_tag} python-package/dist/*.whl
+    ${dockerRun} ${container_type} ${docker_binary} bash -c "auditwheel repair --plat ${wheel_tag} python-package/dist/*.whl && python tests/ci_build/rename_whl.py wheelhouse/*.whl ${commit_id} ${wheel_tag}"
     mv -v wheelhouse/*.whl python-package/dist/
     # Make sure that libgomp.so is vendored in the wheel
-    ${dockerRun} ${container_type} ${docker_binary} bash -c "unzip -l python-package/dist/*.whl | grep libgomp  || exit -1"
+    ${dockerRun} ${container_type} ${docker_binary} bash -c "unzip -l python-package/dist/*.whl | grep libgomp || exit -1"
     """
     echo 'Stashing Python wheel...'
     stash name: "xgboost_whl_arm64_cpu", includes: 'python-package/dist/*.whl'
@@ -221,7 +218,7 @@ def BuildCUDA(args) {
     if (env.BRANCH_NAME != 'master' && !(env.BRANCH_NAME.startsWith('release'))) {
       arch_flag = "-DGPU_COMPUTE_VER=75"
     }
-    def wheel_tag = "manylinux2010_x86_64"
+    def wheel_tag = "manylinux2014_x86_64"
     sh """
     ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/build_via_cmake.sh -DUSE_CUDA=ON -DUSE_NCCL=ON -DOPEN_MP:BOOL=ON -DHIDE_CXX_SYMBOLS=ON ${arch_flag}
     ${dockerRun} ${container_type} ${docker_binary} ${docker_args} bash -c "cd python-package && rm -rf dist/* && python setup.py bdist_wheel --universal"
@@ -230,6 +227,7 @@ def BuildCUDA(args) {
     if (args.cuda_version == ref_cuda_ver) {
       sh """
       ${dockerRun} auditwheel_x86_64 ${docker_binary} auditwheel repair --plat ${wheel_tag} python-package/dist/*.whl
+      ${dockerRun} ${container_type} ${docker_binary} ${docker_args} python tests/ci_build/rename_whl.py wheelhouse/*.whl ${commit_id} ${wheel_tag}
       mv -v wheelhouse/*.whl python-package/dist/
       # Make sure that libgomp.so is vendored in the wheel
       ${dockerRun} auditwheel_x86_64 ${docker_binary} bash -c "unzip -l python-package/dist/*.whl | grep libgomp  || exit -1"
@@ -251,9 +249,9 @@ def BuildCUDA(args) {
       docker_args = "--build-arg CUDA_VERSION_ARG=${args.cuda_version}"
       sh """
       rm -rf build/
-      ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/build_via_cmake.sh --conda-env=gpu_test -DUSE_CUDA=ON -DUSE_NCCL=ON -DPLUGIN_RMM=ON ${arch_flag}
+      ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/build_via_cmake.sh --conda-env=gpu_test -DUSE_CUDA=ON -DUSE_NCCL=ON -DPLUGIN_RMM=ON -DBUILD_WITH_CUDA_CUB=ON ${arch_flag}
       ${dockerRun} ${container_type} ${docker_binary} ${docker_args} bash -c "cd python-package && rm -rf dist/* && python setup.py bdist_wheel --universal"
-      ${dockerRun} ${container_type} ${docker_binary} ${docker_args} python tests/ci_build/rename_whl.py python-package/dist/*.whl ${commit_id} manylinux2010_x86_64
+      ${dockerRun} ${container_type} ${docker_binary} ${docker_args} python tests/ci_build/rename_whl.py python-package/dist/*.whl ${commit_id} manylinux2014_x86_64
       """
       echo 'Stashing Python wheel...'
       stash name: "xgboost_whl_rmm_cuda${args.cuda_version}", includes: 'python-package/dist/*.whl'
@@ -264,6 +262,24 @@ def BuildCUDA(args) {
   }
 }
 
+def BuildRPackageWithCUDA(args) {
+  node('linux && cpu_build') {
+    unstash name: 'srcs'
+    def container_type = 'gpu_build_r_centos7'
+    def docker_binary = "docker"
+    def docker_args = "--build-arg CUDA_VERSION_ARG=${args.cuda_version}"
+    if (env.BRANCH_NAME == 'master' || env.BRANCH_NAME.startsWith('release')) {
+      sh """
+      ${dockerRun} ${container_type} ${docker_binary} ${docker_args} tests/ci_build/build_r_pkg_with_cuda.sh ${commit_id}
+      """
+      echo 'Uploading R tarball...'
+      path = ("${BRANCH_NAME}" == 'master') ? '' : "${BRANCH_NAME}/"
+      s3Upload bucket: 'xgboost-nightly-builds', path: path, acl: 'PublicRead', includePathPattern:'xgboost_r_gpu_linux_*.tar.gz'
+    }
+    deleteDir()
+  }
+}
+
 def BuildJVMPackagesWithCUDA(args) {
   node('linux && mgpu') {
     unstash name: 'srcs'
@@ -429,7 +445,7 @@ def DeployJVMPackages(args) {
     if (env.BRANCH_NAME == 'master' || env.BRANCH_NAME.startsWith('release')) {
       echo 'Deploying to xgboost-maven-repo S3 repo...'
       sh """
-      ${dockerRun} jvm_gpu_build docker --build-arg CUDA_VERSION_ARG=10.0 tests/ci_build/deploy_jvm_packages.sh ${args.spark_version}
+      ${dockerRun} jvm_gpu_build docker --build-arg CUDA_VERSION_ARG=10.1 tests/ci_build/deploy_jvm_packages.sh ${args.spark_version}
       """
     }
     deleteDir()

Jenkinsfile-win64

@@ -40,7 +40,8 @@ pipeline {
       steps {
         script {
           parallel ([
-            'build-win64-cuda10.1': { BuildWin64() }
+            'build-win64-cuda10.1': { BuildWin64() },
+            'build-rpkg-win64-cuda10.1': { BuildRPackageWithCUDAWin64() }
           ])
         }
       }
@@ -75,6 +76,7 @@ def checkoutSrcs() {
 
 def BuildWin64() {
   node('win64 && cuda10_unified') {
+    deleteDir()
     unstash name: 'srcs'
     echo "Building XGBoost for Windows AMD64 target..."
     bat "nvcc --version"
@@ -115,8 +117,26 @@ def BuildWin64() {
   }
 }
 
+def BuildRPackageWithCUDAWin64() {
+  node('win64 && cuda10_unified') {
+    deleteDir()
+    unstash name: 'srcs'
+    bat "nvcc --version"
+    if (env.BRANCH_NAME == 'master' || env.BRANCH_NAME.startsWith('release')) {
+      bat """
+      bash tests/ci_build/build_r_pkg_with_cuda_win64.sh ${commit_id}
+      """
+      echo 'Uploading R tarball...'
+      path = ("${BRANCH_NAME}" == 'master') ? '' : "${BRANCH_NAME}/"
+      s3Upload bucket: 'xgboost-nightly-builds', path: path, acl: 'PublicRead', includePathPattern:'xgboost_r_gpu_win64_*.tar.gz'
+    }
+    deleteDir()
+  }
+}
+
 def TestWin64() {
   node('win64 && cuda10_unified') {
+    deleteDir()
     unstash name: 'srcs'
     unstash name: 'xgboost_whl'
     unstash name: 'xgboost_cli'
@@ -127,7 +147,7 @@ def TestWin64() {
     bat "build\\testxgboost.exe"
     echo "Installing Python dependencies..."
     def env_name = 'win64_' + UUID.randomUUID().toString().replaceAll('-', '')
-    bat "conda env create -n ${env_name} --file=tests/ci_build/conda_env/win64_test.yml"
+    bat "conda activate && mamba env create -n ${env_name} --file=tests/ci_build/conda_env/win64_test.yml"
     echo "Installing Python wheel..."
     bat """
     conda activate ${env_name} && for /R %%i in (python-package\\dist\\*.whl) DO python -m pip install "%%i"

Makefile

@@ -91,8 +91,13 @@ endif
 # If any of the dask tests failed, contributor won't see the other error.
 mypy:
 	cd python-package; \
-	mypy ./xgboost/dask.py ../tests/python/test_with_dask.py --follow-imports=silent; \
-	mypy ../tests/python-gpu/test_gpu_with_dask.py --follow-imports=silent; \
+	mypy ./xgboost/dask.py && \
+	mypy ./xgboost/rabit.py && \
+	mypy ../demo/guide-python/external_memory.py && \
+	mypy ../tests/python-gpu/test_gpu_with_dask.py && \
+	mypy ../tests/python/test_data_iterator.py && \
+	mypy ../tests/python-gpu/test_gpu_data_iterator.py && \
+	mypy ./xgboost/sklearn.py || exit 1; \
 	mypy . || true ;
 
 clean:

NEWS.md

@@ -3,6 +3,247 @@ XGBoost Change Log
 
 This file records the changes in xgboost library in reverse chronological order.
 
+## v1.4.2 (2021.05.13)
+This is a patch release for Python package with following fixes:
+
+* Handle the latest version of cupy.ndarray in inplace_predict. (#6933)
+* Ensure output array from predict_leaf is (n_samples, ) when there's only 1 tree. 1.4.0 outputs (n_samples, 1). (#6889)
+* Fix empty dataset handling with multi-class AUC. (#6947)
+* Handle object type from pandas in inplace_predict. (#6927)
+
+
+## v1.4.1 (2021.04.20)
+This is a bug fix release.
+
+* Fix GPU implementation of AUC on some large datasets. (#6866)
+
+## v1.4.0 (2021.04.12)
+
+### Introduction of pre-built binary package for R, with GPU support
+Starting with release 1.4.0, users now have the option of installing `{xgboost}` without
+having to build it from the source. This is particularly advantageous for users who want
+to take advantage of the GPU algorithm (`gpu_hist`), as previously they'd have to build
+`{xgboost}` from the source using CMake and NVCC. Now installing `{xgboost}` with GPU
+support is as easy as: `R CMD INSTALL ./xgboost_r_gpu_linux.tar.gz`. (#6827)
+
+See the instructions at https://xgboost.readthedocs.io/en/latest/build.html
+
+### Improvements on prediction functions
+XGBoost has many prediction types including shap value computation and inplace prediction.
+In 1.4 we overhauled the underlying prediction functions for C API and Python API with an
+unified interface. (#6777, #6693, #6653, #6662, #6648, #6668, #6804)
+* Starting with 1.4, sklearn interface prediction will use inplace predict by default when
+  input data is supported.
+* Users can use inplace predict with `dart` booster and enable GPU acceleration just
+  like `gbtree`.
+* Also all prediction functions with tree models are now thread-safe.  Inplace predict is
+  improved with `base_margin` support.
+* A new set of C predict functions are exposed in the public interface.
+* A user-visible change is a newly added parameter called `strict_shape`.  See
+  https://xgboost.readthedocs.io/en/latest/prediction.html for more details.
+
+
+### Improvement on Dask interface
+* Starting with 1.4, the Dask interface is considered to be feature-complete, which means
+  all of the models found in the single node Python interface are now supported in Dask,
+  including but not limited to ranking and random forest.  Also, the prediction function
+  is significantly faster and supports shap value computation.
+  - Most of the parameters found in single node sklearn interface are supported by
+    Dask interface. (#6471, #6591)
+  - Implements learning to rank.  On the Dask interface, we use the newly added support of
+    query ID to enable group structure. (#6576)
+  - The Dask interface has Python type hints support. (#6519)
+  - All models can be safely pickled. (#6651)
+  - Random forest estimators are now supported. (#6602)
+  - Shap value computation is now supported. (#6575, #6645, #6614)
+  - Evaluation result is printed on the scheduler process. (#6609)
+  - `DaskDMatrix` (and device quantile dmatrix) now accepts all meta-information. (#6601)
+
+* Prediction optimization.  We enhanced and speeded up the prediction function for the
+  Dask interface.  See the latest Dask tutorial page in our document for an overview of
+  how you can optimize it even further. (#6650, #6645, #6648, #6668)
+
+* Bug fixes
+  - If you are using the latest Dask and distributed where `distributed.MultiLock` is
+    present, XGBoost supports training multiple models on the same cluster in
+    parallel. (#6743)
+  - A bug fix for when using `dask.client` to launch async task, XGBoost might use a
+    different client object internally. (#6722)
+
+* Other improvements on documents, blogs, tutorials, and demos. (#6389, #6366, #6687,
+  #6699, #6532, #6501)
+
+### Python package
+With changes from Dask and general improvement on prediction, we have made some
+enhancements on the general Python interface and IO for booster information.  Starting
+from 1.4, booster feature names and types can be saved into the JSON model.  Also some
+model attributes like `best_iteration`, `best_score` are restored upon model load.  On
+sklearn interface, some attributes are now implemented as Python object property with
+better documents.
+
+* Breaking change: All `data` parameters in prediction functions are renamed to `X`
+  for better compliance to sklearn estimator interface guidelines.
+* Breaking change: XGBoost used to generate some pseudo feature names with `DMatrix`
+  when inputs like `np.ndarray` don't have column names.  The procedure is removed to
+  avoid conflict with other inputs. (#6605)
+* Early stopping with training continuation is now supported. (#6506)
+* Optional import for Dask and cuDF are now lazy. (#6522)
+* As mentioned in the prediction improvement summary, the sklearn interface uses inplace
+  prediction whenever possible. (#6718)
+* Booster information like feature names and feature types are now saved into the JSON
+  model file. (#6605)
+* All `DMatrix` interfaces including `DeviceQuantileDMatrix` and counterparts in Dask
+  interface (as mentioned in the Dask changes summary) now accept all the meta-information
+  like `group` and `qid` in their constructor for better consistency. (#6601)
+* Booster attributes are restored upon model load so users don't have to call `attr`
+  manually. (#6593)
+* On sklearn interface, all models accept `base_margin` for evaluation datasets. (#6591)
+* Improvements over the setup script including smaller sdist size and faster installation
+  if the C++ library is already built (#6611, #6694, #6565).
+
+* Bug fixes for Python package:
+  - Don't validate feature when number of rows is 0. (#6472)
+  - Move metric configuration into booster. (#6504)
+  - Calling XGBModel.fit() should clear the Booster by default (#6562)
+  - Support `_estimator_type`. (#6582)
+  - [dask, sklearn] Fix predict proba. (#6566, #6817)
+  - Restore unknown data support. (#6595)
+  - Fix learning rate scheduler with cv. (#6720)
+  - Fixes small typo in sklearn documentation (#6717)
+  - [python-package] Fix class Booster: feature_types = None (#6705)
+  - Fix divide by 0 in feature importance when no split is found. (#6676)
+
+
+### JVM package
+* [jvm-packages] fix early stopping doesn't work even without custom_eval setting (#6738)
+* fix potential TaskFailedListener's callback won't be called (#6612)
+* [jvm] Add ability to load booster direct from byte array (#6655)
+* [jvm-packages] JVM library loader extensions (#6630)
+
+### R package
+* R documentation: Make construction of DMatrix consistent.
+* Fix R documentation for xgb.train. (#6764)
+
+### ROC-AUC
+We re-implemented the ROC-AUC metric in XGBoost.  The new implementation supports
+multi-class classification and has better support for learning to rank tasks that are not
+binary.  Also, it has a better-defined average on distributed environments with additional
+handling for invalid datasets. (#6749, #6747, #6797)
+
+### Global configuration.
+Starting from 1.4, XGBoost's Python, R and C interfaces support a new global configuration
+model where users can specify some global parameters.  Currently, supported parameters are
+`verbosity` and `use_rmm`.  The latter is experimental, see rmm plugin demo and
+related README file for details. (#6414, #6656)
+
+### Other New features.
+* Better handling for input data types that support `__array_interface__`.  For some
+  data types including GPU inputs and `scipy.sparse.csr_matrix`, XGBoost employs
+  `__array_interface__` for processing the underlying data.  Starting from 1.4, XGBoost
+  can accept arbitrary array strides (which means column-major is supported) without
+  making data copies, potentially reducing a significant amount of memory consumption.
+  Also version 3 of `__cuda_array_interface__` is now supported.  (#6776, #6765, #6459,
+  #6675)
+* Improved parameter validation, now feeding XGBoost with parameters that contain
+  whitespace will trigger an error. (#6769)
+* For Python and R packages, file paths containing the home indicator `~` are supported.
+* As mentioned in the Python changes summary, the JSON model can now save feature
+  information of the trained booster.  The JSON schema is updated accordingly. (#6605)
+* Development of categorical data support is continued.  Newly added weighted data support
+  and `dart` booster support. (#6508, #6693)
+* As mentioned in Dask change summary, ranking now supports the `qid` parameter for
+  query groups. (#6576)
+* `DMatrix.slice` can now consume a numpy array. (#6368)
+
+### Other breaking changes
+* Aside from the feature name generation, there are 2 breaking changes:
+  - Drop saving binary format for memory snapshot. (#6513, #6640)
+  - Change default evaluation metric for binary:logitraw objective to logloss (#6647)
+
+### CPU Optimization
+* Aside from the general changes on predict function, some optimizations are applied on
+  CPU implementation. (#6683, #6550, #6696, #6700)
+* Also performance for sampling initialization in `hist` is improved. (#6410)
+
+### Notable fixes in the core library
+These fixes do not reside in particular language bindings:
+* Fixes for gamma regression.  This includes checking for invalid input values, fixes for
+  gamma deviance metric, and better floating point guard for gamma negative log-likelihood
+  metric. (#6778, #6537, #6761)
+* Random forest with `gpu_hist` might generate low accuracy in previous versions. (#6755)
+* Fix a bug in GPU sketching when data size exceeds limit of 32-bit integer. (#6826)
+* Memory consumption fix for row-major adapters (#6779)
+* Don't estimate sketch batch size when rmm is used. (#6807) (#6830)
+* Fix in-place predict with missing value. (#6787)
+* Re-introduce double buffer in UpdatePosition, to fix perf regression in gpu_hist (#6757)
+* Pass correct split_type to GPU predictor (#6491)
+* Fix DMatrix feature names/types IO. (#6507)
+* Use view for `SparsePage` exclusively to avoid some data access races. (#6590)
+* Check for invalid data. (#6742)
+* Fix relocatable include in CMakeList (#6734) (#6737)
+* Fix DMatrix slice with feature types. (#6689)
+
+### Other deprecation notices:
+
+* This release will be the last release to support CUDA 10.0. (#6642)
+
+* Starting in the next release, the Python package will require Pip 19.3+ due to the use
+  of manylinux2014 tag. Also, CentOS 6, RHEL 6 and other old distributions will not be
+  supported.
+
+### Known issue:
+
+MacOS build of the JVM packages doesn't support multi-threading out of the box. To enable
+multi-threading with JVM packages, MacOS users will need to build the JVM packages from
+the source. See https://xgboost.readthedocs.io/en/latest/jvm/index.html#installation-from-source
+
+
+### Doc
+* Dedicated page for `tree_method` parameter is added. (#6564, #6633)
+* [doc] Add FLAML as a fast tuning tool for XGBoost  (#6770)
+* Add document for tests directory. [skip ci] (#6760)
+* Fix doc string of config.py to use correct `versionadded` (#6458)
+* Update demo for prediction. (#6789)
+* [Doc] Document that AUCPR is for binary classification/ranking (#5899)
+* Update the C API comments (#6457)
+* Fix document. [skip ci] (#6669)
+
+### Maintenance: Testing, continuous integration
+* Use CPU input for test_boost_from_prediction. (#6818)
+* [CI] Upload xgboost4j.dll to S3 (#6781)
+* Update dmlc-core submodule (#6745)
+* [CI] Use manylinux2010_x86_64 container to vendor libgomp (#6485)
+* Add conda-forge badge (#6502)
+* Fix merge conflict. (#6512)
+* [CI] Split up main.yml, add mypy. (#6515)
+* [Breaking] Upgrade cuDF and RMM to 0.18 nightlies; require RMM 0.18+ for RMM plugin (#6510)
+* "featue_map" typo changed to  "feature_map" (#6540)
+* Add script for generating release tarball. (#6544)
+* Add credentials to .gitignore (#6559)
+* Remove warnings in tests. (#6554)
+* Update dmlc-core submodule and conform to new API (#6431)
+* Suppress hypothesis health check for dask client. (#6589)
+* Fix pylint. (#6714)
+* [CI] Clear R package cache (#6746)
+* Exclude dmlc test on github action. (#6625)
+* Tests for regression metrics with weights. (#6729)
+* Add helper script and doc for releasing pip package. (#6613)
+* Support pylint 2.7.0 (#6726)
+* Remove R cache in github action. (#6695)
+* [CI] Do not mix up stashed executable built for ARM and x86_64 platforms (#6646)
+* [CI] Add ARM64 test to Jenkins pipeline (#6643)
+* Disable s390x and arm64 tests on travis for now. (#6641)
+* Move sdist test to action. (#6635)
+* [dask] Rework base margin test. (#6627)
+
+
+### Maintenance: Refactor code for legibility and maintainability
+* Improve OpenMP exception handling (#6680)
+* Improve string view to reduce string allocation. (#6644)
+* Simplify Span checks. (#6685)
+* Use generic dispatching routine for array interface. (#6672)
+
+
 ## v1.3.0 (2020.12.08)
 
 ### XGBoost4J-Spark: Exceptions should cancel jobs gracefully instead of killing SparkContext (#6019).
@@ -873,7 +1114,7 @@ This release marks a major milestone for the XGBoost project.
 * Specify version macro in CMake. (#4730)
 * Include dmlc-tracker into XGBoost Python package (#4731)
 * [CI] Use long key ID for Ubuntu repository fingerprints. (#4783)
-* Remove plugin, cuda related code in automake & autoconf files (#4789)
+* Remove plugin, CUDA related code in automake & autoconf files (#4789)
 * Skip related tests when scikit-learn is not installed. (#4791)
 * Ignore vscode and clion files (#4866)
 * Use bundled Google Test by default (#4900)
@@ -904,7 +1145,7 @@ This release marks a major milestone for the XGBoost project.
 ### Usability Improvements, Documentation
 * Add Random Forest API to Python API doc (#4500)
 * Fix Python demo and doc. (#4545)
-* Remove doc about not supporting cuda 10.1 (#4578)
+* Remove doc about not supporting CUDA 10.1 (#4578)
 * Address some sphinx warnings and errors, add doc for building doc. (#4589)
 * Add instruction to run formatting checks locally (#4591)
 * Fix docstring for `XGBModel.predict()` (#4592)
@@ -919,7 +1160,7 @@ This release marks a major milestone for the XGBoost project.
 * Update XGBoost4J-Spark doc (#4804)
 * Regular formatting for evaluation metrics (#4803)
 * [jvm-packages] Refine documentation for handling missing values in XGBoost4J-Spark (#4805)
-* Monitor for distributed envorinment (#4829). This is useful for identifying performance bottleneck.
+* Monitor for distributed environment (#4829). This is useful for identifying performance bottleneck.
 * Add check for length of weights and produce a good error message (#4872)
 * Fix DMatrix doc (#4884)
 * Export C++ headers in CMake installation (#4897)
@@ -1391,7 +1632,7 @@ This release is packed with many new features and bug fixes.
 ### Known issues
 * Quantile sketcher fails to produce any quantile for some edge cases (#2943)
 * The `hist` algorithm leaks memory when used with learning rate decay callback (#3579)
-* Using custom evaluation funciton together with early stopping causes assertion failure in XGBoost4J-Spark (#3595)
+* Using custom evaluation function together with early stopping causes assertion failure in XGBoost4J-Spark (#3595)
 * Early stopping doesn't work with `gblinear` learner (#3789)
 * Label and weight vectors are not reshared upon the change in number of GPUs (#3794). To get around this issue, delete the `DMatrix` object and re-load.
 * The `DMatrix` Python objects are initialized with incorrect values when given array slices (#3841)
@@ -1485,7 +1726,7 @@ This version is only applicable for the Python package. The content is identical
   - 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.
+  - `XGBClassifier.predict_proba()` no longer accepts parameter `output_margin`. The parameter 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.
@@ -1501,7 +1742,7 @@ This version is only applicable for the Python package. The content is identical
   - 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.
+    - Create an abstract 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:
@@ -1529,7 +1770,7 @@ This version is only applicable for the Python package. The content is identical
 * Refactored gbm to allow more friendly cache strategy
   - Specialized some prediction routine
 * Robust `DMatrix` construction from a sparse matrix
-* Faster consturction of `DMatrix` from 2D NumPy matrices: elide copies, use of multiple threads
+* Faster construction of `DMatrix` from 2D NumPy matrices: elide copies, use of multiple threads
 * Automatically remove nan from input data when it is sparse.
   - This can solve some of user reported problem of istart != hist.size
 * Fix the single-instance prediction function to obtain correct predictions
@@ -1557,7 +1798,7 @@ This version is only applicable for the Python package. The content is identical
   - Faster, histogram-based tree algorithm (`tree_method='hist'`) .
   - GPU/CUDA accelerated tree algorithms (`tree_method='gpu_hist'` or `'gpu_exact'`), including the GPU-based predictor.
   - Monotonic constraints: when other features are fixed, force the prediction to be monotonic increasing with respect to a certain specified feature.
-  - Faster gradient caculation using AVX SIMD
+  - Faster gradient calculation using AVX SIMD
   - Ability to export models in JSON format
   - Support for Tweedie regression
   - Additional dropout options for DART: binomial+1, epsilon

R-package/DESCRIPTION

@@ -1,8 +1,8 @@
 Package: xgboost
 Type: Package
 Title: Extreme Gradient Boosting
-Version: 1.4.0.1
-Date: 2020-08-28
+Version: 1.5.0.1
+Date: 2021-09-25
 Authors@R: c(
   person("Tianqi", "Chen", role = c("aut"),
          email = "tianqi.tchen@gmail.com"),
@@ -62,7 +62,6 @@ Imports:
     Matrix (>= 1.1-0),
     methods,
     data.table (>= 1.9.6),
-    magrittr (>= 1.5),
     jsonlite (>= 1.0),
 RoxygenNote: 7.1.1
 SystemRequirements: GNU make, C++14

R-package/NAMESPACE

@@ -82,7 +82,6 @@ importFrom(graphics,points)
 importFrom(graphics,title)
 importFrom(jsonlite,fromJSON)
 importFrom(jsonlite,toJSON)
-importFrom(magrittr,"%>%")
 importFrom(stats,median)
 importFrom(stats,predict)
 importFrom(utils,head)

R-package/R/callbacks.R

@@ -188,7 +188,7 @@ cb.reset.parameters <- function(new_params) {
   pnames <- gsub("\\.", "_", names(new_params))
   nrounds <- NULL
 
-  # run some checks in the begining
+  # run some checks in the beginning
   init <- function(env) {
     nrounds <<- env$end_iteration - env$begin_iteration + 1
 
@@ -263,10 +263,7 @@ cb.reset.parameters <- function(new_params) {
 #' \itemize{
 #' \item \code{best_score} the evaluation score at the best iteration
 #' \item \code{best_iteration} at which boosting iteration the best score has occurred (1-based index)
-#' \item \code{best_ntreelimit} to use with the \code{ntreelimit} parameter in \code{predict}.
-#'      It differs from \code{best_iteration} in multiclass or random forest settings.
 #' }
-#'
 #' The Same values are also stored as xgb-attributes:
 #' \itemize{
 #' \item \code{best_iteration} is stored as a 0-based iteration index (for interoperability of binary models)
@@ -498,13 +495,12 @@ cb.cv.predict <- function(save_models = FALSE) {
         rep(NA_real_, N)
       }
 
-    ntreelimit <- NVL(env$basket$best_ntreelimit,
-                      env$end_iteration * env$num_parallel_tree)
+    iterationrange <- c(1, NVL(env$basket$best_iteration, env$end_iteration) + 1)
     if (NVL(env$params[['booster']], '') == 'gblinear') {
-      ntreelimit <- 0 # must be 0 for gblinear
+      iterationrange <- c(1, 1)  # must be 0 for gblinear
     }
     for (fd in env$bst_folds) {
-      pr <- predict(fd$bst, fd$watchlist[[2]], ntreelimit = ntreelimit, reshape = TRUE)
+      pr <- predict(fd$bst, fd$watchlist[[2]], iterationrange = iterationrange, reshape = TRUE)
       if (is.matrix(pred)) {
         pred[fd$index, ] <- pr
       } else {
@@ -533,7 +529,7 @@ 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.
+#' @param sparse when set to FALSE/TRUE, 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.
@@ -560,7 +556,6 @@ cb.cv.predict <- function(save_models = FALSE) {
 #' #
 #' # 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"))
@@ -581,7 +576,7 @@ cb.cv.predict <- function(save_models = FALSE) {
 #' 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')
+#' matplot(xgb.gblinear.history(bst), 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.
@@ -590,7 +585,7 @@ cb.cv.predict <- function(save_models = FALSE) {
 #' 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')
+#' matplot(xgb.gblinear.history(bst)[[3]], type = 'l')
 #'
 #'
 #' #### Multiclass classification:
@@ -603,15 +598,15 @@ cb.cv.predict <- function(save_models = FALSE) {
 #' 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')
+#' matplot(xgb.gblinear.history(bst, class_index = 0), type = 'l')
+#' matplot(xgb.gblinear.history(bst, class_index = 1), type = 'l')
+#' matplot(xgb.gblinear.history(bst, class_index = 2), 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')
+#' # 1st fold of 1st class
+#' matplot(xgb.gblinear.history(bst, class_index = 0)[[1]], type = 'l')
 #'
 #' @export
 cb.gblinear.history <- function(sparse=FALSE) {
@@ -642,9 +637,14 @@ cb.gblinear.history <- function(sparse=FALSE) {
     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))
+      # second lapply transposes the list
+      coefs <<- lapply(
+        X = lapply(
+          X = seq_along(coefs[[1]]),
+          FUN = function(i) lapply(coefs, "[[", i)
+        ),
+        FUN = list2mat
+      )
     }
   }
 

R-package/R/utils.R

@@ -1,6 +1,6 @@
 #
-# This file is for the low level reuseable utility functions
-# that are not supposed to be visibe to a user.
+# This file is for the low level reusable utility functions
+# that are not supposed to be visible to a user.
 #
 
 #
@@ -178,7 +178,8 @@ xgb.iter.eval <- function(booster_handle, watchlist, iter, feval = NULL) {
   } else {
     res <- sapply(seq_along(watchlist), function(j) {
       w <- watchlist[[j]]
-      preds <- predict(booster_handle, w, outputmargin = TRUE, ntreelimit = 0) # predict using all trees
+      ## predict using all trees
+      preds <- predict(booster_handle, w, outputmargin = TRUE, iterationrange = c(1, 1))
       eval_res <- feval(preds, w)
       out <- eval_res$value
       names(out) <- paste0(evnames[j], "-", eval_res$metric)
@@ -284,7 +285,7 @@ xgb.createFolds <- function(y, k = 10)
     for (i in seq_along(numInClass)) {
       ## create a vector of integers from 1:k as many times as possible without
       ## going over the number of samples in the class. Note that if the number
-      ## of samples in a class is less than k, nothing is producd here.
+      ## of samples in a class is less than k, nothing is produced here.
       seqVector <- rep(seq_len(k), numInClass[i] %/% k)
       ## add enough random integers to get  length(seqVector) == numInClass[i]
       if (numInClass[i] %% k > 0) seqVector <- c(seqVector, sample.int(k, numInClass[i] %% k))

R-package/R/xgb.Booster.R

@@ -1,7 +1,7 @@
 # Construct an internal xgboost Booster and return a handle to it.
 # internal utility function
 xgb.Booster.handle <- function(params = list(), cachelist = list(),
-                               modelfile = NULL) {
+                               modelfile = NULL, handle = NULL) {
   if (typeof(cachelist) != "list" ||
       !all(vapply(cachelist, inherits, logical(1), what = 'xgb.DMatrix'))) {
     stop("cachelist must be a list of xgb.DMatrix objects")
@@ -20,7 +20,7 @@ xgb.Booster.handle <- function(params = list(), cachelist = list(),
       return(handle)
     } else if (typeof(modelfile) == "raw") {
       ## A memory buffer
-      bst <- xgb.unserialize(modelfile)
+      bst <- xgb.unserialize(modelfile, handle)
       xgb.parameters(bst) <- params
       return (bst)
     } else if (inherits(modelfile, "xgb.Booster")) {
@@ -129,7 +129,7 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
     stop("argument type must be xgb.Booster")
 
   if (is.null.handle(object$handle)) {
-    object$handle <- xgb.Booster.handle(modelfile = object$raw)
+    object$handle <- xgb.Booster.handle(modelfile = object$raw, handle = object$handle)
   } else {
     if (is.null(object$raw) && saveraw) {
       object$raw <- xgb.serialize(object$handle)
@@ -168,8 +168,7 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' @param outputmargin whether the prediction should be returned in the for of original untransformed
 #'        sum of predictions from boosting iterations' results. E.g., setting \code{outputmargin=TRUE} for
 #'        logistic regression would result in predictions for log-odds instead of probabilities.
-#' @param ntreelimit limit the number of model's trees or boosting iterations used in prediction (see Details).
-#'        It will use all the trees by default (\code{NULL} value).
+#' @param ntreelimit Deprecated, use \code{iterationrange} instead.
 #' @param predleaf whether predict leaf index.
 #' @param predcontrib whether to return feature contributions to individual predictions (see Details).
 #' @param approxcontrib whether to use a fast approximation for feature contributions (see Details).
@@ -179,16 +178,19 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #'        or predinteraction flags is TRUE.
 #' @param training whether is the prediction result used for training.  For dart booster,
 #'        training predicting will perform dropout.
+#' @param iterationrange Specifies which layer of trees are used in prediction.  For
+#'        example, if a random forest is trained with 100 rounds.  Specifying
+#'        `iteration_range=(1, 21)`, then only the forests built during [1, 21) (half open set)
+#'        rounds are used in this prediction.  It's 1-based index just like R vector.  When set
+#'        to \code{c(1, 1)} XGBoost will use all trees.
+#' @param strict_shape  Default is \code{FALSE}. When it's set to \code{TRUE}, output
+#'        type and shape of prediction are invariant to model type.
+#'
 #' @param ... Parameters passed to \code{predict.xgb.Booster}
 #'
 #' @details
-#' Note that \code{ntreelimit} is not necessarily equal to the number of boosting iterations
-#' and it is not necessarily equal to the number of trees in a model.
-#' E.g., in a random forest-like model, \code{ntreelimit} would limit the number of trees.
-#' But for multiclass classification, while there are multiple trees per iteration,
-#' \code{ntreelimit} limits the number of boosting iterations.
 #'
-#' Also note that \code{ntreelimit} would currently do nothing for predictions from gblinear,
+#' Note that \code{iterationrange} would currently do nothing for predictions from gblinear,
 #' since gblinear doesn't keep its boosting history.
 #'
 #' One possible practical applications of the \code{predleaf} option is to use the model
@@ -209,7 +211,8 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' of the most important features first. See below about the format of the returned results.
 #'
 #' @return
-#' For regression or binary classification, it returns a vector of length \code{nrows(newdata)}.
+#' The return type is different depending whether \code{strict_shape} is set to \code{TRUE}.  By default,
+#' for regression or binary classification, it returns a vector of length \code{nrows(newdata)}.
 #' For multiclass classification, either a \code{num_class * nrows(newdata)} vector or
 #' a \code{(nrows(newdata), num_class)} dimension matrix is returned, depending on
 #' the \code{reshape} value.
@@ -231,6 +234,13 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' For a multiclass case, a list of \code{num_class} elements is returned, where each element is
 #' such an array.
 #'
+#' When \code{strict_shape} is set to \code{TRUE}, the output is always an array.  For
+#' normal prediction, the output is a 2-dimension array \code{(num_class, nrow(newdata))}.
+#'
+#' For \code{predcontrib = TRUE}, output is \code{(ncol(newdata) + 1, num_class, nrow(newdata))}
+#' For \code{predinteraction = TRUE}, output is \code{(ncol(newdata) + 1, ncol(newdata) + 1, num_class, nrow(newdata))}
+#' For \code{predleaf = TRUE}, output is \code{(n_trees_in_forest, num_class, n_iterations, nrow(newdata))}
+#'
 #' @seealso
 #' \code{\link{xgb.train}}.
 #'
@@ -253,7 +263,7 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' # use all trees by default
 #' pred <- predict(bst, test$data)
 #' # use only the 1st tree
-#' pred1 <- predict(bst, test$data, ntreelimit = 1)
+#' pred1 <- predict(bst, test$data, iterationrange = c(1, 2))
 #'
 #' # Predicting tree leafs:
 #' # the result is an nsamples X ntrees matrix
@@ -305,31 +315,14 @@ xgb.Booster.complete <- function(object, saveraw = TRUE) {
 #' all.equal(pred, pred_labels)
 #' # prediction from using only 5 iterations should result
 #' # in the same error as seen in iteration 5:
-#' pred5 <- predict(bst, as.matrix(iris[, -5]), ntreelimit=5)
+#' pred5 <- predict(bst, as.matrix(iris[, -5]), iterationrange=c(1, 6))
 #' sum(pred5 != lb)/length(lb)
 #'
-#'
-#' ## random forest-like model of 25 trees for binary classification:
-#'
-#' set.seed(11)
-#' bst <- xgboost(data = train$data, label = train$label, max_depth = 5,
-#'                nthread = 2, nrounds = 1, objective = "binary:logistic",
-#'                num_parallel_tree = 25, subsample = 0.6, colsample_bytree = 0.1)
-#' # Inspect the prediction error vs number of trees:
-#' lb <- test$label
-#' dtest <- xgb.DMatrix(test$data, label=lb)
-#' err <- sapply(1:25, function(n) {
-#'   pred <- predict(bst, dtest, ntreelimit=n)
-#'   sum((pred > 0.5) != lb)/length(lb)
-#' })
-#' plot(err, type='l', ylim=c(0,0.1), xlab='#trees')
-#'
 #' @rdname predict.xgb.Booster
 #' @export
 predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FALSE, ntreelimit = NULL,
                                 predleaf = FALSE, predcontrib = FALSE, approxcontrib = FALSE, predinteraction = FALSE,
-                                reshape = FALSE, training = FALSE, ...) {
-
+                                reshape = FALSE, training = FALSE, iterationrange = NULL, strict_shape = FALSE, ...) {
   object <- xgb.Booster.complete(object, saveraw = FALSE)
   if (!inherits(newdata, "xgb.DMatrix"))
     newdata <- xgb.DMatrix(newdata, missing = missing)
@@ -337,62 +330,114 @@ predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FA
       !is.null(colnames(newdata)) &&
       !identical(object[["feature_names"]], colnames(newdata)))
     stop("Feature names stored in `object` and `newdata` are different!")
-  if (is.null(ntreelimit))
-    ntreelimit <- NVL(object$best_ntreelimit, 0)
-  if (NVL(object$params[['booster']], '') == 'gblinear')
+
+  if (NVL(object$params[['booster']], '') == 'gblinear' || is.null(ntreelimit))
     ntreelimit <- 0
-  if (ntreelimit < 0)
-    stop("ntreelimit cannot be negative")
 
-  option <- 0L + 1L * as.logical(outputmargin) + 2L * as.logical(predleaf) + 4L * as.logical(predcontrib) +
-    8L * as.logical(approxcontrib) + 16L * as.logical(predinteraction)
-
-  ret <- .Call(XGBoosterPredict_R, object$handle, newdata, option[1],
-               as.integer(ntreelimit), as.integer(training))
-
-  n_ret <- length(ret)
-  n_row <- nrow(newdata)
-  npred_per_case <- n_ret / n_row
-
-  if (n_ret %% n_row != 0)
-    stop("prediction length ", n_ret, " is not multiple of nrows(newdata) ", n_row)
-
-  if (predleaf) {
-    ret <- if (n_ret == n_row) {
-      matrix(ret, ncol = 1)
+  if (ntreelimit != 0 && is.null(iterationrange)) {
+    ## only ntreelimit, initialize iteration range
+    iterationrange <- c(0, 0)
+  } else if (ntreelimit == 0 && !is.null(iterationrange)) {
+    ## only iteration range, handle 1-based indexing
+    iterationrange <- c(iterationrange[1] - 1, iterationrange[2] - 1)
+  } else if (ntreelimit != 0 && !is.null(iterationrange)) {
+    ## both are specified, let libgxgboost throw an error
+  } else {
+    ## no limit is supplied, use best
+    if (is.null(object$best_iteration)) {
+      iterationrange <- c(0, 0)
     } else {
-      matrix(ret, nrow = n_row, byrow = TRUE)
+      ## We don't need to + 1 as R is 1-based index.
+      iterationrange <- c(0, as.integer(object$best_iteration))
     }
-  } else if (predcontrib) {
-    n_col1 <- ncol(newdata) + 1
-    n_group <- npred_per_case / n_col1
-    cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "BIAS") else NULL
-    ret <- if (n_ret == n_row) {
-      matrix(ret, ncol = 1, dimnames = list(NULL, cnames))
-    } else if (n_group == 1) {
-      matrix(ret, nrow = n_row, byrow = TRUE, dimnames = list(NULL, cnames))
-    } else {
-      arr <- array(ret, c(n_col1, n_group, n_row),
-                   dimnames = list(cnames, NULL, NULL)) %>% aperm(c(2, 3, 1)) # [group, row, col]
-      lapply(seq_len(n_group), function(g) arr[g, , ])
+  }
+  ## Handle the 0 length values.
+  box <- function(val) {
+    if (length(val) == 0) {
+      cval <- vector(, 1)
+      cval[0] <- val
+      return(cval)
+    }
+    return (val)
+  }
+
+  ## We set strict_shape to TRUE then drop the dimensions conditionally
+  args <- list(
+    training = box(training),
+    strict_shape = box(TRUE),
+    iteration_begin = box(as.integer(iterationrange[1])),
+    iteration_end = box(as.integer(iterationrange[2])),
+    ntree_limit = box(as.integer(ntreelimit)),
+    type = box(as.integer(0))
+  )
+
+  set_type <- function(type) {
+    if (args$type != 0) {
+      stop("One type of prediction at a time.")
+    }
+    return(box(as.integer(type)))
+  }
+  if (outputmargin) {
+    args$type <- set_type(1)
+  }
+  if (predcontrib) {
+    args$type <- set_type(if (approxcontrib) 3 else 2)
+  }
+  if (predinteraction) {
+    args$type <- set_type(if (approxcontrib) 5 else 4)
+  }
+  if (predleaf) {
+    args$type <- set_type(6)
+  }
+
+  predts <- .Call(
+    XGBoosterPredictFromDMatrix_R, object$handle, newdata, jsonlite::toJSON(args, auto_unbox = TRUE)
+  )
+  names(predts) <- c("shape", "results")
+  shape <- predts$shape
+  ret <- predts$results
+
+  n_row <- nrow(newdata)
+  if (n_row != shape[1]) {
+    stop("Incorrect predict shape.")
+  }
+
+  arr <- array(data = ret, dim = rev(shape))
+
+  cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "BIAS") else NULL
+  if (predcontrib) {
+    dimnames(arr) <- list(cnames, NULL, NULL)
+    if (!strict_shape) {
+      arr <- aperm(a = arr, perm = c(2, 3, 1)) # [group, row, col]
     }
   } else if (predinteraction) {
-    n_col1 <- ncol(newdata) + 1
-    n_group <- npred_per_case / n_col1^2
-    cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "BIAS") else NULL
-    ret <- if (n_ret == n_row) {
-      matrix(ret, ncol = 1, dimnames = list(NULL, cnames))
-    } else if (n_group == 1) {
-      array(ret, c(n_col1, n_col1, n_row), dimnames = list(cnames, cnames, NULL)) %>% aperm(c(3, 1, 2))
-    } else {
-      arr <- array(ret, c(n_col1, n_col1, n_group, n_row),
-                   dimnames = list(cnames, cnames, NULL, NULL)) %>% aperm(c(3, 4, 1, 2)) # [group, row, col1, col2]
-      lapply(seq_len(n_group), function(g) arr[g, , , ])
+    dimnames(arr) <- list(cnames, cnames, NULL, NULL)
+    if (!strict_shape) {
+      arr <- aperm(a = arr, perm = c(3, 4, 1, 2)) # [group, row, col, col]
     }
-  } else if (reshape && npred_per_case > 1) {
-    ret <- matrix(ret, nrow = n_row, byrow = TRUE)
   }
-  return(ret)
+
+  if (!strict_shape) {
+    n_groups <- shape[2]
+    if (predleaf) {
+      arr <- matrix(arr, nrow = n_row, byrow = TRUE)
+    } else if (predcontrib && n_groups != 1) {
+      arr <- lapply(seq_len(n_groups), function(g) arr[g, , ])
+    } else if (predinteraction && n_groups != 1) {
+      arr <- lapply(seq_len(n_groups), function(g) arr[g, , , ])
+    } else if (!reshape && n_groups != 1) {
+      arr <- ret
+    } else if (reshape && n_groups != 1) {
+      arr <- matrix(arr, ncol = n_groups, byrow = TRUE)
+    }
+    arr <- drop(arr)
+    if (length(dim(arr)) == 1) {
+      arr <- as.vector(arr)
+    } else if (length(dim(arr)) == 2) {
+      arr <- as.matrix(arr)
+    }
+  }
+  return(arr)
 }
 
 #' @rdname predict.xgb.Booster

R-package/R/xgb.DMatrix.R

@@ -1,7 +1,7 @@
 #' Construct xgb.DMatrix object
 #'
 #' Construct xgb.DMatrix object from either a dense matrix, a sparse matrix, or a local file.
-#' Supported input file formats are either a libsvm text file or a binary file that was created previously by
+#' Supported input file formats are either a LIBSVM text file or a binary file that was created previously by
 #' \code{\link{xgb.DMatrix.save}}).
 #'
 #' @param data a \code{matrix} object (either numeric or integer), a \code{dgCMatrix} object, or a character
@@ -20,7 +20,7 @@
 #' dtrain <- xgb.DMatrix('xgb.DMatrix.data')
 #' if (file.exists('xgb.DMatrix.data')) file.remove('xgb.DMatrix.data')
 #' @export
-xgb.DMatrix <- function(data, info = list(), missing = NA, silent = FALSE, ...) {
+xgb.DMatrix <- function(data, info = list(), missing = NA, silent = FALSE, nthread = NULL, ...) {
   cnames <- NULL
   if (typeof(data) == "character") {
     if (length(data) > 1)
@@ -29,7 +29,7 @@ xgb.DMatrix <- function(data, info = list(), missing = NA, silent = FALSE, ...)
     data <- path.expand(data)
     handle <- .Call(XGDMatrixCreateFromFile_R, data, as.integer(silent))
   } else if (is.matrix(data)) {
-    handle <- .Call(XGDMatrixCreateFromMat_R, data, missing)
+    handle <- .Call(XGDMatrixCreateFromMat_R, data, missing, as.integer(NVL(nthread, -1)))
     cnames <- colnames(data)
   } else if (inherits(data, "dgCMatrix")) {
     handle <- .Call(XGDMatrixCreateFromCSC_R, data@p, data@i, data@x, nrow(data))
@@ -51,12 +51,12 @@ xgb.DMatrix <- function(data, info = list(), missing = NA, silent = FALSE, ...)
 
 # get dmatrix from data, label
 # internal helper method
-xgb.get.DMatrix <- function(data, label = NULL, missing = NA, weight = NULL) {
+xgb.get.DMatrix <- function(data, label = NULL, missing = NA, weight = NULL, nthread = NULL) {
   if (inherits(data, "dgCMatrix") || is.matrix(data)) {
     if (is.null(label)) {
       stop("label must be provided when data is a matrix")
     }
-    dtrain <- xgb.DMatrix(data, label = label, missing = missing)
+    dtrain <- xgb.DMatrix(data, label = label, missing = missing, nthread = nthread)
     if (!is.null(weight)){
       setinfo(dtrain, "weight", weight)
     }
@@ -161,9 +161,9 @@ dimnames.xgb.DMatrix <- function(x) {
 #' The \code{name} field can be one of the following:
 #'
 #' \itemize{
-#'     \item \code{label}: label Xgboost learn from ;
+#'     \item \code{label}: label XGBoost learn from ;
 #'     \item \code{weight}: to do a weight rescale ;
-#'     \item \code{base_margin}: base margin is the base prediction Xgboost will boost from ;
+#'     \item \code{base_margin}: base margin is the base prediction XGBoost will boost from ;
 #'     \item \code{nrow}: number of rows of the \code{xgb.DMatrix}.
 #'
 #' }
@@ -216,9 +216,9 @@ getinfo.xgb.DMatrix <- function(object, name, ...) {
 #' The \code{name} field can be one of the following:
 #'
 #' \itemize{
-#'     \item \code{label}: label Xgboost learn from ;
+#'     \item \code{label}: label XGBoost learn from ;
 #'     \item \code{weight}: to do a weight rescale ;
-#'     \item \code{base_margin}: base margin is the base prediction Xgboost will boost from ;
+#'     \item \code{base_margin}: base margin is the base prediction XGBoost will boost from ;
 #'     \item \code{group}: number of rows in each group (to use with \code{rank:pairwise} objective).
 #' }
 #'

R-package/R/xgb.cv.R

@@ -101,9 +101,7 @@
 #'         parameter or randomly generated.
 #'   \item \code{best_iteration} iteration number with the best evaluation metric value
 #'         (only available with early stopping).
-#'   \item \code{best_ntreelimit} the \code{ntreelimit} value corresponding to the best iteration,
-#'         which could further be used in \code{predict} method
-#'         (only available with early stopping).
+#'   \item \code{best_ntreelimit} and the \code{ntreelimit} Deprecated attributes, use \code{best_iteration} instead.
 #'   \item \code{pred} CV prediction values available when \code{prediction} is set.
 #'         It is either vector or matrix (see \code{\link{cb.cv.predict}}).
 #'   \item \code{models} a list of the CV folds' models. It is only available with the explicit

R-package/R/xgb.importance.R

@@ -96,40 +96,44 @@ xgb.importance <- function(feature_names = NULL, model = NULL, trees = NULL,
   if (!(is.null(feature_names) || is.character(feature_names)))
     stop("feature_names: Has to be a character vector")
 
-  model_text_dump <- xgb.dump(model = model, with_stats = TRUE)
-
-  # linear model
-  if (model_text_dump[2] == "bias:"){
-    weights <- which(model_text_dump == "weight:") %>%
-               {model_text_dump[(. + 1):length(model_text_dump)]} %>%
-               as.numeric
-
-    num_class <- NVL(model$params$num_class, 1)
-    if (is.null(feature_names))
-      feature_names <- seq(to = length(weights) / num_class) - 1
-    if (length(feature_names) * num_class != length(weights))
-      stop("feature_names length does not match the number of features used in the model")
-
-    result <- if (num_class == 1) {
-      data.table(Feature = feature_names, Weight = weights)[order(-abs(Weight))]
+  model <- xgb.Booster.complete(model)
+  config <- jsonlite::fromJSON(xgb.config(model))
+  if (config$learner$gradient_booster$name == "gblinear") {
+    args <- list(importance_type = "weight", feature_names = feature_names)
+    results <- .Call(
+      XGBoosterFeatureScore_R, model$handle, jsonlite::toJSON(args, auto_unbox = TRUE, null = "null")
+    )
+    names(results) <- c("features", "shape", "weight")
+    n_classes <-  if (length(results$shape) == 2) { results$shape[2] } else { 0 }
+    importance <- if (n_classes == 0) {
+      data.table(Feature = results$features, Weight = results$weight)[order(-abs(Weight))]
     } else {
-      data.table(Feature = rep(feature_names, each = num_class),
-                 Weight = weights,
-                 Class = seq_len(num_class) - 1)[order(Class, -abs(Weight))]
+      data.table(
+        Feature = rep(results$features, each = n_classes), Weight = results$weight, Class = seq_len(n_classes) - 1
+      )[order(Class, -abs(Weight))]
     }
-  } else { # tree model
-      result <- xgb.model.dt.tree(feature_names = feature_names,
-                                  text = model_text_dump,
-                                  trees = trees)[
-          Feature != "Leaf", .(Gain = sum(Quality),
-                               Cover = sum(Cover),
-                               Frequency = .N), by = Feature][
-        , `:=`(Gain = Gain / sum(Gain),
-               Cover = Cover / sum(Cover),
-               Frequency = Frequency / sum(Frequency))][
-          order(Gain, decreasing = TRUE)]
+  } else {
+    concatenated <- list()
+    output_names <- vector()
+    for (importance_type in c("weight", "gain", "cover")) {
+      args <- list(importance_type = importance_type, feature_names = feature_names)
+      results <- .Call(
+        XGBoosterFeatureScore_R, model$handle, jsonlite::toJSON(args, auto_unbox = TRUE, null = "null")
+      )
+      names(results) <- c("features", "shape", importance_type)
+      concatenated[
+        switch(importance_type, "weight" = "Frequency", "gain" = "Gain", "cover" = "Cover")
+      ] <- results[importance_type]
+      output_names <- results$features
+    }
+    importance <- data.table(
+        Feature = output_names,
+        Gain = concatenated$Gain / sum(concatenated$Gain),
+        Cover = concatenated$Cover / sum(concatenated$Cover),
+        Frequency = concatenated$Frequency / sum(concatenated$Frequency)
+    )[order(Gain, decreasing = TRUE)]
   }
-  result
+  importance
 }
 
 # Avoid error messages during CRAN check.

R-package/R/xgb.plot.multi.trees.R

@@ -75,8 +75,8 @@ xgb.plot.multi.trees <- function(model, feature_names = NULL, features_keep = 5,
   while (tree.matrix[, sum(is.na(abs.node.position))] > 0) {
     yes.row.nodes <- tree.matrix[abs.node.position %in% precedent.nodes & !is.na(Yes)]
     no.row.nodes <- tree.matrix[abs.node.position %in% precedent.nodes & !is.na(No)]
-    yes.nodes.abs.pos <- yes.row.nodes[, abs.node.position] %>% paste0("_0")
-    no.nodes.abs.pos <- no.row.nodes[, abs.node.position] %>% paste0("_1")
+    yes.nodes.abs.pos <- paste0(yes.row.nodes[, abs.node.position], "_0")
+    no.nodes.abs.pos <- paste0(no.row.nodes[, abs.node.position], "_1")
 
     tree.matrix[ID %in% yes.row.nodes[, Yes], abs.node.position := yes.nodes.abs.pos]
     tree.matrix[ID %in% no.row.nodes[, No], abs.node.position := no.nodes.abs.pos]
@@ -92,19 +92,28 @@ xgb.plot.multi.trees <- function(model, feature_names = NULL, features_keep = 5,
   nodes.dt <- tree.matrix[
         , .(Quality = sum(Quality))
         , by = .(abs.node.position, Feature)
-      ][, .(Text = paste0(Feature[1:min(length(Feature), features_keep)],
-                          " (",
-                          format(Quality[1:min(length(Quality), features_keep)], digits = 5),
-                          ")") %>%
-                   paste0(collapse = "\n"))
-        , by = abs.node.position]
+      ][, .(Text = paste0(
+              paste0(
+                Feature[1:min(length(Feature), features_keep)],
+                " (",
+                format(Quality[1:min(length(Quality), features_keep)], digits = 5),
+                ")"
+              ),
+              collapse = "\n"
+            )
+          )
+        , by = abs.node.position
+      ]
 
-  edges.dt <- tree.matrix[Feature != "Leaf", .(abs.node.position, Yes)] %>%
-    list(tree.matrix[Feature != "Leaf", .(abs.node.position, No)]) %>%
-    rbindlist() %>%
-    setnames(c("From", "To")) %>%
-    .[, .N, .(From, To)] %>%
-    .[, N := NULL]
+  edges.dt <- data.table::rbindlist(
+    l = list(
+      tree.matrix[Feature != "Leaf", .(abs.node.position, Yes)],
+      tree.matrix[Feature != "Leaf", .(abs.node.position, No)]
+    )
+  )
+  data.table::setnames(edges.dt, c("From", "To"))
+  edges.dt <- edges.dt[, .N, .(From, To)]
+  edges.dt[, N := NULL]
 
   nodes <- DiagrammeR::create_node_df(
     n = nrow(nodes.dt),
@@ -120,21 +129,25 @@ xgb.plot.multi.trees <- function(model, feature_names = NULL, features_keep = 5,
       nodes_df = nodes,
       edges_df = edges,
       attr_theme = NULL
-      ) %>%
-    DiagrammeR::add_global_graph_attrs(
+  )
+  graph <- DiagrammeR::add_global_graph_attrs(
+      graph = graph,
       attr_type = "graph",
       attr  = c("layout", "rankdir"),
       value = c("dot", "LR")
-      ) %>%
-    DiagrammeR::add_global_graph_attrs(
+  )
+  graph <- DiagrammeR::add_global_graph_attrs(
+      graph = graph,
       attr_type = "node",
       attr  = c("color", "fillcolor", "style", "shape", "fontname"),
       value = c("DimGray", "beige", "filled", "rectangle", "Helvetica")
-      ) %>%
-    DiagrammeR::add_global_graph_attrs(
+  )
+  graph <- DiagrammeR::add_global_graph_attrs(
+      graph = graph,
       attr_type = "edge",
       attr  = c("color", "arrowsize", "arrowhead", "fontname"),
-      value = c("DimGray", "1.5", "vee", "Helvetica"))
+      value = c("DimGray", "1.5", "vee", "Helvetica")
+  )
 
   if (!render) return(invisible(graph))
 

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

@@ -33,7 +33,7 @@
 #' @param col_loess a color to use for the loess curves.
 #' @param span_loess the \code{span} parameter in \code{\link[stats]{loess}}'s call.
 #' @param which whether to do univariate or bivariate plotting. NOTE: only 1D is implemented so far.
-#' @param plot whether a plot should be drawn. If FALSE, only a lits of matrices is returned.
+#' @param plot whether a plot should be drawn. If FALSE, only a list of matrices is returned.
 #' @param ... other parameters passed to \code{plot}.
 #'
 #' @details
@@ -157,7 +157,7 @@ xgb.plot.shap <- function(data, shap_contrib = NULL, features = NULL, top_n = 1,
       plot(x2plot, y, pch = pch, xlab = f, col = col, xlim = x_lim, ylim = y_lim, ylab = ylab, ...)
       grid()
       if (plot_loess) {
-        # compress x to 3 digits, and mean-aggredate y
+        # compress x to 3 digits, and mean-aggregate y
         zz <- data.table(x = signif(x, 3), y)[, .(.N, y = mean(y)), x]
         if (nrow(zz) <= 5) {
           lines(zz$x, zz$y, col = col_loess)

R-package/R/xgb.plot.tree.R

@@ -99,33 +99,41 @@ xgb.plot.tree <- function(feature_names = NULL, model = NULL, trees = NULL, plot
     fontcolor = "black")
 
   edges <- DiagrammeR::create_edge_df(
-    from  = match(dt[Feature != "Leaf", c(ID)] %>% rep(2), dt$ID),
+    from  = match(rep(dt[Feature != "Leaf", c(ID)], 2), dt$ID),
     to    = match(dt[Feature != "Leaf", c(Yes, No)], dt$ID),
-    label = dt[Feature != "Leaf", paste("<", Split)] %>%
-            c(rep("", nrow(dt[Feature != "Leaf"]))),
-    style = dt[Feature != "Leaf", ifelse(Missing == Yes, "bold", "solid")] %>%
-            c(dt[Feature != "Leaf", ifelse(Missing == No, "bold", "solid")]),
+    label = c(
+      dt[Feature != "Leaf", paste("<", Split)],
+      rep("", nrow(dt[Feature != "Leaf"]))
+    ),
+    style = c(
+      dt[Feature != "Leaf", ifelse(Missing == Yes, "bold", "solid")],
+      dt[Feature != "Leaf", ifelse(Missing == No, "bold", "solid")]
+    ),
     rel   = "leading_to")
 
   graph <- DiagrammeR::create_graph(
       nodes_df = nodes,
       edges_df = edges,
       attr_theme = NULL
-      ) %>%
-    DiagrammeR::add_global_graph_attrs(
+  )
+  graph <- DiagrammeR::add_global_graph_attrs(
+      graph = graph,
       attr_type = "graph",
       attr  = c("layout", "rankdir"),
       value = c("dot", "LR")
-      ) %>%
-    DiagrammeR::add_global_graph_attrs(
+  )
+  graph <- DiagrammeR::add_global_graph_attrs(
+      graph = graph,
       attr_type = "node",
       attr  = c("color", "style", "fontname"),
       value = c("DimGray", "filled", "Helvetica")
-      ) %>%
-    DiagrammeR::add_global_graph_attrs(
+  )
+  graph <- DiagrammeR::add_global_graph_attrs(
+      graph = graph,
       attr_type = "edge",
       attr  = c("color", "arrowsize", "arrowhead", "fontname"),
-      value = c("DimGray", "1.5", "vee", "Helvetica"))
+      value = c("DimGray", "1.5", "vee", "Helvetica")
+  )
 
   if (!render) return(invisible(graph))
 

R-package/R/xgb.train.R

@@ -26,7 +26,7 @@
 #'   \item \code{colsample_bytree} subsample ratio of columns when constructing each tree. Default: 1
 #'   \item \code{lambda} L2 regularization term on weights. Default: 1
 #'   \item \code{alpha} L1 regularization term on weights. (there is no L1 reg on bias because it is not important). Default: 0
-#'   \item \code{num_parallel_tree} Experimental parameter. number of trees to grow per round. Useful to test Random Forest through Xgboost (set \code{colsample_bytree < 1}, \code{subsample  < 1}  and \code{round = 1}) accordingly. Default: 1
+#'   \item \code{num_parallel_tree} Experimental parameter. number of trees to grow per round. Useful to test Random Forest through XGBoost (set \code{colsample_bytree < 1}, \code{subsample  < 1}  and \code{round = 1}) accordingly. Default: 1
 #'   \item \code{monotone_constraints} A numerical vector consists of \code{1}, \code{0} and \code{-1} with its length equals to the number of features in the training data. \code{1} is increasing, \code{-1} is decreasing and \code{0} is no constraint.
 #'   \item \code{interaction_constraints} A list of vectors specifying feature indices of permitted interactions. Each item of the list represents one permitted interaction where specified features are allowed to interact with each other. Feature index values should start from \code{0} (\code{0} references the first column).  Leave argument unspecified for no interaction constraints.
 #' }
@@ -51,10 +51,10 @@
 #'     \item \code{binary:logistic} logistic regression for binary classification. Output probability.
 #'     \item \code{binary:logitraw} logistic regression for binary classification, output score before logistic transformation.
 #'     \item \code{binary:hinge}: hinge loss for binary classification. This makes predictions of 0 or 1, rather than producing probabilities.
-#'     \item \code{count:poisson}: poisson regression for count data, output mean of poisson distribution. \code{max_delta_step} is set to 0.7 by default in poisson regression (used to safeguard optimization).
+#'     \item \code{count:poisson}: Poisson regression for count data, output mean of Poisson distribution. \code{max_delta_step} is set to 0.7 by default in poisson regression (used to safeguard optimization).
 #'     \item \code{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 \code{h(t) = h0(t) * HR)}.
 #'     \item \code{survival:aft}: Accelerated failure time model for censored survival time data. See \href{https://xgboost.readthedocs.io/en/latest/tutorials/aft_survival_analysis.html}{Survival Analysis with Accelerated Failure Time} for details.
-#'     \item \code{aft_loss_distribution}: Probabilty Density Function used by \code{survival:aft} and \code{aft-nloglik} metric.
+#'     \item \code{aft_loss_distribution}: Probability Density Function used by \code{survival:aft} and \code{aft-nloglik} metric.
 #'     \item \code{multi:softmax} set xgboost to do multiclass classification using the softmax objective. Class is represented by a number and should be from 0 to \code{num_class - 1}.
 #'     \item \code{multi:softprob} same as softmax, but prediction outputs a vector of ndata * nclass elements, which can be further reshaped to ndata, nclass matrix. The result contains predicted probabilities of each data point belonging to each class.
 #'     \item \code{rank:pairwise} set xgboost to do ranking task by minimizing the pairwise loss.
@@ -126,11 +126,11 @@
 #' Parallelization is automatically enabled if \code{OpenMP} is present.
 #' Number of threads can also be manually specified via \code{nthread} parameter.
 #'
-#' The evaluation metric is chosen automatically by Xgboost (according to the objective)
+#' The evaluation metric is chosen automatically by XGBoost (according to the objective)
 #' when the \code{eval_metric} parameter is not provided.
 #' User may set one or several \code{eval_metric} parameters.
 #' Note that when using a customized metric, only this single metric can be used.
-#' The following is the list of built-in metrics for which Xgboost provides optimized implementation:
+#' The following is the list of built-in metrics for which XGBoost provides optimized implementation:
 #'   \itemize{
 #'      \item \code{rmse} root mean square error. \url{https://en.wikipedia.org/wiki/Root_mean_square_error}
 #'      \item \code{logloss} negative log-likelihood. \url{https://en.wikipedia.org/wiki/Log-likelihood}
@@ -171,9 +171,6 @@
 #'         explicitly passed.
 #'   \item \code{best_iteration} iteration number with the best evaluation metric value
 #'         (only available with early stopping).
-#'   \item \code{best_ntreelimit} the \code{ntreelimit} value corresponding to the best iteration,
-#'         which could further be used in \code{predict} method
-#'         (only available with early stopping).
 #'   \item \code{best_score} the best evaluation metric value during early stopping.
 #'         (only available with early stopping).
 #'   \item \code{feature_names} names of the training dataset features

R-package/R/xgb.unserialize.R

@@ -1,11 +1,21 @@
 #' Load the instance back from \code{\link{xgb.serialize}}
 #'
 #' @param buffer the buffer containing booster instance saved by \code{\link{xgb.serialize}}
+#' @param handle An \code{xgb.Booster.handle} object which will be overwritten with
+#' the new deserialized object. Must be a null handle (e.g. when loading the model through
+#' `readRDS`). If not provided, a new handle will be created.
+#' @return An \code{xgb.Booster.handle} object.
 #'
 #' @export
-xgb.unserialize <- function(buffer) {
+xgb.unserialize <- function(buffer, handle = NULL) {
   cachelist <- list()
-  handle <- .Call(XGBoosterCreate_R, cachelist)
+  if (is.null(handle)) {
+    handle <- .Call(XGBoosterCreate_R, cachelist)
+  } else {
+    if (!is.null.handle(handle))
+      stop("'handle' is not null/empty. Cannot overwrite existing handle.")
+    .Call(XGBoosterCreateInEmptyObj_R, cachelist, handle)
+  }
   tryCatch(
     .Call(XGBoosterUnserializeFromBuffer_R, handle, buffer),
     error = function(e) {

R-package/R/xgboost.R

@@ -10,7 +10,7 @@ xgboost <- function(data = NULL, label = NULL, missing = NA, weight = NULL,
                     save_period = NULL, save_name = "xgboost.model",
                     xgb_model = NULL, callbacks = list(), ...) {
 
-  dtrain <- xgb.get.DMatrix(data, label, missing, weight)
+  dtrain <- xgb.get.DMatrix(data, label, missing, weight, nthread = params$nthread)
 
   watchlist <- list(train = dtrain)
 
@@ -90,7 +90,6 @@ NULL
 #' @importFrom data.table setkey
 #' @importFrom data.table setkeyv
 #' @importFrom data.table setnames
-#' @importFrom magrittr %>%
 #' @importFrom jsonlite fromJSON
 #' @importFrom jsonlite toJSON
 #' @importFrom utils object.size str tail

R-package/README.md

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

R-package/cleanup

@@ -1,4 +1,3 @@
 #!/bin/sh
 
 rm -f src/Makevars
-rm -f CMakeLists.txt

R-package/demo/00Index

@@ -1,6 +1,6 @@
 basic_walkthrough               Basic feature walkthrough
 caret_wrapper                   Use xgboost to train in caret library
-custom_objective                Cutomize loss function, and evaluation metric
+custom_objective                Customize loss function, and evaluation metric
 boost_from_prediction           Boosting from existing prediction
 predict_first_ntree             Predicting using first n trees
 generalized_linear_model        Generalized Linear Model
@@ -8,8 +8,8 @@ cross_validation                Cross validation
 create_sparse_matrix            Create Sparse Matrix
 predict_leaf_indices            Predicting the corresponding leaves
 early_stopping                  Early Stop in training
-poisson_regression              Poisson Regression on count data
-tweedie_regression              Tweddie Regression
+poisson_regression              Poisson regression on count data
+tweedie_regression              Tweedie regression
 gpu_accelerated                 GPU-accelerated tree building algorithms
 interaction_constraints         Interaction constraints among features
 

R-package/demo/README.md

@@ -2,7 +2,7 @@ XGBoost R Feature Walkthrough
 ====
 * [Basic walkthrough of wrappers](basic_walkthrough.R)
 * [Train a xgboost model from caret library](caret_wrapper.R)
-* [Cutomize loss function, and evaluation metric](custom_objective.R)
+* [Customize loss function, and evaluation metric](custom_objective.R)
 * [Boosting from existing prediction](boost_from_prediction.R)
 * [Predicting using first n trees](predict_first_ntree.R)
 * [Generalized Linear Model](generalized_linear_model.R)

R-package/demo/basic_walkthrough.R

@@ -40,7 +40,7 @@ print("Train xgboost with verbose 2, also print information about tree")
 bst <- xgboost(data = dtrain, max_depth = 2, eta = 1, nrounds = 2,
                nthread = 2, objective = "binary:logistic", verbose = 2)
 
-# you can also specify data as file path to a LibSVM format input
+# you can also specify data as file path to a LIBSVM format input
 # since we do not have this file with us, the following line is just for illustration
 # bst <- xgboost(data = 'agaricus.train.svm', max_depth = 2, eta = 1, nrounds = 2,objective = "binary:logistic")
 

R-package/demo/create_sparse_matrix.R

@@ -2,17 +2,17 @@ require(xgboost)
 require(Matrix)
 require(data.table)
 if (!require(vcd)) {
-  install.packages('vcd') #Available in Cran. Used for its dataset with categorical values.
+  install.packages('vcd') #Available in CRAN. Used for its dataset with categorical values.
   require(vcd)
 }
-# According to its documentation, Xgboost works only on numbers.
+# According to its documentation, XGBoost works only on numbers.
 # Sometimes the dataset we have to work on have categorical data.
 # A categorical variable is one which have a fixed number of values. By example, if for each observation a variable called "Colour" can have only "red", "blue" or "green" as value, it is a categorical variable.
 #
 # In R, categorical variable is called Factor.
 # Type ?factor in console for more information.
 #
-# In this demo we will see how to transform a dense dataframe with categorical variables to a sparse matrix before analyzing it in Xgboost.
+# In this demo we will see how to transform a dense dataframe with categorical variables to a sparse matrix before analyzing it in XGBoost.
 # The method we are going to see is usually called "one hot encoding".
 
 #load Arthritis dataset in memory.
@@ -25,13 +25,13 @@ df <- data.table(Arthritis, keep.rownames = FALSE)
 cat("Print the dataset\n")
 print(df)
 
-# 2 columns have factor type, one has ordinal type (ordinal variable is a categorical variable with values wich can be ordered, here: None > Some > Marked).
+# 2 columns have factor type, one has ordinal type (ordinal variable is a categorical variable with values which can be ordered, here: None > Some > Marked).
 cat("Structure of the dataset\n")
 str(df)
 
 # Let's add some new categorical features to see if it helps. Of course these feature are highly correlated to the Age feature. Usually it's not a good thing in ML, but Tree algorithms (including boosted trees) are able to select the best features, even in case of highly correlated features.
 
-# For the first feature we create groups of age by rounding the real age. Note that we transform it to factor (categorical data) so the algorithm treat them as independant values.
+# For the first feature we create groups of age by rounding the real age. Note that we transform it to factor (categorical data) so the algorithm treat them as independent values.
 df[, AgeDiscret := as.factor(round(Age / 10, 0))]
 
 # Here is an even stronger simplification of the real age with an arbitrary split at 30 years old. I choose this value based on nothing. We will see later if simplifying the information based on arbitrary values is a good strategy (I am sure you already have an idea of how well it will work!).

R-package/demo/cross_validation.R

@@ -22,10 +22,10 @@ xgb.cv(param, dtrain, nrounds, nfold = 5,
        metrics = 'error', showsd = FALSE)
 
 ###
-# you can also do cross validation with cutomized loss function
+# you can also do cross validation with customized loss function
 # See custom_objective.R
 ##
-print ('running cross validation, with cutomsized loss function')
+print ('running cross validation, with customized loss function')
 
 logregobj <- function(preds, dtrain) {
   labels <- getinfo(dtrain, "label")

R-package/demo/custom_objective.R

@@ -12,7 +12,7 @@ watchlist <- list(eval = dtest, train = dtrain)
 num_round <- 2
 
 # user define objective function, given prediction, return gradient and second order gradient
-# this is loglikelihood loss
+# this is log likelihood loss
 logregobj <- function(preds, dtrain) {
   labels <- getinfo(dtrain, "label")
   preds <- 1 / (1 + exp(-preds))
@@ -23,9 +23,9 @@ logregobj <- function(preds, dtrain) {
 
 # user defined evaluation function, return a pair metric_name, result
 # NOTE: when you do customized loss function, the default prediction value is margin
-# this may make buildin evalution metric not function properly
+# this may make builtin evaluation metric not function properly
 # for example, we are doing logistic loss, the prediction is score before logistic transformation
-# the buildin evaluation error assumes input is after logistic transformation
+# the builtin evaluation error assumes input is after logistic transformation
 # Take this in mind when you use the customization, and maybe you need write customized evaluation function
 evalerror <- function(preds, dtrain) {
   labels <- getinfo(dtrain, "label")

R-package/demo/early_stopping.R

@@ -11,7 +11,7 @@ param <- list(max_depth = 2, eta = 1, nthread = 2, verbosity = 0)
 watchlist <- list(eval = dtest)
 num_round <- 20
 # user define objective function, given prediction, return gradient and second order gradient
-# this is loglikelihood loss
+# this is log likelihood loss
 logregobj <- function(preds, dtrain) {
   labels <- getinfo(dtrain, "label")
   preds <- 1 / (1 + exp(-preds))
@@ -21,9 +21,9 @@ logregobj <- function(preds, dtrain) {
 }
 # user defined evaluation function, return a pair metric_name, result
 # NOTE: when you do customized loss function, the default prediction value is margin
-# this may make buildin evalution metric not function properly
+# this may make builtin evaluation metric not function properly
 # for example, we are doing logistic loss, the prediction is score before logistic transformation
-# the buildin evaluation error assumes input is after logistic transformation
+# the builtin evaluation error assumes input is after logistic transformation
 # Take this in mind when you use the customization, and maybe you need write customized evaluation function
 evalerror <- function(preds, dtrain) {
   labels <- getinfo(dtrain, "label")

R-package/man/cb.early.stop.Rd

@@ -38,10 +38,7 @@ The following additional fields are assigned to the model's R object:
 \itemize{
 \item \code{best_score} the evaluation score at the best iteration
 \item \code{best_iteration} at which boosting iteration the best score has occurred (1-based index)
-\item \code{best_ntreelimit} to use with the \code{ntreelimit} parameter in \code{predict}.
-     It differs from \code{best_iteration} in multiclass or random forest settings.
 }
-
 The Same values are also stored as xgb-attributes:
 \itemize{
 \item \code{best_iteration} is stored as a 0-based iteration index (for interoperability of binary models)

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

@@ -8,7 +8,7 @@ during its training.}
 cb.gblinear.history(sparse = FALSE)
 }
 \arguments{
-\item{sparse}{when set to FALSE/TURE, a dense/sparse matrix is used to store the result.
+\item{sparse}{when set to FALSE/TRUE, 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.}
@@ -36,7 +36,6 @@ Callback function expects the following values to be set in its calling frame:
 #
 # 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"))
@@ -57,7 +56,7 @@ matplot(coef_path, type = 'l')
 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')
+matplot(xgb.gblinear.history(bst), 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.
@@ -66,7 +65,7 @@ xgb.gblinear.history(bst) \%>\% matplot(type = 'l')
 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')
+matplot(xgb.gblinear.history(bst)[[3]], type = 'l')
 
 
 #### Multiclass classification:
@@ -79,15 +78,15 @@ param <- list(booster = "gblinear", objective = "multi:softprob", num_class = 3,
 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')
+matplot(xgb.gblinear.history(bst, class_index = 0), type = 'l')
+matplot(xgb.gblinear.history(bst, class_index = 1), type = 'l')
+matplot(xgb.gblinear.history(bst, class_index = 2), 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')
+# 1st fold of 1st class
+matplot(xgb.gblinear.history(bst, class_index = 0)[[1]], type = 'l')
 
 }
 \seealso{

R-package/man/getinfo.Rd

@@ -23,9 +23,9 @@ Get information of an xgb.DMatrix object
 The \code{name} field can be one of the following:
 
 \itemize{
-    \item \code{label}: label Xgboost learn from ;
+    \item \code{label}: label XGBoost learn from ;
     \item \code{weight}: to do a weight rescale ;
-    \item \code{base_margin}: base margin is the base prediction Xgboost will boost from ;
+    \item \code{base_margin}: base margin is the base prediction XGBoost will boost from ;
     \item \code{nrow}: number of rows of the \code{xgb.DMatrix}.
 
 }
@@ -34,8 +34,7 @@ The \code{name} field can be one of the following:
 }
 \examples{
 data(agaricus.train, package='xgboost')
-train <- agaricus.train
-dtrain <- xgb.DMatrix(train$data, label=train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 
 labels <- getinfo(dtrain, 'label')
 setinfo(dtrain, 'label', 1-labels)

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

@@ -17,6 +17,8 @@
   predinteraction = FALSE,
   reshape = FALSE,
   training = FALSE,
+  iterationrange = NULL,
+  strict_shape = FALSE,
   ...
 )
 
@@ -34,8 +36,7 @@ missing values in data (e.g., sometimes 0 or some other extreme value is used).}
 sum of predictions from boosting iterations' results. E.g., setting \code{outputmargin=TRUE} for
 logistic regression would result in predictions for log-odds instead of probabilities.}
 
-\item{ntreelimit}{limit the number of model's trees or boosting iterations used in prediction (see Details).
-It will use all the trees by default (\code{NULL} value).}
+\item{ntreelimit}{Deprecated, use \code{iterationrange} instead.}
 
 \item{predleaf}{whether predict leaf index.}
 
@@ -52,10 +53,20 @@ or predinteraction flags is TRUE.}
 \item{training}{whether is the prediction result used for training.  For dart booster,
 training predicting will perform dropout.}
 
+\item{iterationrange}{Specifies which layer of trees are used in prediction.  For
+example, if a random forest is trained with 100 rounds.  Specifying
+`iteration_range=(1, 21)`, then only the forests built during [1, 21) (half open set)
+rounds are used in this prediction.  It's 1-based index just like R vector.  When set
+to \code{c(1, 1)} XGBoost will use all trees.}
+
+\item{strict_shape}{Default is \code{FALSE}. When it's set to \code{TRUE}, output
+type and shape of prediction are invariant to model type.}
+
 \item{...}{Parameters passed to \code{predict.xgb.Booster}}
 }
 \value{
-For regression or binary classification, it returns a vector of length \code{nrows(newdata)}.
+The return type is different depending whether \code{strict_shape} is set to \code{TRUE}.  By default,
+for regression or binary classification, it returns a vector of length \code{nrows(newdata)}.
 For multiclass classification, either a \code{num_class * nrows(newdata)} vector or
 a \code{(nrows(newdata), num_class)} dimension matrix is returned, depending on
 the \code{reshape} value.
@@ -76,18 +87,19 @@ two dimensions. The "+ 1" columns corresponds to bias. Summing this array along
 produce practically the same result as predict with \code{predcontrib = TRUE}.
 For a multiclass case, a list of \code{num_class} elements is returned, where each element is
 such an array.
+
+When \code{strict_shape} is set to \code{TRUE}, the output is always an array.  For
+normal prediction, the output is a 2-dimension array \code{(num_class, nrow(newdata))}.
+
+For \code{predcontrib = TRUE}, output is \code{(ncol(newdata) + 1, num_class, nrow(newdata))}
+For \code{predinteraction = TRUE}, output is \code{(ncol(newdata) + 1, ncol(newdata) + 1, num_class, nrow(newdata))}
+For \code{predleaf = TRUE}, output is \code{(n_trees_in_forest, num_class, n_iterations, nrow(newdata))}
 }
 \description{
 Predicted values based on either xgboost model or model handle object.
 }
 \details{
-Note that \code{ntreelimit} is not necessarily equal to the number of boosting iterations
-and it is not necessarily equal to the number of trees in a model.
-E.g., in a random forest-like model, \code{ntreelimit} would limit the number of trees.
-But for multiclass classification, while there are multiple trees per iteration,
-\code{ntreelimit} limits the number of boosting iterations.
-
-Also note that \code{ntreelimit} would currently do nothing for predictions from gblinear,
+Note that \code{iterationrange} would currently do nothing for predictions from gblinear,
 since gblinear doesn't keep its boosting history.
 
 One possible practical applications of the \code{predleaf} option is to use the model
@@ -120,7 +132,7 @@ bst <- xgboost(data = train$data, label = train$label, max_depth = 2,
 # use all trees by default
 pred <- predict(bst, test$data)
 # use only the 1st tree
-pred1 <- predict(bst, test$data, ntreelimit = 1)
+pred1 <- predict(bst, test$data, iterationrange = c(1, 2))
 
 # Predicting tree leafs:
 # the result is an nsamples X ntrees matrix
@@ -172,25 +184,9 @@ str(pred)
 all.equal(pred, pred_labels)
 # prediction from using only 5 iterations should result
 # in the same error as seen in iteration 5:
-pred5 <- predict(bst, as.matrix(iris[, -5]), ntreelimit=5)
+pred5 <- predict(bst, as.matrix(iris[, -5]), iterationrange=c(1, 6))
 sum(pred5 != lb)/length(lb)
 
-
-## random forest-like model of 25 trees for binary classification:
-
-set.seed(11)
-bst <- xgboost(data = train$data, label = train$label, max_depth = 5,
-               nthread = 2, nrounds = 1, objective = "binary:logistic",
-               num_parallel_tree = 25, subsample = 0.6, colsample_bytree = 0.1)
-# Inspect the prediction error vs number of trees:
-lb <- test$label
-dtest <- xgb.DMatrix(test$data, label=lb)
-err <- sapply(1:25, function(n) {
-  pred <- predict(bst, dtest, ntreelimit=n)
-  sum((pred > 0.5) != lb)/length(lb)
-})
-plot(err, type='l', ylim=c(0,0.1), xlab='#trees')
-
 }
 \references{
 Scott M. Lundberg, Su-In Lee, "A Unified Approach to Interpreting Model Predictions", NIPS Proceedings 2017, \url{https://arxiv.org/abs/1705.07874}

R-package/man/print.xgb.DMatrix.Rd

@@ -19,8 +19,7 @@ Currently it displays dimensions and presence of info-fields and colnames.
 }
 \examples{
 data(agaricus.train, package='xgboost')
-train <- agaricus.train
-dtrain <- xgb.DMatrix(train$data, label=train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 
 dtrain
 print(dtrain, verbose=TRUE)

R-package/man/setinfo.Rd

@@ -25,16 +25,15 @@ Set information of an xgb.DMatrix object
 The \code{name} field can be one of the following:
 
 \itemize{
-    \item \code{label}: label Xgboost learn from ;
+    \item \code{label}: label XGBoost learn from ;
     \item \code{weight}: to do a weight rescale ;
-    \item \code{base_margin}: base margin is the base prediction Xgboost will boost from ;
+    \item \code{base_margin}: base margin is the base prediction XGBoost will boost from ;
     \item \code{group}: number of rows in each group (to use with \code{rank:pairwise} objective).
 }
 }
 \examples{
 data(agaricus.train, package='xgboost')
-train <- agaricus.train
-dtrain <- xgb.DMatrix(train$data, label=train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 
 labels <- getinfo(dtrain, 'label')
 setinfo(dtrain, 'label', 1-labels)

R-package/man/slice.xgb.DMatrix.Rd

@@ -28,8 +28,7 @@ original xgb.DMatrix object
 }
 \examples{
 data(agaricus.train, package='xgboost')
-train <- agaricus.train
-dtrain <- xgb.DMatrix(train$data, label=train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 
 dsub <- slice(dtrain, 1:42)
 labels1 <- getinfo(dsub, 'label')

R-package/man/xgb.DMatrix.Rd

@@ -22,13 +22,12 @@ It is useful when a 0 or some other extreme value represents missing values in d
 }
 \description{
 Construct xgb.DMatrix object from either a dense matrix, a sparse matrix, or a local file.
-Supported input file formats are either a libsvm text file or a binary file that was created previously by
+Supported input file formats are either a LIBSVM text file or a binary file that was created previously by
 \code{\link{xgb.DMatrix.save}}).
 }
 \examples{
 data(agaricus.train, package='xgboost')
-train <- agaricus.train
-dtrain <- xgb.DMatrix(train$data, label=train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 xgb.DMatrix.save(dtrain, 'xgb.DMatrix.data')
 dtrain <- xgb.DMatrix('xgb.DMatrix.data')
 if (file.exists('xgb.DMatrix.data')) file.remove('xgb.DMatrix.data')

R-package/man/xgb.DMatrix.save.Rd

@@ -16,8 +16,7 @@ Save xgb.DMatrix object to binary file
 }
 \examples{
 data(agaricus.train, package='xgboost')
-train <- agaricus.train
-dtrain <- xgb.DMatrix(train$data, label=train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 xgb.DMatrix.save(dtrain, 'xgb.DMatrix.data')
 dtrain <- xgb.DMatrix('xgb.DMatrix.data')
 if (file.exists('xgb.DMatrix.data')) file.remove('xgb.DMatrix.data')

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

@@ -59,8 +59,8 @@ a rule on certain features."
 \examples{
 data(agaricus.train, package='xgboost')
 data(agaricus.test, package='xgboost')
-dtrain <- xgb.DMatrix(data = agaricus.train$data, label = agaricus.train$label)
-dtest <- xgb.DMatrix(data = agaricus.test$data, label = agaricus.test$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
+dtest <- with(agaricus.test, xgb.DMatrix(data, label = label))
 
 param <- list(max_depth=2, eta=1, silent=1, objective='binary:logistic')
 nrounds = 4

R-package/man/xgb.cv.Rd

@@ -135,9 +135,7 @@ An object of class \code{xgb.cv.synchronous} with the following elements:
         parameter or randomly generated.
   \item \code{best_iteration} iteration number with the best evaluation metric value
         (only available with early stopping).
-  \item \code{best_ntreelimit} the \code{ntreelimit} value corresponding to the best iteration,
-        which could further be used in \code{predict} method
-        (only available with early stopping).
+  \item \code{best_ntreelimit} and the \code{ntreelimit} Deprecated attributes, use \code{best_iteration} instead.
   \item \code{pred} CV prediction values available when \code{prediction} is set.
         It is either vector or matrix (see \code{\link{cb.cv.predict}}).
   \item \code{models} a list of the CV folds' models. It is only available with the explicit
@@ -160,7 +158,7 @@ Adapted from \url{https://en.wikipedia.org/wiki/Cross-validation_\%28statistics\
 }
 \examples{
 data(agaricus.train, package='xgboost')
-dtrain <- xgb.DMatrix(agaricus.train$data, label = agaricus.train$label)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
 cv <- xgb.cv(data = dtrain, nrounds = 3, nthread = 2, nfold = 5, metrics = list("rmse","auc"),
                   max_depth = 3, eta = 1, objective = "binary:logistic")
 print(cv)

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

@@ -87,7 +87,7 @@ more than 5 distinct values.}
 
 \item{which}{whether to do univariate or bivariate plotting. NOTE: only 1D is implemented so far.}
 
-\item{plot}{whether a plot should be drawn. If FALSE, only a lits of matrices is returned.}
+\item{plot}{whether a plot should be drawn. If FALSE, only a list of matrices is returned.}
 
 \item{...}{other parameters passed to \code{plot}.}
 }

R-package/man/xgb.train.Rd

@@ -54,7 +54,7 @@ xgboost(
 
 2. Booster Parameters
 
-2.1. Parameter for Tree Booster
+2.1. Parameters for Tree Booster
 
 \itemize{
   \item \code{eta} control the learning rate: scale the contribution of each tree by a factor of \code{0 < eta < 1} when it is added to the current approximation. Used to prevent overfitting by making the boosting process more conservative. Lower value for \code{eta} implies larger value for \code{nrounds}: low \code{eta} value means model more robust to overfitting but slower to compute. Default: 0.3
@@ -63,12 +63,14 @@ xgboost(
   \item \code{min_child_weight} minimum sum of instance weight (hessian) needed in a child. If the tree partition step results in a leaf node with the sum of instance weight less than min_child_weight, then the building process will give up further partitioning. In linear regression mode, this simply corresponds to minimum number of instances needed to be in each node. The larger, the more conservative the algorithm will be. Default: 1
   \item \code{subsample} subsample ratio of the training instance. Setting it to 0.5 means that xgboost randomly collected half of the data instances to grow trees and this will prevent overfitting. It makes computation shorter (because less data to analyse). It is advised to use this parameter with \code{eta} and increase \code{nrounds}. Default: 1
   \item \code{colsample_bytree} subsample ratio of columns when constructing each tree. Default: 1
-  \item \code{num_parallel_tree} Experimental parameter. number of trees to grow per round. Useful to test Random Forest through Xgboost (set \code{colsample_bytree < 1}, \code{subsample  < 1}  and \code{round = 1}) accordingly. Default: 1
+  \item \code{lambda} L2 regularization term on weights. Default: 1
+  \item \code{alpha} L1 regularization term on weights. (there is no L1 reg on bias because it is not important). Default: 0
+  \item \code{num_parallel_tree} Experimental parameter. number of trees to grow per round. Useful to test Random Forest through XGBoost (set \code{colsample_bytree < 1}, \code{subsample  < 1}  and \code{round = 1}) accordingly. Default: 1
   \item \code{monotone_constraints} A numerical vector consists of \code{1}, \code{0} and \code{-1} with its length equals to the number of features in the training data. \code{1} is increasing, \code{-1} is decreasing and \code{0} is no constraint.
   \item \code{interaction_constraints} A list of vectors specifying feature indices of permitted interactions. Each item of the list represents one permitted interaction where specified features are allowed to interact with each other. Feature index values should start from \code{0} (\code{0} references the first column).  Leave argument unspecified for no interaction constraints.
 }
 
-2.2. Parameter for Linear Booster
+2.2. Parameters for Linear Booster
 
 \itemize{
   \item \code{lambda} L2 regularization term on weights. Default: 0
@@ -88,10 +90,10 @@ xgboost(
     \item \code{binary:logistic} logistic regression for binary classification. Output probability.
     \item \code{binary:logitraw} logistic regression for binary classification, output score before logistic transformation.
     \item \code{binary:hinge}: hinge loss for binary classification. This makes predictions of 0 or 1, rather than producing probabilities.
-    \item \code{count:poisson}: poisson regression for count data, output mean of poisson distribution. \code{max_delta_step} is set to 0.7 by default in poisson regression (used to safeguard optimization).
+    \item \code{count:poisson}: Poisson regression for count data, output mean of Poisson distribution. \code{max_delta_step} is set to 0.7 by default in poisson regression (used to safeguard optimization).
     \item \code{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 \code{h(t) = h0(t) * HR)}.
     \item \code{survival:aft}: Accelerated failure time model for censored survival time data. See \href{https://xgboost.readthedocs.io/en/latest/tutorials/aft_survival_analysis.html}{Survival Analysis with Accelerated Failure Time} for details.
-    \item \code{aft_loss_distribution}: Probabilty Density Function used by \code{survival:aft} and \code{aft-nloglik} metric.
+    \item \code{aft_loss_distribution}: Probability Density Function used by \code{survival:aft} and \code{aft-nloglik} metric.
     \item \code{multi:softmax} set xgboost to do multiclass classification using the softmax objective. Class is represented by a number and should be from 0 to \code{num_class - 1}.
     \item \code{multi:softprob} same as softmax, but prediction outputs a vector of ndata * nclass elements, which can be further reshaped to ndata, nclass matrix. The result contains predicted probabilities of each data point belonging to each class.
     \item \code{rank:pairwise} set xgboost to do ranking task by minimizing the pairwise loss.
@@ -185,9 +187,6 @@ An object of class \code{xgb.Booster} with the following elements:
         explicitly passed.
   \item \code{best_iteration} iteration number with the best evaluation metric value
         (only available with early stopping).
-  \item \code{best_ntreelimit} the \code{ntreelimit} value corresponding to the best iteration,
-        which could further be used in \code{predict} method
-        (only available with early stopping).
   \item \code{best_score} the best evaluation metric value during early stopping.
         (only available with early stopping).
   \item \code{feature_names} names of the training dataset features
@@ -209,11 +208,11 @@ than the \code{xgboost} interface.
 Parallelization is automatically enabled if \code{OpenMP} is present.
 Number of threads can also be manually specified via \code{nthread} parameter.
 
-The evaluation metric is chosen automatically by Xgboost (according to the objective)
+The evaluation metric is chosen automatically by XGBoost (according to the objective)
 when the \code{eval_metric} parameter is not provided.
 User may set one or several \code{eval_metric} parameters.
 Note that when using a customized metric, only this single metric can be used.
-The following is the list of built-in metrics for which Xgboost provides optimized implementation:
+The following is the list of built-in metrics for which XGBoost provides optimized implementation:
   \itemize{
      \item \code{rmse} root mean square error. \url{https://en.wikipedia.org/wiki/Root_mean_square_error}
      \item \code{logloss} negative log-likelihood. \url{https://en.wikipedia.org/wiki/Log-likelihood}
@@ -242,8 +241,8 @@ The following callbacks are automatically created when certain parameters are se
 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)
+dtrain <- with(agaricus.train, xgb.DMatrix(data, label = label))
+dtest <- with(agaricus.test, xgb.DMatrix(data, label = label))
 watchlist <- list(train = dtrain, eval = dtest)
 
 ## A simple xgb.train example:

R-package/man/xgb.unserialize.Rd

@@ -4,10 +4,17 @@
 \alias{xgb.unserialize}
 \title{Load the instance back from \code{\link{xgb.serialize}}}
 \usage{
-xgb.unserialize(buffer)
+xgb.unserialize(buffer, handle = NULL)
 }
 \arguments{
 \item{buffer}{the buffer containing booster instance saved by \code{\link{xgb.serialize}}}
+
+\item{handle}{An \code{xgb.Booster.handle} object which will be overwritten with
+the new deserialized object. Must be a null handle (e.g. when loading the model through
+`readRDS`). If not provided, a new handle will be created.}
+}
+\value{
+An \code{xgb.Booster.handle} object.
 }
 \description{
 Load the instance back from \code{\link{xgb.serialize}}

R-package/src/Makevars.in

@@ -17,9 +17,9 @@ 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@ @ENDIAN_FLAG@ -pthread
+PKG_CXXFLAGS= @OPENMP_CXXFLAGS@ @ENDIAN_FLAG@ -pthread $(CXX_VISIBILITY)
 PKG_LIBS = @OPENMP_CXXFLAGS@ @OPENMP_LIB@ @ENDIAN_FLAG@ @BACKTRACE_LIB@ -pthread
 OBJECTS= ./xgboost_R.o ./xgboost_custom.o ./xgboost_assert.o ./init.o \
          $(PKGROOT)/amalgamation/xgboost-all0.o $(PKGROOT)/amalgamation/dmlc-minimum0.o \
-         $(PKGROOT)/rabit/src/engine.o $(PKGROOT)/rabit/src/c_api.o \
+         $(PKGROOT)/rabit/src/engine.o $(PKGROOT)/rabit/src/rabit_c_api.o \
          $(PKGROOT)/rabit/src/allreduce_base.o

R-package/src/Makevars.win

@@ -3,7 +3,7 @@ PKGROOT=./
 ENABLE_STD_THREAD=0
 # _*_ mode: Makefile; _*_
 
-# This file is only used for windows compilation from github
+# This file is only used for Windows compilation from GitHub
 # It will be replaced with Makevars.in for the CRAN version
 .PHONY: all xgblib
 all: $(SHLIB)
@@ -33,7 +33,7 @@ PKG_CXXFLAGS= $(SHLIB_OPENMP_CXXFLAGS) $(SHLIB_PTHREAD_FLAGS)
 PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS) $(SHLIB_PTHREAD_FLAGS)
 OBJECTS= ./xgboost_R.o ./xgboost_custom.o ./xgboost_assert.o ./init.o \
          $(PKGROOT)/amalgamation/xgboost-all0.o $(PKGROOT)/amalgamation/dmlc-minimum0.o \
-         $(PKGROOT)/rabit/src/engine.o $(PKGROOT)/rabit/src/c_api.o \
+         $(PKGROOT)/rabit/src/engine.o $(PKGROOT)/rabit/src/rabit_c_api.o \
          $(PKGROOT)/rabit/src/allreduce_base.o
 
 $(OBJECTS) : xgblib

R-package/src/init.c

@@ -9,6 +9,7 @@
 #include <Rinternals.h>
 #include <stdlib.h>
 #include <R_ext/Rdynload.h>
+#include <R_ext/Visibility.h>
 
 /* FIXME:
 Check these declarations against the C/Fortran source code.
@@ -17,6 +18,7 @@ Check these declarations against the C/Fortran source code.
 /* .Call calls */
 extern SEXP XGBoosterBoostOneIter_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGBoosterCreate_R(SEXP);
+extern SEXP XGBoosterCreateInEmptyObj_R(SEXP, SEXP);
 extern SEXP XGBoosterDumpModel_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGBoosterEvalOneIter_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGBoosterGetAttrNames_R(SEXP);
@@ -29,6 +31,7 @@ extern SEXP XGBoosterSerializeToBuffer_R(SEXP handle);
 extern SEXP XGBoosterUnserializeFromBuffer_R(SEXP handle, SEXP raw);
 extern SEXP XGBoosterModelToRaw_R(SEXP);
 extern SEXP XGBoosterPredict_R(SEXP, SEXP, SEXP, SEXP, SEXP);
+extern SEXP XGBoosterPredictFromDMatrix_R(SEXP, SEXP, SEXP);
 extern SEXP XGBoosterSaveModel_R(SEXP, SEXP);
 extern SEXP XGBoosterSetAttr_R(SEXP, SEXP, SEXP);
 extern SEXP XGBoosterSetParam_R(SEXP, SEXP, SEXP);
@@ -36,7 +39,7 @@ extern SEXP XGBoosterUpdateOneIter_R(SEXP, SEXP, SEXP);
 extern SEXP XGCheckNullPtr_R(SEXP);
 extern SEXP XGDMatrixCreateFromCSC_R(SEXP, SEXP, SEXP, SEXP);
 extern SEXP XGDMatrixCreateFromFile_R(SEXP, SEXP);
-extern SEXP XGDMatrixCreateFromMat_R(SEXP, SEXP);
+extern SEXP XGDMatrixCreateFromMat_R(SEXP, SEXP, SEXP);
 extern SEXP XGDMatrixGetInfo_R(SEXP, SEXP);
 extern SEXP XGDMatrixNumCol_R(SEXP);
 extern SEXP XGDMatrixNumRow_R(SEXP);
@@ -45,10 +48,12 @@ extern SEXP XGDMatrixSetInfo_R(SEXP, SEXP, SEXP);
 extern SEXP XGDMatrixSliceDMatrix_R(SEXP, SEXP);
 extern SEXP XGBSetGlobalConfig_R(SEXP);
 extern SEXP XGBGetGlobalConfig_R();
+extern SEXP XGBoosterFeatureScore_R(SEXP, SEXP);
 
 static const R_CallMethodDef CallEntries[] = {
   {"XGBoosterBoostOneIter_R",     (DL_FUNC) &XGBoosterBoostOneIter_R,     4},
   {"XGBoosterCreate_R",           (DL_FUNC) &XGBoosterCreate_R,           1},
+  {"XGBoosterCreateInEmptyObj_R", (DL_FUNC) &XGBoosterCreateInEmptyObj_R, 2},
   {"XGBoosterDumpModel_R",        (DL_FUNC) &XGBoosterDumpModel_R,        4},
   {"XGBoosterEvalOneIter_R",      (DL_FUNC) &XGBoosterEvalOneIter_R,      4},
   {"XGBoosterGetAttrNames_R",     (DL_FUNC) &XGBoosterGetAttrNames_R,     1},
@@ -61,6 +66,7 @@ static const R_CallMethodDef CallEntries[] = {
   {"XGBoosterUnserializeFromBuffer_R", (DL_FUNC) &XGBoosterUnserializeFromBuffer_R, 2},
   {"XGBoosterModelToRaw_R",       (DL_FUNC) &XGBoosterModelToRaw_R,       1},
   {"XGBoosterPredict_R",          (DL_FUNC) &XGBoosterPredict_R,          5},
+  {"XGBoosterPredictFromDMatrix_R", (DL_FUNC) &XGBoosterPredictFromDMatrix_R, 3},
   {"XGBoosterSaveModel_R",        (DL_FUNC) &XGBoosterSaveModel_R,        2},
   {"XGBoosterSetAttr_R",          (DL_FUNC) &XGBoosterSetAttr_R,          3},
   {"XGBoosterSetParam_R",         (DL_FUNC) &XGBoosterSetParam_R,         3},
@@ -68,7 +74,7 @@ static const R_CallMethodDef CallEntries[] = {
   {"XGCheckNullPtr_R",            (DL_FUNC) &XGCheckNullPtr_R,            1},
   {"XGDMatrixCreateFromCSC_R",    (DL_FUNC) &XGDMatrixCreateFromCSC_R,    4},
   {"XGDMatrixCreateFromFile_R",   (DL_FUNC) &XGDMatrixCreateFromFile_R,   2},
-  {"XGDMatrixCreateFromMat_R",    (DL_FUNC) &XGDMatrixCreateFromMat_R,    2},
+  {"XGDMatrixCreateFromMat_R",    (DL_FUNC) &XGDMatrixCreateFromMat_R,    3},
   {"XGDMatrixGetInfo_R",          (DL_FUNC) &XGDMatrixGetInfo_R,          2},
   {"XGDMatrixNumCol_R",           (DL_FUNC) &XGDMatrixNumCol_R,           1},
   {"XGDMatrixNumRow_R",           (DL_FUNC) &XGDMatrixNumRow_R,           1},
@@ -77,13 +83,14 @@ static const R_CallMethodDef CallEntries[] = {
   {"XGDMatrixSliceDMatrix_R",     (DL_FUNC) &XGDMatrixSliceDMatrix_R,     2},
   {"XGBSetGlobalConfig_R",        (DL_FUNC) &XGBSetGlobalConfig_R,        1},
   {"XGBGetGlobalConfig_R",        (DL_FUNC) &XGBGetGlobalConfig_R,        0},
+  {"XGBoosterFeatureScore_R",     (DL_FUNC) &XGBoosterFeatureScore_R,     2},
   {NULL, NULL, 0}
 };
 
 #if defined(_WIN32)
 __declspec(dllexport)
 #endif  // defined(_WIN32)
-void R_init_xgboost(DllInfo *dll) {
+void attribute_visible R_init_xgboost(DllInfo *dll) {
   R_registerRoutines(dll, NULL, CallEntries, NULL, NULL);
   R_useDynamicSymbols(dll, FALSE);
 }

R-package/src/xgboost-win.def

@@ -0,0 +1,3 @@
+LIBRARY xgboost.dll
+EXPORTS
+ R_init_xgboost

R-package/src/xgboost_R.cc

@@ -9,6 +9,8 @@
 #include <cstring>
 #include <cstdio>
 #include <sstream>
+
+#include "../../src/common/threading_utils.h"
 #include "./xgboost_R.h"
 
 /*!
@@ -38,11 +40,11 @@
 
 using namespace dmlc;
 
-SEXP XGCheckNullPtr_R(SEXP handle) {
+XGB_DLL SEXP XGCheckNullPtr_R(SEXP handle) {
   return ScalarLogical(R_ExternalPtrAddr(handle) == NULL);
 }
 
-void _DMatrixFinalizer(SEXP ext) {
+XGB_DLL void _DMatrixFinalizer(SEXP ext) {
   R_API_BEGIN();
   if (R_ExternalPtrAddr(ext) == NULL) return;
   CHECK_CALL(XGDMatrixFree(R_ExternalPtrAddr(ext)));
@@ -50,14 +52,14 @@ void _DMatrixFinalizer(SEXP ext) {
   R_API_END();
 }
 
-SEXP XGBSetGlobalConfig_R(SEXP json_str) {
+XGB_DLL SEXP XGBSetGlobalConfig_R(SEXP json_str) {
   R_API_BEGIN();
   CHECK_CALL(XGBSetGlobalConfig(CHAR(asChar(json_str))));
   R_API_END();
   return R_NilValue;
 }
 
-SEXP XGBGetGlobalConfig_R() {
+XGB_DLL SEXP XGBGetGlobalConfig_R() {
   const char* json_str;
   R_API_BEGIN();
   CHECK_CALL(XGBGetGlobalConfig(&json_str));
@@ -65,7 +67,7 @@ SEXP XGBGetGlobalConfig_R() {
   return mkString(json_str);
 }
 
-SEXP XGDMatrixCreateFromFile_R(SEXP fname, SEXP silent) {
+XGB_DLL SEXP XGDMatrixCreateFromFile_R(SEXP fname, SEXP silent) {
   SEXP ret;
   R_API_BEGIN();
   DMatrixHandle handle;
@@ -77,8 +79,7 @@ SEXP XGDMatrixCreateFromFile_R(SEXP fname, SEXP silent) {
   return ret;
 }
 
-SEXP XGDMatrixCreateFromMat_R(SEXP mat,
-                              SEXP missing) {
+XGB_DLL SEXP XGDMatrixCreateFromMat_R(SEXP mat, SEXP missing, SEXP n_threads) {
   SEXP ret;
   R_API_BEGIN();
   SEXP dim = getAttrib(mat, R_DimSymbol);
@@ -94,7 +95,9 @@ SEXP XGDMatrixCreateFromMat_R(SEXP mat,
   }
   std::vector<float> data(nrow * ncol);
   dmlc::OMPException exc;
-  #pragma omp parallel for schedule(static)
+  int32_t threads = xgboost::common::OmpGetNumThreads(asInteger(n_threads));
+
+#pragma omp parallel for schedule(static) num_threads(threads)
   for (omp_ulong i = 0; i < nrow; ++i) {
     exc.Run([&]() {
       for (size_t j = 0; j < ncol; ++j) {
@@ -104,7 +107,8 @@ SEXP XGDMatrixCreateFromMat_R(SEXP mat,
   }
   exc.Rethrow();
   DMatrixHandle handle;
-  CHECK_CALL(XGDMatrixCreateFromMat(BeginPtr(data), nrow, ncol, asReal(missing), &handle));
+  CHECK_CALL(XGDMatrixCreateFromMat_omp(BeginPtr(data), nrow, ncol,
+                                        asReal(missing), &handle, threads));
   ret = PROTECT(R_MakeExternalPtr(handle, R_NilValue, R_NilValue));
   R_RegisterCFinalizerEx(ret, _DMatrixFinalizer, TRUE);
   R_API_END();
@@ -112,10 +116,8 @@ SEXP XGDMatrixCreateFromMat_R(SEXP mat,
   return ret;
 }
 
-SEXP XGDMatrixCreateFromCSC_R(SEXP indptr,
-                              SEXP indices,
-                              SEXP data,
-                              SEXP num_row) {
+XGB_DLL SEXP XGDMatrixCreateFromCSC_R(SEXP indptr, SEXP indices, SEXP data,
+                                      SEXP num_row) {
   SEXP ret;
   R_API_BEGIN();
   const int *p_indptr = INTEGER(indptr);
@@ -151,7 +153,7 @@ SEXP XGDMatrixCreateFromCSC_R(SEXP indptr,
   return ret;
 }
 
-SEXP XGDMatrixSliceDMatrix_R(SEXP handle, SEXP idxset) {
+XGB_DLL SEXP XGDMatrixSliceDMatrix_R(SEXP handle, SEXP idxset) {
   SEXP ret;
   R_API_BEGIN();
   int len = length(idxset);
@@ -171,7 +173,7 @@ SEXP XGDMatrixSliceDMatrix_R(SEXP handle, SEXP idxset) {
   return ret;
 }
 
-SEXP XGDMatrixSaveBinary_R(SEXP handle, SEXP fname, SEXP silent) {
+XGB_DLL SEXP XGDMatrixSaveBinary_R(SEXP handle, SEXP fname, SEXP silent) {
   R_API_BEGIN();
   CHECK_CALL(XGDMatrixSaveBinary(R_ExternalPtrAddr(handle),
                                  CHAR(asChar(fname)),
@@ -180,7 +182,7 @@ SEXP XGDMatrixSaveBinary_R(SEXP handle, SEXP fname, SEXP silent) {
   return R_NilValue;
 }
 
-SEXP XGDMatrixSetInfo_R(SEXP handle, SEXP field, SEXP array) {
+XGB_DLL SEXP XGDMatrixSetInfo_R(SEXP handle, SEXP field, SEXP array) {
   R_API_BEGIN();
   int len = length(array);
   const char *name = CHAR(asChar(field));
@@ -214,7 +216,7 @@ SEXP XGDMatrixSetInfo_R(SEXP handle, SEXP field, SEXP array) {
   return R_NilValue;
 }
 
-SEXP XGDMatrixGetInfo_R(SEXP handle, SEXP field) {
+XGB_DLL SEXP XGDMatrixGetInfo_R(SEXP handle, SEXP field) {
   SEXP ret;
   R_API_BEGIN();
   bst_ulong olen;
@@ -232,7 +234,7 @@ SEXP XGDMatrixGetInfo_R(SEXP handle, SEXP field) {
   return ret;
 }
 
-SEXP XGDMatrixNumRow_R(SEXP handle) {
+XGB_DLL SEXP XGDMatrixNumRow_R(SEXP handle) {
   bst_ulong nrow;
   R_API_BEGIN();
   CHECK_CALL(XGDMatrixNumRow(R_ExternalPtrAddr(handle), &nrow));
@@ -240,7 +242,7 @@ SEXP XGDMatrixNumRow_R(SEXP handle) {
   return ScalarInteger(static_cast<int>(nrow));
 }
 
-SEXP XGDMatrixNumCol_R(SEXP handle) {
+XGB_DLL SEXP XGDMatrixNumCol_R(SEXP handle) {
   bst_ulong ncol;
   R_API_BEGIN();
   CHECK_CALL(XGDMatrixNumCol(R_ExternalPtrAddr(handle), &ncol));
@@ -255,7 +257,7 @@ void _BoosterFinalizer(SEXP ext) {
   R_ClearExternalPtr(ext);
 }
 
-SEXP XGBoosterCreate_R(SEXP dmats) {
+XGB_DLL SEXP XGBoosterCreate_R(SEXP dmats) {
   SEXP ret;
   R_API_BEGIN();
   int len = length(dmats);
@@ -272,7 +274,22 @@ SEXP XGBoosterCreate_R(SEXP dmats) {
   return ret;
 }
 
-SEXP XGBoosterSetParam_R(SEXP handle, SEXP name, SEXP val) {
+XGB_DLL SEXP XGBoosterCreateInEmptyObj_R(SEXP dmats, SEXP R_handle) {
+  R_API_BEGIN();
+  int len = length(dmats);
+  std::vector<void*> dvec;
+  for (int i = 0; i < len; ++i) {
+    dvec.push_back(R_ExternalPtrAddr(VECTOR_ELT(dmats, i)));
+  }
+  BoosterHandle handle;
+  CHECK_CALL(XGBoosterCreate(BeginPtr(dvec), dvec.size(), &handle));
+  R_SetExternalPtrAddr(R_handle, handle);
+  R_RegisterCFinalizerEx(R_handle, _BoosterFinalizer, TRUE);
+  R_API_END();
+  return R_NilValue;
+}
+
+XGB_DLL SEXP XGBoosterSetParam_R(SEXP handle, SEXP name, SEXP val) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterSetParam(R_ExternalPtrAddr(handle),
                                CHAR(asChar(name)),
@@ -281,7 +298,7 @@ SEXP XGBoosterSetParam_R(SEXP handle, SEXP name, SEXP val) {
   return R_NilValue;
 }
 
-SEXP XGBoosterUpdateOneIter_R(SEXP handle, SEXP iter, SEXP dtrain) {
+XGB_DLL SEXP XGBoosterUpdateOneIter_R(SEXP handle, SEXP iter, SEXP dtrain) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterUpdateOneIter(R_ExternalPtrAddr(handle),
                                   asInteger(iter),
@@ -290,7 +307,7 @@ SEXP XGBoosterUpdateOneIter_R(SEXP handle, SEXP iter, SEXP dtrain) {
   return R_NilValue;
 }
 
-SEXP XGBoosterBoostOneIter_R(SEXP handle, SEXP dtrain, SEXP grad, SEXP hess) {
+XGB_DLL SEXP XGBoosterBoostOneIter_R(SEXP handle, SEXP dtrain, SEXP grad, SEXP hess) {
   R_API_BEGIN();
   CHECK_EQ(length(grad), length(hess))
       << "gradient and hess must have same length";
@@ -313,7 +330,7 @@ SEXP XGBoosterBoostOneIter_R(SEXP handle, SEXP dtrain, SEXP grad, SEXP hess) {
   return R_NilValue;
 }
 
-SEXP XGBoosterEvalOneIter_R(SEXP handle, SEXP iter, SEXP dmats, SEXP evnames) {
+XGB_DLL SEXP XGBoosterEvalOneIter_R(SEXP handle, SEXP iter, SEXP dmats, SEXP evnames) {
   const char *ret;
   R_API_BEGIN();
   CHECK_EQ(length(dmats), length(evnames))
@@ -338,8 +355,8 @@ SEXP XGBoosterEvalOneIter_R(SEXP handle, SEXP iter, SEXP dmats, SEXP evnames) {
   return mkString(ret);
 }
 
-SEXP XGBoosterPredict_R(SEXP handle, SEXP dmat, SEXP option_mask,
-                        SEXP ntree_limit, SEXP training) {
+XGB_DLL SEXP XGBoosterPredict_R(SEXP handle, SEXP dmat, SEXP option_mask,
+                                SEXP ntree_limit, SEXP training) {
   SEXP ret;
   R_API_BEGIN();
   bst_ulong olen;
@@ -359,21 +376,60 @@ SEXP XGBoosterPredict_R(SEXP handle, SEXP dmat, SEXP option_mask,
   return ret;
 }
 
-SEXP XGBoosterLoadModel_R(SEXP handle, SEXP fname) {
+XGB_DLL SEXP XGBoosterPredictFromDMatrix_R(SEXP handle, SEXP dmat, SEXP json_config)  {
+  SEXP r_out_shape;
+  SEXP r_out_result;
+  SEXP r_out;
+
+  R_API_BEGIN();
+  char const *c_json_config = CHAR(asChar(json_config));
+
+  bst_ulong out_dim;
+  bst_ulong const *out_shape;
+  float const *out_result;
+  CHECK_CALL(XGBoosterPredictFromDMatrix(R_ExternalPtrAddr(handle),
+                                         R_ExternalPtrAddr(dmat), c_json_config,
+                                         &out_shape, &out_dim, &out_result));
+
+  r_out_shape = PROTECT(allocVector(INTSXP, out_dim));
+  size_t len = 1;
+  for (size_t i = 0; i < out_dim; ++i) {
+    INTEGER(r_out_shape)[i] = out_shape[i];
+    len *= out_shape[i];
+  }
+  r_out_result = PROTECT(allocVector(REALSXP, len));
+
+#pragma omp parallel for
+  for (omp_ulong i = 0; i < len; ++i) {
+    REAL(r_out_result)[i] = out_result[i];
+  }
+
+  r_out = PROTECT(allocVector(VECSXP, 2));
+
+  SET_VECTOR_ELT(r_out, 0, r_out_shape);
+  SET_VECTOR_ELT(r_out, 1, r_out_result);
+
+  R_API_END();
+  UNPROTECT(3);
+
+  return r_out;
+}
+
+XGB_DLL SEXP XGBoosterLoadModel_R(SEXP handle, SEXP fname) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterLoadModel(R_ExternalPtrAddr(handle), CHAR(asChar(fname))));
   R_API_END();
   return R_NilValue;
 }
 
-SEXP XGBoosterSaveModel_R(SEXP handle, SEXP fname) {
+XGB_DLL SEXP XGBoosterSaveModel_R(SEXP handle, SEXP fname) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterSaveModel(R_ExternalPtrAddr(handle), CHAR(asChar(fname))));
   R_API_END();
   return R_NilValue;
 }
 
-SEXP XGBoosterModelToRaw_R(SEXP handle) {
+XGB_DLL SEXP XGBoosterModelToRaw_R(SEXP handle) {
   SEXP ret;
   R_API_BEGIN();
   bst_ulong olen;
@@ -388,7 +444,7 @@ SEXP XGBoosterModelToRaw_R(SEXP handle) {
   return ret;
 }
 
-SEXP XGBoosterLoadModelFromRaw_R(SEXP handle, SEXP raw) {
+XGB_DLL SEXP XGBoosterLoadModelFromRaw_R(SEXP handle, SEXP raw) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterLoadModelFromBuffer(R_ExternalPtrAddr(handle),
                                           RAW(raw),
@@ -397,7 +453,7 @@ SEXP XGBoosterLoadModelFromRaw_R(SEXP handle, SEXP raw) {
   return R_NilValue;
 }
 
-SEXP XGBoosterSaveJsonConfig_R(SEXP handle) {
+XGB_DLL SEXP XGBoosterSaveJsonConfig_R(SEXP handle) {
   const char* ret;
   R_API_BEGIN();
   bst_ulong len {0};
@@ -408,14 +464,14 @@ SEXP XGBoosterSaveJsonConfig_R(SEXP handle) {
   return mkString(ret);
 }
 
-SEXP XGBoosterLoadJsonConfig_R(SEXP handle, SEXP value) {
+XGB_DLL SEXP XGBoosterLoadJsonConfig_R(SEXP handle, SEXP value) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterLoadJsonConfig(R_ExternalPtrAddr(handle), CHAR(asChar(value))));
   R_API_END();
   return R_NilValue;
 }
 
-SEXP XGBoosterSerializeToBuffer_R(SEXP handle) {
+XGB_DLL SEXP XGBoosterSerializeToBuffer_R(SEXP handle) {
   SEXP ret;
   R_API_BEGIN();
   bst_ulong out_len;
@@ -430,7 +486,7 @@ SEXP XGBoosterSerializeToBuffer_R(SEXP handle) {
   return ret;
 }
 
-SEXP XGBoosterUnserializeFromBuffer_R(SEXP handle, SEXP raw) {
+XGB_DLL SEXP XGBoosterUnserializeFromBuffer_R(SEXP handle, SEXP raw) {
   R_API_BEGIN();
   CHECK_CALL(XGBoosterUnserializeFromBuffer(R_ExternalPtrAddr(handle),
                                  RAW(raw),
@@ -439,7 +495,7 @@ SEXP XGBoosterUnserializeFromBuffer_R(SEXP handle, SEXP raw) {
   return R_NilValue;
 }
 
-SEXP XGBoosterDumpModel_R(SEXP handle, SEXP fmap, SEXP with_stats, SEXP dump_format) {
+XGB_DLL SEXP XGBoosterDumpModel_R(SEXP handle, SEXP fmap, SEXP with_stats, SEXP dump_format) {
   SEXP out;
   R_API_BEGIN();
   bst_ulong olen;
@@ -476,7 +532,7 @@ SEXP XGBoosterDumpModel_R(SEXP handle, SEXP fmap, SEXP with_stats, SEXP dump_for
   return out;
 }
 
-SEXP XGBoosterGetAttr_R(SEXP handle, SEXP name) {
+XGB_DLL SEXP XGBoosterGetAttr_R(SEXP handle, SEXP name) {
   SEXP out;
   R_API_BEGIN();
   int success;
@@ -496,7 +552,7 @@ SEXP XGBoosterGetAttr_R(SEXP handle, SEXP name) {
   return out;
 }
 
-SEXP XGBoosterSetAttr_R(SEXP handle, SEXP name, SEXP val) {
+XGB_DLL SEXP XGBoosterSetAttr_R(SEXP handle, SEXP name, SEXP val) {
   R_API_BEGIN();
   const char *v = isNull(val) ? nullptr : CHAR(asChar(val));
   CHECK_CALL(XGBoosterSetAttr(R_ExternalPtrAddr(handle),
@@ -505,7 +561,7 @@ SEXP XGBoosterSetAttr_R(SEXP handle, SEXP name, SEXP val) {
   return R_NilValue;
 }
 
-SEXP XGBoosterGetAttrNames_R(SEXP handle) {
+XGB_DLL SEXP XGBoosterGetAttrNames_R(SEXP handle) {
   SEXP out;
   R_API_BEGIN();
   bst_ulong len;
@@ -524,3 +580,51 @@ SEXP XGBoosterGetAttrNames_R(SEXP handle) {
   UNPROTECT(1);
   return out;
 }
+
+XGB_DLL SEXP XGBoosterFeatureScore_R(SEXP handle, SEXP json_config) {
+  SEXP out_features_sexp;
+  SEXP out_scores_sexp;
+  SEXP out_shape_sexp;
+  SEXP r_out;
+
+  R_API_BEGIN();
+  char const *c_json_config = CHAR(asChar(json_config));
+  bst_ulong out_n_features;
+  char const **out_features;
+
+  bst_ulong out_dim;
+  bst_ulong const *out_shape;
+  float const *out_scores;
+
+  CHECK_CALL(XGBoosterFeatureScore(R_ExternalPtrAddr(handle), c_json_config,
+                                   &out_n_features, &out_features,
+                                   &out_dim, &out_shape, &out_scores));
+
+  out_shape_sexp = PROTECT(allocVector(INTSXP, out_dim));
+  size_t len = 1;
+  for (size_t i = 0; i < out_dim; ++i) {
+    INTEGER(out_shape_sexp)[i] = out_shape[i];
+    len *= out_shape[i];
+  }
+
+  out_scores_sexp = PROTECT(allocVector(REALSXP, len));
+#pragma omp parallel for
+  for (omp_ulong i = 0; i < len; ++i) {
+    REAL(out_scores_sexp)[i] = out_scores[i];
+  }
+
+  out_features_sexp = PROTECT(allocVector(STRSXP, out_n_features));
+  for (size_t i = 0; i < out_n_features; ++i) {
+    SET_STRING_ELT(out_features_sexp, i, mkChar(out_features[i]));
+  }
+
+  r_out = PROTECT(allocVector(VECSXP, 3));
+  SET_VECTOR_ELT(r_out, 0, out_features_sexp);
+  SET_VECTOR_ELT(r_out, 1, out_shape_sexp);
+  SET_VECTOR_ELT(r_out, 2, out_scores_sexp);
+
+  R_API_END();
+  UNPROTECT(4);
+
+  return r_out;
+}

R-package/src/xgboost_R.h

@@ -47,10 +47,12 @@ XGB_DLL SEXP XGDMatrixCreateFromFile_R(SEXP fname, SEXP silent);
  * This assumes the matrix is stored in column major format
  * \param data R Matrix object
  * \param missing which value to represent missing value
+ * \param n_threads Number of threads used to construct DMatrix from dense matrix.
  * \return created dmatrix
  */
 XGB_DLL SEXP XGDMatrixCreateFromMat_R(SEXP mat,
-                                      SEXP missing);
+                                      SEXP missing,
+                                      SEXP n_threads);
 /*!
  * \brief create a matrix content from CSC format
  * \param indptr pointer to column headers
@@ -116,6 +118,14 @@ XGB_DLL SEXP XGDMatrixNumCol_R(SEXP handle);
  */
 XGB_DLL SEXP XGBoosterCreate_R(SEXP dmats);
 
+
+/*!
+ * \brief create xgboost learner, saving the pointer into an existing R object
+ * \param dmats a list of dmatrix handles that will be cached
+ * \param R_handle a clean R external pointer (not holding any object)
+ */
+XGB_DLL SEXP XGBoosterCreateInEmptyObj_R(SEXP dmats, SEXP R_handle);
+
 /*!
  * \brief set parameters
  * \param handle handle
@@ -156,7 +166,7 @@ XGB_DLL SEXP XGBoosterBoostOneIter_R(SEXP handle, SEXP dtrain, SEXP grad, SEXP h
 XGB_DLL SEXP XGBoosterEvalOneIter_R(SEXP handle, SEXP iter, SEXP dmats, SEXP evnames);
 
 /*!
- * \brief make prediction based on dmat
+ * \brief (Deprecated) make prediction based on dmat
  * \param handle handle
  * \param dmat data matrix
  * \param option_mask output_margin:1 predict_leaf:2
@@ -165,6 +175,16 @@ XGB_DLL SEXP XGBoosterEvalOneIter_R(SEXP handle, SEXP iter, SEXP dmats, SEXP evn
  */
 XGB_DLL SEXP XGBoosterPredict_R(SEXP handle, SEXP dmat, SEXP option_mask,
                                 SEXP ntree_limit, SEXP training);
+
+/*!
+ * \brief Run prediction on DMatrix, replacing `XGBoosterPredict_R`
+ * \param handle handle
+ * \param dmat data matrix
+ * \param json_config See `XGBoosterPredictFromDMatrix` in xgboost c_api.h
+ *
+ * \return A list containing 2 vectors, first one for shape while second one for prediction result.
+ */
+XGB_DLL SEXP XGBoosterPredictFromDMatrix_R(SEXP handle, SEXP dmat, SEXP json_config);
 /*!
  * \brief load model from existing file
  * \param handle handle
@@ -257,4 +277,12 @@ XGB_DLL SEXP XGBoosterSetAttr_R(SEXP handle, SEXP name, SEXP val);
  */
 XGB_DLL SEXP XGBoosterGetAttrNames_R(SEXP handle);
 
+/*!
+ * \brief Get feature scores from the model.
+ * \param json_config See `XGBoosterFeatureScore` in xgboost c_api.h
+ * \return A vector with the first element as feature names, second element as shape of
+ *         feature scores and thrid element as feature scores.
+ */
+XGB_DLL SEXP XGBoosterFeatureScore_R(SEXP handle, SEXP json_config);
+
 #endif  // XGBOOST_WRAPPER_R_H_ // NOLINT(*)

R-package/tests/testthat/test_basic.R

@@ -34,6 +34,10 @@ test_that("train and predict binary classification", {
   err_pred1 <- sum((pred1 > 0.5) != train$label) / length(train$label)
   err_log <- bst$evaluation_log[1, train_error]
   expect_lt(abs(err_pred1 - err_log), 10e-6)
+
+  pred2 <- predict(bst, train$data, iterationrange = c(1, 2))
+  expect_length(pred1, 6513)
+  expect_equal(pred1, pred2)
 })
 
 test_that("parameter validation works", {
@@ -143,6 +147,24 @@ test_that("train and predict softprob", {
   pred_labels <- max.col(mpred) - 1
   err <- sum(pred_labels != lb) / length(lb)
   expect_equal(bst$evaluation_log[1, train_merror], err, tolerance = 5e-6)
+
+  mpred1 <- predict(bst, as.matrix(iris[, -5]), reshape = TRUE, iterationrange = c(1, 2))
+  expect_equal(mpred, mpred1)
+
+  d <- cbind(
+    x1 = rnorm(100),
+    x2 = rnorm(100),
+    x3 = rnorm(100)
+  )
+  y <- sample.int(10, 100, replace = TRUE) - 1
+  dtrain <- xgb.DMatrix(data = d, info = list(label = y))
+  booster <- xgb.train(
+    params = list(tree_method = "hist"), data = dtrain, nrounds = 4, num_class = 10,
+    objective = "multi:softprob"
+  )
+  predt <- predict(booster, as.matrix(d), reshape = TRUE, strict_shape = FALSE)
+  expect_equal(ncol(predt), 10)
+  expect_equal(rowSums(predt), rep(1, 100), tolerance = 1e-7)
 })
 
 test_that("train and predict softmax", {
@@ -182,10 +204,8 @@ test_that("train and predict RF", {
   pred_err_20 <- sum((pred > 0.5) != lb) / length(lb)
   expect_equal(pred_err_20, pred_err)
 
-  #pred <- predict(bst, train$data, ntreelimit = 1)
-  #pred_err_1 <- sum((pred > 0.5) != lb)/length(lb)
-  #expect_lt(pred_err, pred_err_1)
-  #expect_lt(pred_err, 0.08)
+  pred1 <- predict(bst, train$data, iterationrange = c(1, 2))
+  expect_equal(pred, pred1)
 })
 
 test_that("train and predict RF with softprob", {
@@ -331,7 +351,7 @@ test_that("train and predict with non-strict classes", {
   expect_error(pr <- predict(bst, train_dense), regexp = NA)
   expect_equal(pr0, pr)
 
-  # when someone inhertis from xgb.Booster, it should still be possible to use it as xgb.Booster
+  # when someone inherits from xgb.Booster, it should still be possible to use it as xgb.Booster
   class(bst) <- c('super.Booster', 'xgb.Booster')
   expect_error(pr <- predict(bst, train_dense), regexp = NA)
   expect_equal(pr0, pr)
@@ -346,7 +366,7 @@ test_that("max_delta_step works", {
   bst1 <- xgb.train(param, dtrain, nrounds, watchlist, verbose = 1)
   # model with restricted max_delta_step
   bst2 <- xgb.train(param, dtrain, nrounds, watchlist, verbose = 1, max_delta_step = 1)
-  # the no-restriction model is expected to have consistently lower loss during the initial interations
+  # the no-restriction model is expected to have consistently lower loss during the initial iterations
   expect_true(all(bst1$evaluation_log$train_logloss < bst2$evaluation_log$train_logloss))
   expect_lt(mean(bst1$evaluation_log$train_logloss) / mean(bst2$evaluation_log$train_logloss), 0.8)
 })
@@ -385,3 +405,57 @@ test_that("Configuration works", {
   reloaded_config <- xgb.config(bst)
   expect_equal(config, reloaded_config);
 })
+
+test_that("strict_shape works", {
+  n_rounds <- 2
+
+  test_strict_shape <- function(bst, X, n_groups) {
+    predt <- predict(bst, X, strict_shape = TRUE)
+    margin <- predict(bst, X, outputmargin = TRUE, strict_shape = TRUE)
+    contri <- predict(bst, X, predcontrib = TRUE, strict_shape = TRUE)
+    interact <- predict(bst, X, predinteraction = TRUE, strict_shape = TRUE)
+    leaf <- predict(bst, X, predleaf = TRUE, strict_shape = TRUE)
+
+    n_rows <- nrow(X)
+    n_cols <- ncol(X)
+
+    expect_equal(dim(predt), c(n_groups, n_rows))
+    expect_equal(dim(margin), c(n_groups, n_rows))
+    expect_equal(dim(contri), c(n_cols + 1, n_groups, n_rows))
+    expect_equal(dim(interact), c(n_cols + 1, n_cols + 1, n_groups, n_rows))
+    expect_equal(dim(leaf), c(1, n_groups, n_rounds, n_rows))
+
+    if (n_groups != 1) {
+      for (g in seq_len(n_groups)) {
+        expect_lt(max(abs(colSums(contri[, g, ]) - margin[g, ])), 1e-5)
+      }
+    }
+  }
+
+  test_iris <- function() {
+    y <- as.numeric(iris$Species) - 1
+    X <- as.matrix(iris[, -5])
+
+    bst <- xgboost(data = X, label = y,
+                   max_depth = 2, nrounds = n_rounds,
+                   objective = "multi:softprob", num_class = 3, eval_metric = "merror")
+
+    test_strict_shape(bst, X, 3)
+  }
+
+
+  test_agaricus <- function() {
+    data(agaricus.train, package = 'xgboost')
+    X <- agaricus.train$data
+    y <- agaricus.train$label
+
+    bst <- xgboost(data = X, label = y, max_depth = 2,
+                   nrounds = n_rounds, objective = "binary:logistic",
+                   eval_metric = 'error', eval_metric = 'auc', eval_metric = "logloss")
+
+    test_strict_shape(bst, X, 1)
+  }
+
+  test_iris()
+  test_agaricus()
+})

R-package/tests/testthat/test_helpers.R

@@ -110,7 +110,7 @@ test_that("predict feature contributions works", {
   pred <- predict(bst.GLM, sparse_matrix, outputmargin = TRUE)
   expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5)
   # manual calculation of linear terms
-  coefs <- xgb.dump(bst.GLM)[-c(1, 2, 4)] %>% as.numeric
+  coefs <- as.numeric(xgb.dump(bst.GLM)[-c(1, 2, 4)])
   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),
@@ -130,7 +130,11 @@ test_that("predict feature contributions works", {
   pred <- predict(mbst.GLM, as.matrix(iris[, -5]), outputmargin = TRUE, reshape = TRUE)
   pred_contr <- predict(mbst.GLM, as.matrix(iris[, -5]), predcontrib = TRUE)
   expect_length(pred_contr, 3)
-  coefs_all <- xgb.dump(mbst.GLM)[-c(1, 2, 6)] %>% as.numeric %>% matrix(ncol = 3, byrow = TRUE)
+  coefs_all <- matrix(
+    data = as.numeric(xgb.dump(mbst.GLM)[-c(1, 2, 6)]),
+    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])), float_tolerance)
@@ -238,12 +242,13 @@ if (grepl('Windows', Sys.info()[['sysname']]) ||
 test_that("xgb.Booster serializing as R object works", {
   saveRDS(bst.Tree, 'xgb.model.rds')
   bst <- readRDS('xgb.model.rds')
-  if (file.exists('xgb.model.rds')) file.remove('xgb.model.rds')
   dtrain <- xgb.DMatrix(sparse_matrix, label = label)
   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')
   if (file.exists('xgb.model')) file.remove('xgb.model')
+  bst <- readRDS('xgb.model.rds')
+  if (file.exists('xgb.model.rds')) file.remove('xgb.model.rds')
   nil_ptr <- new("externalptr")
   class(nil_ptr) <- "xgb.Booster.handle"
   expect_true(identical(bst$handle, nil_ptr))

R-package/tests/testthat/test_interactions.R

@@ -1,7 +1,6 @@
 context('Test prediction of feature interactions')
 
 require(xgboost)
-require(magrittr)
 
 set.seed(123)
 
@@ -32,7 +31,7 @@ test_that("predict feature interactions works", {
   cont <- predict(b, dm, predcontrib = TRUE)
   expect_equal(dim(cont), c(N, P + 1))
   # make sure for each row they add up to marginal predictions
-  max(abs(rowSums(cont) - pred)) %>% expect_lt(0.001)
+  expect_lt(max(abs(rowSums(cont) - pred)), 0.001)
   # Hand-construct the 'ground truth' feature contributions:
   gt_cont <- cbind(
       2. * X[, 1],
@@ -52,21 +51,24 @@ test_that("predict feature interactions works", {
   expect_equal(dimnames(intr), list(NULL, cn, cn))
 
   # check the symmetry
-  max(abs(aperm(intr, c(1, 3, 2)) - intr)) %>% expect_lt(0.00001)
+  expect_lt(max(abs(aperm(intr, c(1, 3, 2)) - intr)), 0.00001)
 
   # sums WRT columns must be close to feature contributions
-  max(abs(apply(intr, c(1, 2), sum) - cont)) %>% expect_lt(0.00001)
+  expect_lt(max(abs(apply(intr, c(1, 2), sum) - cont)), 0.00001)
 
   # diagonal terms for features 3,4,5 must be close to zero
-  Reduce(max, sapply(3:P, function(i) max(abs(intr[, i, i])))) %>% expect_lt(0.05)
+  expect_lt(Reduce(max, sapply(3:P, function(i) max(abs(intr[, i, i])))), 0.05)
 
   # BIAS must have no interactions
-  max(abs(intr[, 1:P, P + 1])) %>% expect_lt(0.00001)
+  expect_lt(max(abs(intr[, 1:P, P + 1])), 0.00001)
 
   # interactions other than 2 x 3 must be close to zero
   intr23 <- intr
   intr23[, 2, 3] <- 0
-  Reduce(max, sapply(1:P, function(i) max(abs(intr23[, i, (i + 1):(P + 1)])))) %>% expect_lt(0.05)
+  expect_lt(
+    Reduce(max, sapply(1:P, function(i) max(abs(intr23[, i, (i + 1):(P + 1)])))),
+    0.05
+  )
 
   # Construct the 'ground truth' contributions of interactions directly from the linear terms:
   gt_intr <- array(0, c(N, P + 1, P + 1))
@@ -119,23 +121,39 @@ test_that("multiclass feature interactions work", {
   dm <- xgb.DMatrix(as.matrix(iris[, -5]), label = as.numeric(iris$Species) - 1)
   param <- list(eta = 0.1, max_depth = 4, objective = 'multi:softprob', num_class = 3)
   b <- xgb.train(param, dm, 40)
-  pred <- predict(b, dm, outputmargin = TRUE) %>% array(c(3, 150)) %>% t
+  pred <- t(
+    array(
+      data = predict(b, dm, outputmargin = TRUE),
+      dim = c(3, 150)
+    )
+  )
 
   # SHAP contributions:
   cont <- predict(b, dm, predcontrib = TRUE)
   expect_length(cont, 3)
   # rewrap them as a 3d array
-  cont <- unlist(cont) %>% array(c(150, 5, 3))
+  cont <- array(
+    data = unlist(cont),
+    dim = c(150, 5,  3)
+  )
+
   # make sure for each row they add up to marginal predictions
-  max(abs(apply(cont, c(1, 3), sum) - pred)) %>% expect_lt(0.001)
+  expect_lt(max(abs(apply(cont, c(1, 3), sum) - pred)), 0.001)
 
   # SHAP interaction contributions:
   intr <- predict(b, dm, predinteraction = TRUE)
   expect_length(intr, 3)
   # rewrap them as a 4d array
-  intr <- unlist(intr) %>% array(c(150, 5, 5, 3)) %>% aperm(c(4, 1, 2, 3)) # [grp, row, col, col]
+  intr <- aperm(
+    a = array(
+      data = unlist(intr),
+      dim = c(150, 5, 5, 3)
+    ),
+    perm = c(4, 1, 2, 3)  # [grp, row, col, col]
+  )
+
   # check the symmetry
-  max(abs(aperm(intr, c(1, 2, 4, 3)) - intr)) %>% expect_lt(0.00001)
+  expect_lt(max(abs(aperm(intr, c(1, 2, 4, 3)) - intr)), 0.00001)
   # sums WRT columns must be close to feature contributions
-  max(abs(apply(intr, c(1, 2, 3), sum) - aperm(cont, c(3, 1, 2)))) %>% expect_lt(0.00001)
+  expect_lt(max(abs(apply(intr, c(1, 2, 3), sum) - aperm(cont, c(3, 1, 2)))), 0.00001)
 })

R-package/tests/testthat/test_model_compatibility.R

@@ -83,6 +83,7 @@ test_that("Models from previous versions of XGBoost can be loaded", {
       if (is_rds && compareVersion(model_xgb_ver, '1.1.1.1') < 0) {
         booster <- readRDS(model_file)
         expect_warning(predict(booster, newdata = pred_data))
+        booster <- readRDS(model_file)
         expect_warning(run_booster_check(booster, name))
       } else {
         if (is_rds) {

R-package/tests/testthat/test_monotone.R

@@ -19,5 +19,5 @@ test_that("monotone constraints for regression", {
     pred.ord <- pred[ind]
     expect_true({
         !any(diff(pred.ord) > 0)
-    }, "Monotone Contraint Satisfied")
+    }, "Monotone constraint satisfied")
 })

R-package/tests/testthat/test_poisson_regression.R

@@ -1,9 +1,9 @@
-context('Test poisson regression model')
+context('Test Poisson regression model')
 
 require(xgboost)
 set.seed(1994)
 
-test_that("poisson regression works", {
+test_that("Poisson regression works", {
   data(mtcars)
   bst <- xgboost(data = as.matrix(mtcars[, -11]), label = mtcars[, 11],
                 objective = 'count:poisson', nrounds = 10, verbose = 0)

R-package/vignettes/discoverYourData.Rmd

@@ -1,5 +1,5 @@
 ---
-title: "Understand your dataset with Xgboost"
+title: "Understand your dataset with XGBoost"
 output:
   rmarkdown::html_vignette:
     css: vignette.css
@@ -18,9 +18,9 @@ Understand your dataset with XGBoost
 Introduction
 ------------
 
-The purpose of this vignette is to show you how to use **Xgboost** to discover and understand your own dataset better.
+The purpose of this vignette is to show you how to use **XGBoost** to discover and understand your own dataset better.
 
-This vignette is not about predicting anything (see [Xgboost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)). We will explain how to use **Xgboost** to highlight the *link* between the *features* of your data and the *outcome*.
+This vignette is not about predicting anything (see [XGBoost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)). We will explain how to use **XGBoost** to highlight the *link* between the *features* of your data and the *outcome*.
 
 Package loading:
 
@@ -39,7 +39,7 @@ Preparation of the dataset
 ### Numeric v.s. categorical variables
 
 
-**Xgboost** manages only `numeric` vectors.
+**XGBoost** manages only `numeric` vectors.
 
 What to do when you have *categorical* data?
 
@@ -66,7 +66,7 @@ data(Arthritis)
 df <- data.table(Arthritis, keep.rownames = FALSE)
 ```
 
-> `data.table` is 100% compliant with **R** `data.frame` but its syntax is more consistent and its performance for large dataset is [best in class](https://stackoverflow.com/questions/21435339/data-table-vs-dplyr-can-one-do-something-well-the-other-cant-or-does-poorly) (`dplyr` from **R** and `Pandas` from **Python** [included](https://github.com/Rdatatable/data.table/wiki/Benchmarks-%3A-Grouping)). Some parts of **Xgboost** **R** package use `data.table`.
+> `data.table` is 100% compliant with **R** `data.frame` but its syntax is more consistent and its performance for large dataset is [best in class](https://stackoverflow.com/questions/21435339/data-table-vs-dplyr-can-one-do-something-well-the-other-cant-or-does-poorly) (`dplyr` from **R** and `Pandas` from **Python** [included](https://github.com/Rdatatable/data.table/wiki/Benchmarks-%3A-Grouping)). Some parts of **XGBoost** **R** package use `data.table`.
 
 The first thing we want to do is to have a look to the first few lines of the `data.table`:
 
@@ -166,7 +166,7 @@ output_vector = df[,Improved] == "Marked"
 Build the model
 ---------------
 
-The code below is very usual. For more information, you can look at the documentation of `xgboost` function (or at the vignette [Xgboost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)).
+The code below is very usual. For more information, you can look at the documentation of `xgboost` function (or at the vignette [XGBoost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)).
 
 ```{r}
 bst <- xgboost(data = sparse_matrix, label = output_vector, max_depth = 4,
@@ -176,7 +176,7 @@ bst <- xgboost(data = sparse_matrix, label = output_vector, max_depth = 4,
 
 You can see some `train-error: 0.XXXXX` lines followed by a number. It decreases. Each line shows how well the model explains your data. Lower is better.
 
-A model which fits too well may [overfit](https://en.wikipedia.org/wiki/Overfitting) (meaning it copy/paste too much the past, and won't be that good to predict the future).
+A small value for training error may be a symptom of [overfitting](https://en.wikipedia.org/wiki/Overfitting), meaning the model will not accurately predict the future values.
 
 > Here you can see the numbers decrease until line 7 and then increase.
 >
@@ -304,19 +304,19 @@ Linear model may not be that smart in this scenario.
 Special Note: What about Random Forests™?
 -----------------------------------------
 
-As you may know, [Random Forests™](https://en.wikipedia.org/wiki/Random_forest) algorithm is cousin with boosting and both are part of the [ensemble learning](https://en.wikipedia.org/wiki/Ensemble_learning) family.
+As you may know, [Random Forests](https://en.wikipedia.org/wiki/Random_forest) algorithm is cousin with boosting and both are part of the [ensemble learning](https://en.wikipedia.org/wiki/Ensemble_learning) family.
 
-Both trains several decision trees for one dataset. The *main* difference is that in Random Forests™, trees are independent and in boosting, the tree `N+1` focus its learning on the loss (<=> what has not been well modeled by the tree `N`).
+Both trains several decision trees for one dataset. The *main* difference is that in Random Forests, trees are independent and in boosting, the tree `N+1` focus its learning on the loss (<=> what has not been well modeled by the tree `N`).
 
 This difference have an impact on a corner case in feature importance analysis: the *correlated features*.
 
-Imagine two features perfectly correlated, feature `A` and feature `B`. For one specific tree, if the algorithm needs one of them, it will choose randomly (true in both boosting and Random Forests™).
+Imagine two features perfectly correlated, feature `A` and feature `B`. For one specific tree, if the algorithm needs one of them, it will choose randomly (true in both boosting and Random Forests).
 
-However, in Random Forests™ this random choice will be done for each tree, because each tree is independent from the others. Therefore, approximatively, depending of your parameters, 50% of the trees will choose feature `A` and the other 50% will choose feature `B`. So the *importance* of the information contained in `A` and `B` (which is the same, because they are perfectly correlated) is diluted in `A` and `B`. So you won't easily know this information is important to predict what you want to predict! It is even worse when you have 10 correlated features...
+However, in Random Forests this random choice will be done for each tree, because each tree is independent from the others. Therefore, approximatively, depending of your parameters, 50% of the trees will choose feature `A` and the other 50% will choose feature `B`. So the *importance* of the information contained in `A` and `B` (which is the same, because they are perfectly correlated) is diluted in `A` and `B`. So you won't easily know this information is important to predict what you want to predict! It is even worse when you have 10 correlated features...
 
 In boosting, when a specific link between feature and outcome have been learned by the algorithm, it will try to not refocus on it (in theory it is what happens, reality is not always that simple). Therefore, all the importance will be on feature `A` or on feature `B` (but not both). You will know that one feature have an important role in the link between the observations and the label. It is still up to you to search for the correlated features to the one detected as important if you need to know all of them.
 
-If you want to try Random Forests™ algorithm, you can tweak Xgboost parameters!
+If you want to try Random Forests algorithm, you can tweak XGBoost parameters!
 
 For instance, to compute a model with 1000 trees, with a 0.5 factor on sampling rows and columns:
 
@@ -326,7 +326,7 @@ data(agaricus.test, package='xgboost')
 train <- agaricus.train
 test <- agaricus.test
 
-#Random Forest™ - 1000 trees
+#Random Forest - 1000 trees
 bst <- xgboost(data = train$data, label = train$label, max_depth = 4, num_parallel_tree = 1000, subsample = 0.5, colsample_bytree =0.5, nrounds = 1, objective = "binary:logistic")
 
 #Boosting - 3 rounds
@@ -335,4 +335,4 @@ bst <- xgboost(data = train$data, label = train$label, max_depth = 4, nrounds =
 
 > Note that the parameter `round` is set to `1`.
 
-> [**Random Forests™**](https://www.stat.berkeley.edu/~breiman/RandomForests/cc_papers.htm) is a trademark of Leo Breiman and Adele Cutler and is licensed exclusively to Salford Systems for the commercial release of the software.
+> [**Random Forests**](https://www.stat.berkeley.edu/~breiman/RandomForests/cc_papers.htm) is a trademark of Leo Breiman and Adele Cutler and is licensed exclusively to Salford Systems for the commercial release of the software.

R-package/vignettes/xgboostPresentation.Rmd

@@ -1,5 +1,5 @@
 ---
-title: "Xgboost presentation"
+title: "XGBoost presentation"
 output:
   rmarkdown::html_vignette:
     css: vignette.css
@@ -8,7 +8,7 @@ output:
 bibliography: xgboost.bib
 author: Tianqi Chen, Tong He, Michaël Benesty
 vignette: >
-  %\VignetteIndexEntry{Xgboost presentation}
+  %\VignetteIndexEntry{XGBoost presentation}
   %\VignetteEngine{knitr::rmarkdown}
   \usepackage[utf8]{inputenc}
 ---
@@ -19,9 +19,9 @@ XGBoost R Tutorial
 ## Introduction
 
 
-**Xgboost** is short for e**X**treme **G**radient **Boost**ing package.
+**XGBoost** is short for e**X**treme **G**radient **Boost**ing package.
 
-The purpose of this Vignette is to show you how to use **Xgboost** to build a model and make predictions.
+The purpose of this Vignette is to show you how to use **XGBoost** to build a model and make predictions.
 
 It is an efficient and scalable implementation of gradient boosting framework by @friedman2000additive and @friedman2001greedy. Two solvers are included:
 
@@ -46,10 +46,10 @@ It has several features:
 ## Installation
 
 
-### Github version
+### GitHub version
 
 
-For weekly updated version (highly recommended), install from *Github*:
+For weekly updated version (highly recommended), install from *GitHub*:
 
 ```{r installGithub, eval=FALSE}
 install.packages("drat", repos="https://cran.rstudio.com")
@@ -82,7 +82,7 @@ require(xgboost)
 ### Dataset presentation
 
 
-In this example, we are aiming to predict whether a mushroom can be eaten or not (like in many tutorials, example data are the the same as you will use on in your every day life :-).
+In this example, we are aiming to predict whether a mushroom can be eaten or not (like in many tutorials, example data are the same as you will use on in your every day life :-).
 
 Mushroom data is cited from UCI Machine Learning Repository. @Bache+Lichman:2013.
 
@@ -148,7 +148,7 @@ We will train decision tree model using the following parameters:
 
 * `objective = "binary:logistic"`: we will train a binary classification model ;
 * `max_depth = 2`: the trees won't be deep, because our case is very simple ;
-* `nthread = 2`: the number of cpu threads we are going to use;
+* `nthread = 2`: the number of CPU threads we are going to use;
 * `nrounds = 2`: there will be two passes on the data, the second one will enhance the model by further reducing the difference between ground truth and prediction.
 
 ```{r trainingSparse, message=F, warning=F}
@@ -180,7 +180,7 @@ bstDMatrix <- xgboost(data = dtrain, max_depth = 2, eta = 1, nthread = 2, nround
 
 **XGBoost** has several features to help you to view how the learning progress internally. The purpose is to help you to set the best parameters, which is the key of your model quality.
 
-One of the simplest way to see the training progress is to set the `verbose` option (see below for more advanced technics).
+One of the simplest way to see the training progress is to set the `verbose` option (see below for more advanced techniques).
 
 ```{r trainingVerbose0, message=T, warning=F}
 # verbose = 0, no message
@@ -253,7 +253,7 @@ The most important thing to remember is that **to do a classification, you just
 
 *Multiclass* classification works in a similar way.
 
-This metric is **`r round(err, 2)`** and is pretty low: our yummly mushroom model works well!
+This metric is **`r round(err, 2)`** and is pretty low: our yummy mushroom model works well!
 
 ## Advanced features
 

R-package/vignettes/xgboostfromJSON.Rmd

@@ -16,7 +16,7 @@ XGBoost from JSON
 
 ## Introduction
 
-The purpose of this Vignette is to show you how to correctly load and work with an **Xgboost** model that has been dumped to JSON.  **Xgboost** internally converts all data to [32-bit floats](https://en.wikipedia.org/wiki/Single-precision_floating-point_format), and the values dumped to JSON are decimal representations of these values.  When working with a model that has been parsed from a JSON file, care must be taken to correctly treat:
+The purpose of this Vignette is to show you how to correctly load and work with an **XGBoost** model that has been dumped to JSON.  **XGBoost** internally converts all data to [32-bit floats](https://en.wikipedia.org/wiki/Single-precision_floating-point_format), and the values dumped to JSON are decimal representations of these values.  When working with a model that has been parsed from a JSON file, care must be taken to correctly treat:
 
 - the input data, which should be converted to 32-bit floats
 - any 32-bit floats that were stored in JSON as decimal representations
@@ -172,9 +172,9 @@ bst_from_json_preds <- ifelse(fl(data$dates)<fl(node$split_condition),
 bst_preds == bst_from_json_preds
 ```
 
-None are exactly equal again.  What is going on here?  Well, since we are using the value `1` in the calcuations, we have introduced a double into the calculation.  Because of this, all float values are promoted to 64-bit doubles and the 64-bit version of the exponentiation operator `exp` is also used.  On the other hand, xgboost uses the 32-bit version of the exponentation operator in its [sigmoid function](https://github.com/dmlc/xgboost/blob/54980b8959680a0da06a3fc0ec776e47c8cbb0a1/src/common/math.h#L25-L27).
+None are exactly equal again.  What is going on here?  Well, since we are using the value `1` in the calculations, we have introduced a double into the calculation.  Because of this, all float values are promoted to 64-bit doubles and the 64-bit version of the exponentiation operator `exp` is also used.  On the other hand, xgboost uses the 32-bit version of the exponentiation operator in its [sigmoid function](https://github.com/dmlc/xgboost/blob/54980b8959680a0da06a3fc0ec776e47c8cbb0a1/src/common/math.h#L25-L27).
 
-How do we fix this?  We have to ensure we use the correct datatypes everywhere and the correct operators.  If we use only floats, the float library that we have loaded will ensure the 32-bit float exponention operator is applied. 
+How do we fix this?  We have to ensure we use the correct data types everywhere and the correct operators.  If we use only floats, the float library that we have loaded will ensure the 32-bit float exponentiation operator is applied. 
 ```{r}
 # calculate the predictions casting doubles to floats
 bst_from_json_preds <- ifelse(fl(data$dates)<fl(node$split_condition),

README.md

@@ -2,7 +2,6 @@
 ===========
 [![Build Status](https://xgboost-ci.net/job/xgboost/job/master/badge/icon)](https://xgboost-ci.net/blue/organizations/jenkins/xgboost/activity)
 [![Build Status](https://img.shields.io/travis/dmlc/xgboost.svg?label=build&logo=travis&branch=master)](https://travis-ci.org/dmlc/xgboost)
-[![Build Status](https://ci.appveyor.com/api/projects/status/5ypa8vaed6kpmli8?svg=true)](https://ci.appveyor.com/project/tqchen/xgboost)
 [![XGBoost-CI](https://github.com/dmlc/xgboost/workflows/XGBoost-CI/badge.svg?branch=master)](https://github.com/dmlc/xgboost/actions)
 [![Documentation Status](https://readthedocs.org/projects/xgboost/badge/?version=latest)](https://xgboost.readthedocs.org)
 [![GitHub license](http://dmlc.github.io/img/apache2.svg)](./LICENSE)
@@ -25,7 +24,7 @@ The same code runs on major distributed environment (Kubernetes, Hadoop, SGE, MP
 
 License
 -------
-© Contributors, 2019. Licensed under an [Apache-2](https://github.com/dmlc/xgboost/blob/master/LICENSE) license.
+© Contributors, 2021. Licensed under an [Apache-2](https://github.com/dmlc/xgboost/blob/master/LICENSE) license.
 
 Contribute to XGBoost
 ---------------------

amalgamation/xgboost-all0.cc

@@ -37,17 +37,16 @@
 #include "../src/data/simple_dmatrix.cc"
 #include "../src/data/sparse_page_raw_format.cc"
 #include "../src/data/ellpack_page.cc"
-#include "../src/data/ellpack_page_source.cc"
+#include "../src/data/gradient_index.cc"
+#include "../src/data/gradient_index_page_source.cc"
+#include "../src/data/gradient_index_format.cc"
+#include "../src/data/sparse_page_dmatrix.cc"
+#include "../src/data/proxy_dmatrix.cc"
 
 // prediction
 #include "../src/predictor/predictor.cc"
 #include "../src/predictor/cpu_predictor.cc"
 
-#if DMLC_ENABLE_STD_THREAD
-#include "../src/data/sparse_page_dmatrix.cc"
-#include "../src/data/sparse_page_source.cc"
-#endif
-
 // trees
 #include "../src/tree/param.cc"
 #include "../src/tree/tree_model.cc"

appveyor.yml

@@ -1,71 +0,0 @@
-environment:
-    matrix:
-        - target: msvc
-          ver: 2015
-          generator: "Visual Studio 14 2015 Win64"
-          configuration: Debug
-        - target: msvc
-          ver: 2015
-          generator: "Visual Studio 14 2015 Win64"
-          configuration: Release
-        - target: mingw
-          generator: "Unix Makefiles"
-
-#matrix:
-#    fast_finish: true
-
-platform:
-    - x64
-
-install:
-    - git submodule update --init --recursive
-    # MinGW
-    - set PATH=C:\msys64\mingw64\bin;C:\msys64\usr\bin;%PATH%
-    - gcc -v
-    - ls -l C:\
-    # Miniconda3
-    - call C:\Miniconda3-x64\Scripts\activate.bat
-    - conda info
-    - where python
-    - python --version
-    # do python build for mingw and one of the msvc jobs
-    - set DO_PYTHON=off
-    - if /i "%target%" == "mingw" set DO_PYTHON=on
-    - if /i "%target%_%ver%_%configuration%" == "msvc_2015_Release" set DO_PYTHON=on
-    - if /i "%DO_PYTHON%" == "on" (
-        conda config --set always_yes true &&
-        conda update -q conda &&
-        conda install -y numpy scipy pandas matplotlib pytest scikit-learn graphviz python-graphviz hypothesis
-      )
-    - set PATH=C:\Miniconda3-x64\Library\bin\graphviz;%PATH%
-
-build_script:
-    - cd %APPVEYOR_BUILD_FOLDER%
-    - if /i "%target%" == "msvc" (
-        mkdir build_msvc%ver% &&
-        cd build_msvc%ver% &&
-        cmake .. -G"%generator%" -DCMAKE_CONFIGURATION_TYPES="Release;Debug;" &&
-        msbuild xgboost.sln
-      )
-    - if /i "%target%" == "mingw" (
-        mkdir build_mingw &&
-        cd build_mingw &&
-        cmake .. -G"%generator%" &&
-        make -j2
-      )
-    # Python package
-    - if /i "%DO_PYTHON%" == "on" (
-        cd %APPVEYOR_BUILD_FOLDER%\python-package &&
-        python setup.py install &&
-        mkdir wheel &&
-        python setup.py bdist_wheel --universal --plat-name win-amd64 -d wheel
-      )
-
-test_script:
-    - cd %APPVEYOR_BUILD_FOLDER%
-    - if /i "%DO_PYTHON%" == "on" python -m pytest tests/python
-
-artifacts:
-    # binary Python wheel package
-    - path: '**\*.whl'
-      name: Bits

cmake/Python_version.in

@@ -1 +1 @@
-@xgboost_VERSION_MAJOR@.@xgboost_VERSION_MINOR@.@xgboost_VERSION_PATCH@-SNAPSHOT
+@xgboost_VERSION_MAJOR@.@xgboost_VERSION_MINOR@.@xgboost_VERSION_PATCH@

cmake/RPackageInstall.cmake.in

@@ -27,7 +27,7 @@ file(WRITE "${build_dir}/R-package/src/Makevars.win" "all:")
 
 # Install dependencies
 set(XGB_DEPS_SCRIPT
-    "deps = setdiff(c('data.table', 'magrittr', 'stringi'), rownames(installed.packages())); if(length(deps)>0) install.packages(deps, repo = 'https://cloud.r-project.org/')")
+    "deps = setdiff(c('data.table', 'jsonlite', 'Matrix'), rownames(installed.packages())); if(length(deps)>0) install.packages(deps, repo = 'https://cloud.r-project.org/')")
 check_call(COMMAND "${LIBR_EXECUTABLE}" -q -e "${XGB_DEPS_SCRIPT}")
 
 # Install the XGBoost R package

cmake/Utils.cmake

@@ -90,7 +90,9 @@ function(format_gencode_flags flags out)
   endif()
   # Set up architecture flags
   if(NOT flags)
-    if (CUDA_VERSION VERSION_GREATER_EQUAL "11.0")
+    if (CUDA_VERSION VERSION_GREATER_EQUAL "11.1")
+      set(flags "50;52;60;61;70;75;80;86")
+    elseif (CUDA_VERSION VERSION_GREATER_EQUAL "11.0")
       set(flags "35;50;52;60;61;70;75;80")
     elseif(CUDA_VERSION VERSION_GREATER_EQUAL "10.0")
       set(flags "35;50;52;60;61;70;75")
@@ -130,9 +132,6 @@ endmacro()
 
 # Set CUDA related flags to target.  Must be used after code `format_gencode_flags`.
 function(xgboost_set_cuda_flags target)
-  find_package(OpenMP REQUIRED)
-  target_link_libraries(${target} PUBLIC OpenMP::OpenMP_CXX)
-
   target_compile_options(${target} PRIVATE
     $<$<COMPILE_LANGUAGE:CUDA>:--expt-extended-lambda>
     $<$<COMPILE_LANGUAGE:CUDA>:--expt-relaxed-constexpr>
@@ -155,8 +154,13 @@ function(xgboost_set_cuda_flags target)
     enable_nvtx(${target})
   endif (USE_NVTX)
 
-  target_compile_definitions(${target} PRIVATE -DXGBOOST_USE_CUDA=1 -DTHRUST_IGNORE_CUB_VERSION_CHECK=1)
-  target_include_directories(${target} PRIVATE ${xgboost_SOURCE_DIR}/cub/)
+  if (NOT BUILD_WITH_CUDA_CUB)
+    target_compile_definitions(${target} PRIVATE -DXGBOOST_USE_CUDA=1 -DTHRUST_IGNORE_CUB_VERSION_CHECK=1)
+    target_include_directories(${target} PRIVATE ${xgboost_SOURCE_DIR}/cub/ ${xgboost_SOURCE_DIR}/gputreeshap)
+  else ()
+    target_compile_definitions(${target} PRIVATE -DXGBOOST_USE_CUDA=1)
+    target_include_directories(${target} PRIVATE ${xgboost_SOURCE_DIR}/gputreeshap)
+  endif (NOT BUILD_WITH_CUDA_CUB)
 
   if (MSVC)
     target_compile_options(${target} PRIVATE
@@ -167,16 +171,111 @@ function(xgboost_set_cuda_flags target)
     CUDA_STANDARD 14
     CUDA_STANDARD_REQUIRED ON
     CUDA_SEPARABLE_COMPILATION OFF)
+endfunction(xgboost_set_cuda_flags)
 
-  if (HIDE_CXX_SYMBOLS)
-    target_compile_options(${target} PRIVATE
-      $<$<COMPILE_LANGUAGE:CUDA>:-Xcompiler=-fvisibility=hidden>)
-  endif (HIDE_CXX_SYMBOLS)
-
-  if (USE_NCCL)
-    find_package(Nccl REQUIRED)
+macro(xgboost_link_nccl target)
+  if (BUILD_STATIC_LIB)
+    target_include_directories(${target} PUBLIC ${NCCL_INCLUDE_DIR})
+    target_compile_definitions(${target} PUBLIC -DXGBOOST_USE_NCCL=1)
+    target_link_libraries(${target} PUBLIC ${NCCL_LIBRARY})
+  else ()
     target_include_directories(${target} PRIVATE ${NCCL_INCLUDE_DIR})
     target_compile_definitions(${target} PRIVATE -DXGBOOST_USE_NCCL=1)
-    target_link_libraries(${target} PUBLIC ${NCCL_LIBRARY})
+    target_link_libraries(${target} PRIVATE ${NCCL_LIBRARY})
+  endif (BUILD_STATIC_LIB)
+endmacro(xgboost_link_nccl)
+
+# compile options
+macro(xgboost_target_properties target)
+  set_target_properties(${target} PROPERTIES
+    CXX_STANDARD 14
+    CXX_STANDARD_REQUIRED ON
+    POSITION_INDEPENDENT_CODE ON)
+  if (HIDE_CXX_SYMBOLS)
+    #-- Hide all C++ symbols
+    set_target_properties(${target} PROPERTIES
+      C_VISIBILITY_PRESET hidden
+      CXX_VISIBILITY_PRESET hidden
+      CUDA_VISIBILITY_PRESET hidden
+    )
+  endif (HIDE_CXX_SYMBOLS)
+
+  if (ENABLE_ALL_WARNINGS)
+    target_compile_options(${target} PUBLIC
+      $<IF:$<COMPILE_LANGUAGE:CUDA>,-Xcompiler=-Wall -Xcompiler=-Wextra,-Wall -Wextra>
+    )
+  endif(ENABLE_ALL_WARNINGS)
+
+  target_compile_options(${target}
+    PRIVATE
+    $<$<AND:$<CXX_COMPILER_ID:MSVC>,$<COMPILE_LANGUAGE:CXX>>:/MP>
+    $<$<AND:$<NOT:$<CXX_COMPILER_ID:MSVC>>,$<COMPILE_LANGUAGE:CXX>>:-funroll-loops>)
+
+  if (MSVC)
+    target_compile_options(${target} PRIVATE
+      $<$<NOT:$<COMPILE_LANGUAGE:CUDA>>:/utf-8>
+      -D_CRT_SECURE_NO_WARNINGS
+      -D_CRT_SECURE_NO_DEPRECATE
+      )
+  endif (MSVC)
+
+  if (WIN32 AND MINGW)
+    target_compile_options(${target} PUBLIC -static-libstdc++)
+  endif (WIN32 AND MINGW)
+endmacro(xgboost_target_properties)
+
+# Custom definitions used in xgboost.
+macro(xgboost_target_defs target)
+  if (NOT ${target} STREQUAL "dmlc") # skip dmlc core for custom logging.
+    target_compile_definitions(${target}
+      PRIVATE
+      -DDMLC_LOG_CUSTOMIZE=1
+      $<$<NOT:$<CXX_COMPILER_ID:MSVC>>:_MWAITXINTRIN_H_INCLUDED>)
+  endif ()
+  if (USE_DEBUG_OUTPUT)
+    target_compile_definitions(${target} PRIVATE -DXGBOOST_USE_DEBUG_OUTPUT=1)
+  endif (USE_DEBUG_OUTPUT)
+  if (XGBOOST_MM_PREFETCH_PRESENT)
+    target_compile_definitions(${target}
+      PRIVATE
+      -DXGBOOST_MM_PREFETCH_PRESENT=1)
+  endif(XGBOOST_MM_PREFETCH_PRESENT)
+  if (XGBOOST_BUILTIN_PREFETCH_PRESENT)
+    target_compile_definitions(${target}
+      PRIVATE
+      -DXGBOOST_BUILTIN_PREFETCH_PRESENT=1)
+  endif (XGBOOST_BUILTIN_PREFETCH_PRESENT)
+endmacro(xgboost_target_defs)
+
+# handles dependencies
+macro(xgboost_target_link_libraries target)
+  if (BUILD_STATIC_LIB)
+    target_link_libraries(${target} PUBLIC Threads::Threads ${CMAKE_THREAD_LIBS_INIT})
+  else()
+    target_link_libraries(${target} PRIVATE Threads::Threads ${CMAKE_THREAD_LIBS_INIT})
+  endif (BUILD_STATIC_LIB)
+
+  if (USE_OPENMP)
+    if (BUILD_STATIC_LIB)
+      target_link_libraries(${target} PUBLIC OpenMP::OpenMP_CXX)
+    else()
+      target_link_libraries(${target} PRIVATE OpenMP::OpenMP_CXX)
+    endif (BUILD_STATIC_LIB)
+  endif (USE_OPENMP)
+
+  if (USE_CUDA)
+    xgboost_set_cuda_flags(${target})
+  endif (USE_CUDA)
+
+  if (USE_NCCL)
+    xgboost_link_nccl(${target})
   endif (USE_NCCL)
-endfunction(xgboost_set_cuda_flags)
+
+  if (USE_NVTX)
+    enable_nvtx(${target})
+  endif (USE_NVTX)
+
+  if (RABIT_BUILD_MPI)
+    target_link_libraries(${target} PRIVATE MPI::MPI_CXX)
+  endif (RABIT_BUILD_MPI)
+endmacro(xgboost_target_link_libraries)

cmake/modules/FindNccl.cmake

@@ -29,7 +29,7 @@
 #  NCCL_INCLUDE_DIR, directory containing header
 #  NCCL_LIBRARY, directory containing nccl library
 #  NCCL_LIB_NAME, nccl library name
-#  USE_NCCL_LIB_PATH, when set, NCCL_LIBRARY path is also inspected for the 
+#  USE_NCCL_LIB_PATH, when set, NCCL_LIBRARY path is also inspected for the
 #                     location of the nccl library. This would disable
 #                     switching between static and shared.
 #

cmake/xgboost-config.cmake.in

@@ -1,21 +1,22 @@
 @PACKAGE_INIT@
 
-include(CMakeFindDependencyMacro)
-
 set(USE_OPENMP @USE_OPENMP@)
 set(USE_CUDA @USE_CUDA@)
 set(USE_NCCL @USE_NCCL@)
+set(XGBOOST_BUILD_STATIC_LIB @BUILD_STATIC_LIB@)
 
-find_dependency(Threads)
-if(USE_OPENMP)
-  find_dependency(OpenMP)
-endif()
-if(USE_CUDA)
-  find_dependency(CUDA)
-endif()
-if(USE_NCCL)
-  find_dependency(Nccl)
-endif()
+include(CMakeFindDependencyMacro)
+
+if (XGBOOST_BUILD_STATIC_LIB)
+  find_dependency(Threads)
+  if(USE_OPENMP)
+    find_dependency(OpenMP)
+  endif()
+  if(USE_CUDA)
+    find_dependency(CUDA)
+  endif()
+  # nccl should be linked statically if xgboost is built as static library.
+endif (XGBOOST_BUILD_STATIC_LIB)
 
 if(NOT TARGET xgboost::xgboost)
   include(${CMAKE_CURRENT_LIST_DIR}/XGBoostTargets.cmake)

demo/CLI/binary_classification/README.md

@@ -6,7 +6,7 @@ The script 'runexp.sh' can be used to run the demo. Here we use [mushroom datase
 
 ### Tutorial
 #### Generate Input Data
-XGBoost takes LibSVM format. An example of faked input data is below:
+XGBoost takes LIBSVM format. An example of faked input data is below:
 ```
 1 101:1.2 102:0.03
 0 1:2.1 10001:300 10002:400
@@ -15,7 +15,7 @@ XGBoost takes LibSVM format. An example of faked input data is below:
 Each line represent a single instance, and in the first line '1' is the instance label,'101' and '102' are feature indices, '1.2' and '0.03' are feature values. In the binary classification case, '1' is used to indicate positive samples, and '0' is used to indicate negative samples. We also support probability values in [0,1] as label, to indicate the probability of the instance being positive.
 
 
-First we will transform the dataset into classic LibSVM format and split the data into training set and test set by running:
+First we will transform the dataset into classic LIBSVM format and split the data into training set and test set by running:
 ```
 python mapfeat.py
 python mknfold.py agaricus.txt 1

demo/README.md

@@ -120,7 +120,7 @@ Please send pull requests if you find ones that are missing here.
 - [XGBoost - eXtreme Gradient Boosting](http://www.slideshare.net/ShangxuanZhang/xgboost) by Tong He
 - [How to use XGBoost algorithm in R in easy steps](http://www.analyticsvidhya.com/blog/2016/01/xgboost-algorithm-easy-steps/) by TAVISH SRIVASTAVA ([Chinese Translation 中文翻译](https://segmentfault.com/a/1190000004421821) by [HarryZhu](https://segmentfault.com/u/harryprince))
 - [Kaggle Solution: What’s Cooking ? (Text Mining Competition)](http://www.analyticsvidhya.com/blog/2015/12/kaggle-solution-cooking-text-mining-competition/) by MANISH SARASWAT
-- Better Optimization with Repeated Cross Validation and the XGBoost model - Machine Learning with R) by Manuel Amunategui ([Youtube Link](https://www.youtube.com/watch?v=Og7CGAfSr_Y)) ([Github Link](https://github.com/amunategui/BetterCrossValidation))
+- Better Optimization with Repeated Cross Validation and the XGBoost model - Machine Learning with R) by Manuel Amunategui ([Youtube Link](https://www.youtube.com/watch?v=Og7CGAfSr_Y)) ([GitHub Link](https://github.com/amunategui/BetterCrossValidation))
 - [XGBoost Rossman Parameter Tuning](https://www.kaggle.com/khozzy/rossmann-store-sales/xgboost-parameter-tuning-template/run/90168/notebook) by [Norbert Kozlowski](https://www.kaggle.com/khozzy)
 - [Featurizing log data before XGBoost](http://www.slideshare.net/DataRobot/featurizing-log-data-before-xgboost) by Xavier Conort, Owen Zhang etc
 - [West Nile Virus Competition Benchmarks & Tutorials](http://blog.kaggle.com/2015/07/21/west-nile-virus-competition-benchmarks-tutorials/) by [Anna Montoya](http://blog.kaggle.com/author/annamontoya/)

demo/c-api/CMakeLists.txt

@@ -1,5 +1,23 @@
 cmake_minimum_required(VERSION 3.13)
-project(api-demo LANGUAGES C CXX VERSION 0.0.1)
-find_package(xgboost REQUIRED)
-add_executable(api-demo c-api-demo.c)
-target_link_libraries(api-demo PRIVATE xgboost::xgboost)
+project(xgboost-c-examples)
+
+add_subdirectory(basic)
+add_subdirectory(external-memory)
+add_subdirectory(inference)
+
+enable_testing()
+add_test(
+  NAME test_xgboost_demo_c_basic
+  COMMAND api-demo
+  WORKING_DIRECTORY ${xgboost-c-examples_BINARY_DIR}
+)
+add_test(
+  NAME test_xgboost_demo_c_external_memory
+  COMMAND external-memory-demo
+  WORKING_DIRECTORY ${xgboost-c-examples_BINARY_DIR}
+)
+add_test(
+  NAME test_xgboost_demo_c_inference
+  COMMAND inference-demo
+  WORKING_DIRECTORY ${xgboost-c-examples_BINARY_DIR}
+)

demo/c-api/basic/CMakeLists.txt

@@ -0,0 +1,13 @@
+project(api-demo LANGUAGES C VERSION 0.0.1)
+find_package(xgboost REQUIRED)
+
+# xgboost is built as static libraries, all cxx dependencies need to be linked into the
+# executable.
+if (XGBOOST_BUILD_STATIC_LIB)
+  enable_language(CXX)
+  # find again for those  cxx libraries.
+  find_package(xgboost REQUIRED)
+endif(XGBOOST_BUILD_STATIC_LIB)
+
+add_executable(api-demo c-api-demo.c)
+target_link_libraries(api-demo PRIVATE xgboost::xgboost)

demo/c-api/basic/README.md

@@ -27,4 +27,4 @@ target_link_libraries(api-demo xgboost)
 ```
 
 # make
-You can start by modifying the makefile in this directory to fit your need.
+You can start by modifying the makefile in this directory to fit your need.

demo/c-api/basic/c-api-demo.c

@@ -24,8 +24,8 @@ int main(int argc, char** argv) {
 
   // load the data
   DMatrixHandle dtrain, dtest;
-  safe_xgboost(XGDMatrixCreateFromFile("../data/agaricus.txt.train", silent, &dtrain));
-  safe_xgboost(XGDMatrixCreateFromFile("../data/agaricus.txt.test", silent, &dtest));
+  safe_xgboost(XGDMatrixCreateFromFile("../../data/agaricus.txt.train", silent, &dtrain));
+  safe_xgboost(XGDMatrixCreateFromFile("../../data/agaricus.txt.test", silent, &dtest));
 
   // create the booster
   BoosterHandle booster;

demo/c-api/external-memory/CMakeLists.txt

@@ -0,0 +1,7 @@
+cmake_minimum_required(VERSION 3.13)
+project(external-memory-demo LANGUAGES C VERSION 0.0.1)
+
+find_package(xgboost REQUIRED)
+
+add_executable(external-memory-demo external_memory.c)
+target_link_libraries(external-memory-demo PRIVATE xgboost::xgboost)

demo/c-api/external-memory/README.md

@@ -0,0 +1,16 @@
+Defining a Custom Data Iterator to Load Data from External Memory
+=================================================================
+
+A simple demo for using custom data iterator with XGBoost.  The feature is still
+**experimental** and not ready for production use.  If you are not familiar with C API,
+please read its introduction in our tutorials and visit the basic demo first.
+
+Defining Data Iterator
+----------------------
+
+In the example, we define a custom data iterator with 2 methods: `reset` and `next`.  The
+`next` method passes data into XGBoost and tells XGBoost whether the iterator has reached
+its end, and the `reset` method resets iterations. One important detail when using the C
+API for data iterator is users need to make sure that the data passed into `next` method
+must be kept in memory until the next iteration or `reset` is called.  The external memory
+DMatrix is not limited to training, but also valid for other features like prediction.

demo/c-api/external-memory/external_memory.c

@@ -0,0 +1,180 @@
+/*!
+ * Copyright 2021 XGBoost contributors
+ *
+ * \brief A simple example of using xgboost data callback API.
+ */
+
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+#include <xgboost/c_api.h>
+
+#define safe_xgboost(err)                                                      \
+  if ((err) != 0) {                                                            \
+    fprintf(stderr, "%s:%d: error in %s: %s\n", __FILE__, __LINE__, #err,      \
+            XGBGetLastError());                                                \
+    exit(1);                                                                   \
+  }
+
+#define N_BATCHS 32
+#define BATCH_LEN 512
+
+/* Shorthands. */
+typedef DMatrixHandle DMatrix;
+typedef BoosterHandle Booster;
+
+typedef struct _DataIter {
+  /* Data of each batch. */
+  float **data;
+  /* Labels of each batch */
+  float **labels;
+  /* Length of each batch. */
+  size_t *lengths;
+  /* Total number of batches. */
+  size_t n;
+  /* Current iteration. */
+  size_t cur_it;
+
+  /* Private fields */
+  DMatrix _proxy;
+  char _array[128];
+} DataIter;
+
+#define safe_malloc(ptr)                                                       \
+  if ((ptr) == NULL) {                                                         \
+    fprintf(stderr, "%s:%d: Failed to allocate memory.\n", __FILE__,           \
+            __LINE__);                                                         \
+    exit(1);                                                                   \
+  }
+
+/**
+ * Initialize with random data for demo. In practice the data should be loaded
+ * from external memory.  We just demonstrate how to use the iterator in
+ * XGBoost.
+ *
+ * \param batch_size  Number of elements for each batch.  The demo here is only using 1
+ *                    column.
+ * \param n_batches   Number of batches.
+ */
+void DataIterator_Init(DataIter *self, size_t batch_size, size_t n_batches) {
+  self->n = n_batches;
+
+  self->lengths = (size_t *)malloc(self->n * sizeof(size_t));
+  safe_malloc(self->lengths);
+  for (size_t i = 0; i < self->n; ++i) {
+    self->lengths[i] = batch_size;
+  }
+
+  self->data = (float **)malloc(self->n * sizeof(float *));
+  safe_malloc(self->data);
+  self->labels = (float **)malloc(self->n * sizeof(float *));
+  safe_malloc(self->labels);
+
+  /* Generate some random data. */
+  for (size_t i = 0; i < self->n; ++i) {
+    self->data[i] = (float *)malloc(self->lengths[i] * sizeof(float));
+    safe_malloc(self->data[i]);
+    for (size_t j = 0; j < self->lengths[i]; ++j) {
+      float x = (float)rand() / (float)(RAND_MAX);
+      self->data[i][j] = x;
+    }
+
+    self->labels[i] = (float *)malloc(self->lengths[i] * sizeof(float));
+    safe_malloc(self->labels[i]);
+    for (size_t j = 0; j < self->lengths[i]; ++j) {
+      float y = (float)rand() / (float)(RAND_MAX);
+      self->labels[i][j] = y;
+    }
+  }
+
+  self->cur_it = 0;
+  safe_xgboost(XGProxyDMatrixCreate(&self->_proxy));
+}
+
+void DataIterator_Free(DataIter *self) {
+  for (size_t i = 0; i < self->n; ++i) {
+    free(self->data[i]);
+    free(self->labels[i]);
+  }
+  free(self->data);
+  free(self->lengths);
+  free(self->labels);
+  safe_xgboost(XGDMatrixFree(self->_proxy));
+};
+
+int DataIterator_Next(DataIterHandle handle) {
+  DataIter *self = (DataIter *)(handle);
+  if (self->cur_it == self->n) {
+    self->cur_it = 0;
+    return 0;  /* At end */
+  }
+
+  /* A JSON string encoding array interface (standard from numpy). */
+  char array[] = "{\"data\": [%lu, false], \"shape\":[%lu, 1], \"typestr\": "
+                 "\"<f4\", \"version\": 3}";
+  memset(self->_array, '\0', sizeof(self->_array));
+  sprintf(self->_array, array, (size_t)self->data[self->cur_it],
+          self->lengths[self->cur_it]);
+
+  safe_xgboost(XGProxyDMatrixSetDataDense(self->_proxy, self->_array));
+  /* The data passed in the iterator must remain valid (not being freed until the next
+   * iteration or reset) */
+  safe_xgboost(XGDMatrixSetDenseInfo(self->_proxy, "label",
+                                     self->labels[self->cur_it],
+                                     self->lengths[self->cur_it], 1));
+  self->cur_it++;
+  return 1;  /* Continue. */
+}
+
+void DataIterator_Reset(DataIterHandle handle) {
+  DataIter *self = (DataIter *)(handle);
+  self->cur_it = 0;
+}
+
+/**
+ * Train a regression model and save it into JSON model file.
+ */
+void TrainModel(DMatrix Xy) {
+  /* Create booster for training. */
+  Booster booster;
+  DMatrix cache[] = {Xy};
+  safe_xgboost(XGBoosterCreate(cache, 1, &booster));
+  /* Use approx for external memory training. */
+  safe_xgboost(XGBoosterSetParam(booster, "tree_method", "approx"));
+  safe_xgboost(XGBoosterSetParam(booster, "objective", "reg:squarederror"));
+
+  /* Start training. */
+  const char *validation_names[1] = {"train"};
+  const char *validation_result = NULL;
+  size_t n_rounds = 10;
+  for (size_t i = 0; i < n_rounds; ++i) {
+    safe_xgboost(XGBoosterUpdateOneIter(booster, i, Xy));
+    safe_xgboost(XGBoosterEvalOneIter(booster, i, cache, validation_names, 1,
+                                      &validation_result));
+    printf("%s\n", validation_result);
+  }
+
+  /* Save the model to a JSON file. */
+  safe_xgboost(XGBoosterSaveModel(booster, "model.json"));
+
+  safe_xgboost(XGBoosterFree(booster));
+}
+
+int main() {
+  DataIter iter;
+  DataIterator_Init(&iter, BATCH_LEN, N_BATCHS);
+
+  /* Create DMatrix from iterator.  During training, some cache files with the
+   * prefix "cache-" will be generated in current directory */
+  char config[] = "{\"missing\": NaN, \"cache_prefix\": \"cache\"}";
+  DMatrix Xy;
+  safe_xgboost(XGDMatrixCreateFromCallback(
+      &iter, iter._proxy, DataIterator_Reset, DataIterator_Next, config, &Xy));
+
+  TrainModel(Xy);
+
+  safe_xgboost(XGDMatrixFree(Xy));
+
+  DataIterator_Free(&iter);
+  return 0;
+}

demo/c-api/inference/CMakeLists.txt

@@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.13)
+project(inference-demo LANGUAGES C VERSION 0.0.1)
+find_package(xgboost REQUIRED)
+
+# xgboost is built as static libraries, all cxx dependencies need to be linked into the
+# executable.
+if (XGBOOST_BUILD_STATIC_LIB)
+  enable_language(CXX)
+  # find again for those  cxx libraries.
+  find_package(xgboost REQUIRED)
+endif(XGBOOST_BUILD_STATIC_LIB)
+
+add_executable(inference-demo inference.c)
+target_link_libraries(inference-demo PRIVATE xgboost::xgboost)

demo/c-api/inference/inference.c

@@ -0,0 +1,210 @@
+/*!
+ * Copyright 2021 XGBoost contributors
+ *
+ * \brief A simple example of using prediction functions.
+ */
+#include <stddef.h>
+#include <stdlib.h>
+#include <string.h>
+#include <xgboost/c_api.h>
+
+#define safe_xgboost(err)                                                      \
+  if ((err) != 0) {                                                            \
+    fprintf(stderr, "%s:%d: error in %s: %s\n", __FILE__, __LINE__, #err,      \
+            XGBGetLastError());                                                \
+    exit(1);                                                                   \
+  }
+
+#define safe_malloc(ptr)                                                       \
+  if ((ptr) == NULL) {                                                         \
+    fprintf(stderr, "%s:%d: Failed to allocate memory.\n", __FILE__,           \
+            __LINE__);                                                         \
+    exit(1);                                                                   \
+  }
+
+#define N_SAMPLES 128
+#define N_FEATURES 16
+
+typedef BoosterHandle Booster;
+typedef DMatrixHandle DMatrix;
+
+/* Row-major matrix */
+struct _Matrix {
+  float *data;
+  size_t shape[2];
+
+  /* private members */
+  char _array_intrerface[256];
+};
+
+/* A custom data type for demo. */
+typedef struct _Matrix *Matrix;
+
+/* Initialize matrix, copy data from `data` if it's not NULL. */
+void Matrix_Create(Matrix *self, float const *data, size_t n_samples,
+                   size_t n_features) {
+  if (self == NULL) {
+    fprintf(stderr, "Invalid pointer to %s\n", __func__);
+    exit(-1);
+  }
+
+  *self = (Matrix)malloc(sizeof(struct _Matrix));
+  safe_malloc(*self);
+  (*self)->data = (float *)malloc(n_samples * n_features * sizeof(float));
+  safe_malloc((*self)->data);
+  (*self)->shape[0] = n_samples;
+  (*self)->shape[1] = n_features;
+
+  if (data != NULL) {
+    memcpy((*self)->data, data,
+           (*self)->shape[0] * (*self)->shape[1] * sizeof(float));
+  }
+}
+
+/* Generate random matrix. */
+void Matrix_Random(Matrix *self, size_t n_samples, size_t n_features) {
+  Matrix_Create(self, NULL, n_samples, n_features);
+  for (size_t i = 0; i < n_samples * n_features; ++i) {
+    float x = (float)rand() / (float)(RAND_MAX);
+    (*self)->data[i] = x;
+  }
+}
+
+/* Array interface specified by numpy. */
+char const *Matrix_ArrayInterface(Matrix self) {
+  char const template[] = "{\"data\": [%lu, true], \"shape\": [%lu, %lu], "
+                          "\"typestr\": \"<f4\", \"version\": 3}";
+  memset(self->_array_intrerface, '\0', sizeof(self->_array_intrerface));
+  sprintf(self->_array_intrerface, template, (size_t)self->data, self->shape[0],
+          self->shape[1]);
+  return self->_array_intrerface;
+}
+
+size_t Matrix_NSamples(Matrix self) { return self->shape[0]; }
+
+size_t Matrix_NFeatures(Matrix self) { return self->shape[1]; }
+
+float Matrix_At(Matrix self, size_t i, size_t j) {
+  return self->data[i * self->shape[1] + j];
+}
+
+void Matrix_Print(Matrix self) {
+  for (size_t i = 0; i < Matrix_NSamples(self); i++) {
+    for (size_t j = 0; j < Matrix_NFeatures(self); ++j) {
+      printf("%f, ", Matrix_At(self, i, j));
+    }
+  }
+  printf("\n");
+}
+
+void Matrix_Free(Matrix self) {
+  if (self != NULL) {
+    if (self->data != NULL) {
+      self->shape[0] = 0;
+      self->shape[1] = 0;
+      free(self->data);
+      self->data = NULL;
+    }
+    free(self);
+  }
+}
+
+int main() {
+  Matrix X;
+  Matrix y;
+
+  Matrix_Random(&X, N_SAMPLES, N_FEATURES);
+  Matrix_Random(&y, N_SAMPLES, 1);
+
+  char const *X_interface = Matrix_ArrayInterface(X);
+  char config[] = "{\"nthread\": 16, \"missing\": NaN}";
+  DMatrix Xy;
+  /* Dense means "dense matrix". */
+  safe_xgboost(XGDMatrixCreateFromDense(X_interface, config, &Xy));
+  /* Label must be in a contigious array. */
+  safe_xgboost(XGDMatrixSetDenseInfo(Xy, "label", y->data, y->shape[0], 1));
+
+  DMatrix cache[] = {Xy};
+  Booster booster;
+  /* Train a booster for demo. */
+  safe_xgboost(XGBoosterCreate(cache, 1, &booster));
+
+  size_t n_rounds = 10;
+  for (size_t i = 0; i < n_rounds; ++i) {
+    safe_xgboost(XGBoosterUpdateOneIter(booster, i, Xy));
+  }
+
+  /* Save the trained model in JSON format. */
+  safe_xgboost(XGBoosterSaveModel(booster, "model.json"));
+  safe_xgboost(XGBoosterFree(booster));
+
+  /* Load it back for inference.  The save and load is not required, only shown here for
+   * demonstration purpose. */
+  safe_xgboost(XGBoosterCreate(NULL, 0, &booster));
+  safe_xgboost(XGBoosterLoadModel(booster, "model.json"));
+  {
+    /* Run prediction with DMatrix object. */
+    char const config[] =
+        "{\"training\": false, \"type\": 0, "
+        "\"iteration_begin\": 0, \"iteration_end\": 0, \"strict_shape\": true}";
+    /* Shape of output prediction */
+    uint64_t const *out_shape;
+    /* Dimension of output prediction */
+    uint64_t out_dim;
+    /* Pointer to a thread local contigious array, assigned in prediction function. */
+    float const *out_results;
+
+    safe_xgboost(XGBoosterPredictFromDMatrix(booster, Xy, config, &out_shape,
+                                             &out_dim, &out_results));
+    if (out_dim != 2 || out_shape[0] != N_SAMPLES || out_shape[1] != 1) {
+      fprintf(stderr, "Regression model should output prediction as vector.");
+      exit(-1);
+    }
+
+    Matrix predt;
+    /* Always copy output from XGBoost before calling next API function. */
+    Matrix_Create(&predt, out_results, out_shape[0], out_shape[1]);
+    printf("Results from prediction\n");
+    Matrix_Print(predt);
+    Matrix_Free(predt);
+  }
+
+  {
+    /* Run inplace prediction, which is faster and more memory efficient, but supports
+     * only basic inference types. */
+    char const config[] = "{\"type\": 0, \"iteration_begin\": 0, "
+                          "\"iteration_end\": 0, \"strict_shape\": true, "
+                          "\"cache_id\": 0, \"missing\": NaN}";
+    /* Shape of output prediction */
+    uint64_t const *out_shape;
+    /* Dimension of output prediction */
+    uint64_t out_dim;
+    /* Pointer to a thread local contigious array, assigned in prediction function. */
+    float const *out_results;
+
+    char const *X_interface = Matrix_ArrayInterface(X);
+    safe_xgboost(XGBoosterPredictFromDense(booster, X_interface, config, NULL,
+                                           &out_shape, &out_dim, &out_results));
+
+    if (out_dim != 2 || out_shape[0] != N_SAMPLES || out_shape[1] != 1) {
+      fprintf(stderr,
+              "Regression model should output prediction as vector, %lu, %lu",
+              out_dim, out_shape[0]);
+      exit(-1);
+    }
+
+    Matrix predt;
+    /* Always copy output from XGBoost before calling next API function. */
+    Matrix_Create(&predt, out_results, out_shape[0], out_shape[1]);
+    printf("Results from inplace prediction\n");
+    Matrix_Print(predt);
+    Matrix_Free(predt);
+  }
+
+  XGBoosterFree(booster);
+
+  XGDMatrixFree(Xy);
+  Matrix_Free(X);
+  Matrix_Free(y);
+  return 0;
+}

demo/dask/callbacks.py

@@ -0,0 +1,86 @@
+"""Example of using callbacks in Dask"""
+import numpy as np
+import xgboost as xgb
+from xgboost.dask import DaskDMatrix
+from dask.distributed import Client
+from dask.distributed import LocalCluster
+from dask_ml.datasets import make_regression
+from dask_ml.model_selection import train_test_split
+
+
+def probability_for_going_backward(epoch):
+    return 0.999 / (1.0 + 0.05 * np.log(1.0 + epoch))
+
+
+# All callback functions must inherit from TrainingCallback
+class CustomEarlyStopping(xgb.callback.TrainingCallback):
+    """A custom early stopping class where early stopping is determined stochastically.
+    In the beginning, allow the metric to become worse with a probability of 0.999.
+    As boosting progresses, the probability should be adjusted downward"""
+
+    def __init__(self, *, validation_set, target_metric, maximize, seed):
+        self.validation_set = validation_set
+        self.target_metric = target_metric
+        self.maximize = maximize
+        self.seed = seed
+        self.rng = np.random.default_rng(seed=seed)
+        if maximize:
+            self.better = lambda x, y: x > y
+        else:
+            self.better = lambda x, y: x < y
+
+    def after_iteration(self, model, epoch, evals_log):
+        metric_history = evals_log[self.validation_set][self.target_metric]
+        if len(metric_history) < 2 or self.better(
+            metric_history[-1], metric_history[-2]
+        ):
+            return False  # continue training
+        p = probability_for_going_backward(epoch)
+        go_backward = self.rng.choice(2, size=(1,), replace=True, p=[1 - p, p]).astype(
+            np.bool
+        )[0]
+        print(
+            "The validation metric went into the wrong direction. "
+            + f"Stopping training with probability {1 - p}..."
+        )
+        if go_backward:
+            return False  # continue training
+        else:
+            return True  # stop training
+
+
+def main(client):
+    m = 100000
+    n = 100
+    X, y = make_regression(n_samples=m, n_features=n, chunks=200, random_state=0)
+    X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0)
+
+    dtrain = DaskDMatrix(client, X_train, y_train)
+    dtest = DaskDMatrix(client, X_test, y_test)
+
+    output = xgb.dask.train(
+        client,
+        {
+            "verbosity": 1,
+            "tree_method": "hist",
+            "objective": "reg:squarederror",
+            "eval_metric": "rmse",
+            "max_depth": 6,
+            "learning_rate": 1.0,
+        },
+        dtrain,
+        num_boost_round=1000,
+        evals=[(dtrain, "train"), (dtest, "test")],
+        callbacks=[
+            CustomEarlyStopping(
+                validation_set="test", target_metric="rmse", maximize=False, seed=0
+            )
+        ],
+    )
+
+
+if __name__ == "__main__":
+    # or use other clusters for scaling
+    with LocalCluster(n_workers=4, threads_per_worker=1) as cluster:
+        with Client(cluster) as client:
+            main(client)

demo/dask/cpu_survival.py

@@ -0,0 +1,61 @@
+import xgboost as xgb
+import os
+from xgboost.dask import DaskDMatrix
+import dask.dataframe as dd
+from dask.distributed import Client
+from dask.distributed import LocalCluster
+
+def main(client):
+    # Load an example survival data from CSV into a Dask data frame.
+    # The Veterans' Administration Lung Cancer Trial
+    # The Statistical Analysis of Failure Time Data by Kalbfleisch J. and Prentice R (1980)
+    CURRENT_DIR = os.path.dirname(__file__)
+    df = dd.read_csv(os.path.join(CURRENT_DIR, os.pardir, 'data', 'veterans_lung_cancer.csv'))
+
+    # DaskDMatrix acts like normal DMatrix, works as a proxy for local
+    # DMatrix scatter around workers.
+    # For AFT survival, you'd need to extract the lower and upper bounds for the label
+    # and pass them as arguments to DaskDMatrix.
+    y_lower_bound = df['Survival_label_lower_bound']
+    y_upper_bound = df['Survival_label_upper_bound']
+    X = df.drop(['Survival_label_lower_bound',
+                 'Survival_label_upper_bound'], axis=1)
+    dtrain = DaskDMatrix(client, X, label_lower_bound=y_lower_bound,
+                         label_upper_bound=y_upper_bound)
+
+    # Use train method from xgboost.dask instead of xgboost.  This
+    # distributed version of train returns a dictionary containing the
+    # resulting booster and evaluation history obtained from
+    # evaluation metrics.
+    params = {'verbosity': 1,
+              'objective': 'survival:aft',
+              'eval_metric': 'aft-nloglik',
+              'learning_rate': 0.05,
+              'aft_loss_distribution_scale': 1.20,
+              'aft_loss_distribution': 'normal',
+              'max_depth': 6,
+              'lambda': 0.01,
+              'alpha': 0.02}
+    output = xgb.dask.train(client,
+                            params,
+                            dtrain,
+                            num_boost_round=100,
+                            evals=[(dtrain, 'train')])
+    bst = output['booster']
+    history = output['history']
+
+    # you can pass output directly into `predict` too.
+    prediction = xgb.dask.predict(client, bst, dtrain)
+    print('Evaluation history: ', history)
+
+    # Uncomment the following line to save the model to the disk
+    # bst.save_model('survival_model.json')
+
+    return prediction
+
+
+if __name__ == '__main__':
+    # or use other clusters for scaling
+    with LocalCluster(n_workers=7, threads_per_worker=4) as cluster:
+        with Client(cluster) as client:
+            main(client)

demo/guide-python/README.md

@@ -14,3 +14,5 @@ XGBoost Python Feature Walkthrough
 * [Sklearn access evals result](sklearn_evals_result.py)
 * [Access evals result](evals_result.py)
 * [External Memory](external_memory.py)
+* [Training continuation](continuation.py)
+* [Feature weights for column sampling](feature_weights.py)

demo/guide-python/basic_walkthrough.py

@@ -11,8 +11,8 @@ DEMO_DIR = os.path.join(XGBOOST_ROOT_DIR, 'demo')
 
 # simple example
 # load file from text file, also binary buffer generated by xgboost
-dtrain = xgb.DMatrix(os.path.join(DEMO_DIR, 'data', 'agaricus.txt.train'))
-dtest = xgb.DMatrix(os.path.join(DEMO_DIR, 'data', 'agaricus.txt.test'))
+dtrain = xgb.DMatrix(os.path.join(DEMO_DIR, 'data', 'agaricus.txt.train?indexing_mode=1'))
+dtest = xgb.DMatrix(os.path.join(DEMO_DIR, 'data', 'agaricus.txt.test?indexing_mode=1'))
 
 # specify parameters via map, definition are same as c++ version
 param = {'max_depth': 2, 'eta': 1, 'objective': 'binary:logistic'}

demo/guide-python/categorical.py

@@ -0,0 +1,76 @@
+"""Experimental support for categorical data.  After 1.5 XGBoost `gpu_hist` tree method
+has experimental support for one-hot encoding based tree split.
+
+In before, users need to run an encoder themselves before passing the data into XGBoost,
+which creates a sparse matrix and potentially increase memory usage.  This demo showcases
+the experimental categorical data support, more advanced features are planned.
+
+    .. versionadded:: 1.5.0
+
+"""
+import pandas as pd
+import numpy as np
+import xgboost as xgb
+from typing import Tuple
+
+
+def make_categorical(
+    n_samples: int, n_features: int, n_categories: int, onehot: bool
+) -> Tuple[pd.DataFrame, pd.Series]:
+    """Make some random data for demo."""
+    rng = np.random.RandomState(1994)
+
+    pd_dict = {}
+    for i in range(n_features + 1):
+        c = rng.randint(low=0, high=n_categories, size=n_samples)
+        pd_dict[str(i)] = pd.Series(c, dtype=np.int64)
+
+    df = pd.DataFrame(pd_dict)
+    label = df.iloc[:, 0]
+    df = df.iloc[:, 1:]
+    for i in range(0, n_features):
+        label += df.iloc[:, i]
+    label += 1
+
+    df = df.astype("category")
+    categories = np.arange(0, n_categories)
+    for col in df.columns:
+        df[col] = df[col].cat.set_categories(categories)
+
+    if onehot:
+        return pd.get_dummies(df), label
+    return df, label
+
+
+def main() -> None:
+    # Use builtin categorical data support
+    # For scikit-learn interface, the input data must be pandas DataFrame or cudf
+    # DataFrame with categorical features
+    X, y = make_categorical(100, 10, 4, False)
+    # Specify `enable_categorical` to True.
+    reg = xgb.XGBRegressor(tree_method="gpu_hist", enable_categorical=True)
+    reg.fit(X, y, eval_set=[(X, y)])
+
+    # Pass in already encoded data
+    X_enc, y_enc = make_categorical(100, 10, 4, True)
+    reg_enc = xgb.XGBRegressor(tree_method="gpu_hist")
+    reg_enc.fit(X_enc, y_enc, eval_set=[(X_enc, y_enc)])
+
+    reg_results = np.array(reg.evals_result()["validation_0"]["rmse"])
+    reg_enc_results = np.array(reg_enc.evals_result()["validation_0"]["rmse"])
+
+    # Check that they have same results
+    np.testing.assert_allclose(reg_results, reg_enc_results)
+
+    # Convert to DMatrix for SHAP value
+    booster: xgb.Booster = reg.get_booster()
+    m = xgb.DMatrix(X, enable_categorical=True)  # specify categorical data support.
+    SHAP = booster.predict(m, pred_contribs=True)
+    margin = booster.predict(m, output_margin=True)
+    np.testing.assert_allclose(
+        np.sum(SHAP, axis=len(SHAP.shape) - 1), margin, rtol=1e-3
+    )
+
+
+if __name__ == "__main__":
+    main()

demo/guide-python/continuation.py

@@ -0,0 +1,109 @@
+"""
+Demo for training continuation.
+"""
+
+from sklearn.datasets import load_breast_cancer
+import xgboost
+import pickle
+import tempfile
+import os
+
+
+def training_continuation(tmpdir: str, use_pickle: bool) -> None:
+    """Basic training continuation."""
+    # Train 128 iterations in 1 session
+    X, y = load_breast_cancer(return_X_y=True)
+    clf = xgboost.XGBClassifier(n_estimators=128, use_label_encoder=False)
+    clf.fit(X, y, eval_set=[(X, y)], eval_metric="logloss")
+    print("Total boosted rounds:", clf.get_booster().num_boosted_rounds())
+
+    # Train 128 iterations in 2 sessions, with the first one runs for 32 iterations and
+    # the second one runs for 96 iterations
+    clf = xgboost.XGBClassifier(n_estimators=32, use_label_encoder=False)
+    clf.fit(X, y, eval_set=[(X, y)], eval_metric="logloss")
+    assert clf.get_booster().num_boosted_rounds() == 32
+
+    # load back the model, this could be a checkpoint
+    if use_pickle:
+        path = os.path.join(tmpdir, "model-first-32.pkl")
+        with open(path, "wb") as fd:
+            pickle.dump(clf, fd)
+        with open(path, "rb") as fd:
+            loaded = pickle.load(fd)
+    else:
+        path = os.path.join(tmpdir, "model-first-32.json")
+        clf.save_model(path)
+        loaded = xgboost.XGBClassifier()
+        loaded.load_model(path)
+
+    clf = xgboost.XGBClassifier(n_estimators=128 - 32)
+    clf.fit(X, y, eval_set=[(X, y)], eval_metric="logloss", xgb_model=loaded)
+
+    print("Total boosted rounds:", clf.get_booster().num_boosted_rounds())
+
+    assert clf.get_booster().num_boosted_rounds() == 128
+
+
+def training_continuation_early_stop(tmpdir: str, use_pickle: bool) -> None:
+    """Training continuation with early stopping."""
+    early_stopping_rounds = 5
+    early_stop = xgboost.callback.EarlyStopping(
+        rounds=early_stopping_rounds, save_best=True
+    )
+    n_estimators = 512
+
+    X, y = load_breast_cancer(return_X_y=True)
+    clf = xgboost.XGBClassifier(n_estimators=n_estimators, use_label_encoder=False)
+    clf.fit(X, y, eval_set=[(X, y)], eval_metric="logloss", callbacks=[early_stop])
+    print("Total boosted rounds:", clf.get_booster().num_boosted_rounds())
+    best = clf.best_iteration
+
+    # Train 512 iterations in 2 sessions, with the first one runs for 128 iterations and
+    # the second one runs until early stop.
+    clf = xgboost.XGBClassifier(n_estimators=128, use_label_encoder=False)
+    # Reinitialize the early stop callback
+    early_stop = xgboost.callback.EarlyStopping(
+        rounds=early_stopping_rounds, save_best=True
+    )
+    clf.fit(X, y, eval_set=[(X, y)], eval_metric="logloss", callbacks=[early_stop])
+    assert clf.get_booster().num_boosted_rounds() == 128
+
+    # load back the model, this could be a checkpoint
+    if use_pickle:
+        path = os.path.join(tmpdir, "model-first-128.pkl")
+        with open(path, "wb") as fd:
+            pickle.dump(clf, fd)
+        with open(path, "rb") as fd:
+            loaded = pickle.load(fd)
+    else:
+        path = os.path.join(tmpdir, "model-first-128.json")
+        clf.save_model(path)
+        loaded = xgboost.XGBClassifier(use_label_encoder=False)
+        loaded.load_model(path)
+
+    early_stop = xgboost.callback.EarlyStopping(
+        rounds=early_stopping_rounds, save_best=True
+    )
+    clf = xgboost.XGBClassifier(
+        n_estimators=n_estimators - 128, use_label_encoder=False
+    )
+    clf.fit(
+        X,
+        y,
+        eval_set=[(X, y)],
+        eval_metric="logloss",
+        callbacks=[early_stop],
+        xgb_model=loaded,
+    )
+
+    print("Total boosted rounds:", clf.get_booster().num_boosted_rounds())
+    assert clf.best_iteration == best
+
+
+if __name__ == "__main__":
+    with tempfile.TemporaryDirectory() as tmpdir:
+        training_continuation_early_stop(tmpdir, False)
+        training_continuation_early_stop(tmpdir, True)
+
+        training_continuation(tmpdir, True)
+        training_continuation(tmpdir, False)

demo/guide-python/external_memory.py

@@ -1,22 +1,92 @@
+"""Experimental support for external memory.  This is similar to the one in
+`quantile_data_iterator.py`, but for external memory instead of Quantile DMatrix.  The
+feature is not ready for production use yet.
+
+    .. versionadded:: 1.5.0
+
+"""
 import os
-import xgboost as xgb
+import xgboost
+from typing import Callable, List, Tuple
+import tempfile
+import numpy as np
 
-### simple example for using external memory version
 
-# this is the only difference, add a # followed by a cache prefix name
-# several cache file with the prefix will be generated
-# currently only support convert from libsvm file
-CURRENT_DIR = os.path.dirname(__file__)
-dtrain = xgb.DMatrix(os.path.join(CURRENT_DIR, '../data/agaricus.txt.train#dtrain.cache'))
-dtest = xgb.DMatrix(os.path.join(CURRENT_DIR, '../data/agaricus.txt.test#dtest.cache'))
+def make_batches(
+    n_samples_per_batch: int, n_features: int, n_batches: int
+) -> Tuple[List[np.ndarray], List[np.ndarray]]:
+    """Generate random batches."""
+    X = []
+    y = []
+    rng = np.random.RandomState(1994)
+    for i in range(n_batches):
+        _X = rng.randn(n_samples_per_batch, n_features)
+        _y = rng.randn(n_samples_per_batch)
+        X.append(_X)
+        y.append(_y)
+    return X, y
 
-# specify validations set to watch performance
-param = {'max_depth':2, 'eta':1, 'objective':'binary:logistic'}
 
-# performance notice: set nthread to be the number of your real cpu
-# some cpu offer two threads per core, for example, a 4 core cpu with 8 threads, in such case set nthread=4
-#param['nthread']=num_real_cpu
+class Iterator(xgboost.DataIter):
+    """A custom iterator for loading files in batches."""
+    def __init__(self, file_paths: List[Tuple[str, str]]):
+        self._file_paths = file_paths
+        self._it = 0
+        # XGBoost will generate some cache files under current directory with the prefix
+        # "cache"
+        super().__init__(cache_prefix=os.path.join(".", "cache"))
 
-watchlist = [(dtest, 'eval'), (dtrain, 'train')]
-num_round = 2
-bst = xgb.train(param, dtrain, num_round, watchlist)
+    def load_file(self) -> Tuple[np.ndarray, np.ndarray]:
+        X_path, y_path = self._file_paths[self._it]
+        X = np.loadtxt(X_path)
+        y = np.loadtxt(y_path)
+        assert X.shape[0] == y.shape[0]
+        return X, y
+
+    def next(self, input_data: Callable) -> int:
+        """Advance the iterator by 1 step and pass the data to XGBoost.  This function is
+        called by XGBoost during the construction of ``DMatrix``
+
+        """
+        if self._it == len(self._file_paths):
+            # return 0 to let XGBoost know this is the end of iteration
+            return 0
+
+        # input_data is a function passed in by XGBoost who has the similar signature to
+        # the ``DMatrix`` constructor.
+        X, y = self.load_file()
+        input_data(data=X, label=y)
+        self._it += 1
+        return 1
+
+    def reset(self) -> None:
+        """Reset the iterator to its beginning"""
+        self._it = 0
+
+
+def main(tmpdir: str) -> xgboost.Booster:
+    # generate some random data for demo
+    batches = make_batches(1024, 17, 31)
+    files = []
+    for i, (X, y) in enumerate(zip(*batches)):
+        X_path = os.path.join(tmpdir, "X-" + str(i) + ".txt")
+        np.savetxt(X_path, X)
+        y_path = os.path.join(tmpdir, "y-" + str(i) + ".txt")
+        np.savetxt(y_path, y)
+        files.append((X_path, y_path))
+
+    it = Iterator(files)
+    # For non-data arguments, specify it here once instead of passing them by the `next`
+    # method.
+    missing = np.NaN
+    Xy = xgboost.DMatrix(it, missing=missing, enable_categorical=False)
+
+    # Other tree methods including ``hist`` and ``gpu_hist`` also work, but has some
+    # caveats.  This is still an experimental feature.
+    booster = xgboost.train({"tree_method": "approx"}, Xy)
+    return booster
+
+
+if __name__ == "__main__":
+    with tempfile.TemporaryDirectory() as tmpdir:
+        main(tmpdir)

demo/guide-python/quantile_data_iterator.py

@@ -85,7 +85,7 @@ def main():
     rounds = 100
     it = IterForDMatrixDemo()
 
-    # Use iterator, must be `DeviceQuantileDMatrix`
+    # Use iterator, must be `DeviceQuantileDMatrix` for quantile DMatrix.
     m_with_it = xgboost.DeviceQuantileDMatrix(it)
 
     # Use regular DMatrix.

demo/guide-python/update_process.py

@@ -0,0 +1,90 @@
+"""Demo for using `process_type` with `prune` and `refresh`.  Modifying existing trees is
+not a well established use for XGBoost, so feel free to experiment.
+
+"""
+
+import xgboost as xgb
+from sklearn.datasets import load_boston
+import numpy as np
+
+
+def main():
+    n_rounds = 32
+
+    X, y = load_boston(return_X_y=True)
+
+    # Train a model first
+    X_train = X[: X.shape[0] // 2]
+    y_train = y[: y.shape[0] // 2]
+    Xy = xgb.DMatrix(X_train, y_train)
+    evals_result: xgb.callback.EvaluationMonitor.EvalsLog = {}
+    booster = xgb.train(
+        {"tree_method": "gpu_hist", "max_depth": 6},
+        Xy,
+        num_boost_round=n_rounds,
+        evals=[(Xy, "Train")],
+        evals_result=evals_result,
+    )
+    SHAP = booster.predict(Xy, pred_contribs=True)
+
+    # Refresh the leaf value and tree statistic
+    X_refresh = X[X.shape[0] // 2:]
+    y_refresh = y[y.shape[0] // 2:]
+    Xy_refresh = xgb.DMatrix(X_refresh, y_refresh)
+    # The model will adapt to other half of the data by changing leaf value (no change in
+    # split condition) with refresh_leaf set to True.
+    refresh_result: xgb.callback.EvaluationMonitor.EvalsLog = {}
+    refreshed = xgb.train(
+        {"process_type": "update", "updater": "refresh", "refresh_leaf": True},
+        Xy_refresh,
+        num_boost_round=n_rounds,
+        xgb_model=booster,
+        evals=[(Xy, "Original"), (Xy_refresh, "Train")],
+        evals_result=refresh_result,
+    )
+
+    # Refresh the model without changing the leaf value, but tree statistic including
+    # cover and weight are refreshed.
+    refresh_result: xgb.callback.EvaluationMonitor.EvalsLog = {}
+    refreshed = xgb.train(
+        {"process_type": "update", "updater": "refresh", "refresh_leaf": False},
+        Xy_refresh,
+        num_boost_round=n_rounds,
+        xgb_model=booster,
+        evals=[(Xy, "Original"), (Xy_refresh, "Train")],
+        evals_result=refresh_result,
+    )
+    # Without refreshing the leaf value, resulting trees should be the same with original
+    # model except for accumulated statistic.  The rtol is for floating point error in
+    # prediction.
+    np.testing.assert_allclose(
+        refresh_result["Original"]["rmse"], evals_result["Train"]["rmse"], rtol=1e-5
+    )
+    # But SHAP value is changed as cover in tree nodes are changed.
+    refreshed_SHAP = refreshed.predict(Xy, pred_contribs=True)
+    assert not np.allclose(SHAP, refreshed_SHAP, rtol=1e-3)
+
+    # Prune the trees with smaller max_depth
+    X_update = X_train
+    y_update = y_train
+    Xy_update = xgb.DMatrix(X_update, y_update)
+
+    prune_result: xgb.callback.EvaluationMonitor.EvalsLog = {}
+    pruned = xgb.train(
+        {"process_type": "update", "updater": "prune", "max_depth": 2},
+        Xy_update,
+        num_boost_round=n_rounds,
+        xgb_model=booster,
+        evals=[(Xy, "Original"), (Xy_update, "Train")],
+        evals_result=prune_result,
+    )
+    # Have a smaller model, but similar accuracy.
+    np.testing.assert_allclose(
+        np.array(prune_result["Original"]["rmse"]),
+        np.array(prune_result["Train"]["rmse"]),
+        atol=1e-5
+    )
+
+
+if __name__ == "__main__":
+    main()

demo/rmm_plugin/README.md

@@ -27,5 +27,21 @@ cmake .. -DUSE_CUDA=ON -DUSE_NCCL=ON -DPLUGIN_RMM=ON -DCMAKE_PREFIX_PATH=$CONDA_
 cmake .. -DUSE_CUDA=ON -DUSE_NCCL=ON -DPLUGIN_RMM=ON -DCMAKE_PREFIX_PATH=/path/to/rmm
 ```
 
+# Informing XGBoost about RMM pool
+
+When XGBoost is compiled with RMM, most of the large size allocation will go through RMM
+allocators, but some small allocations in performance critical areas are using a different
+caching allocator so that we can have better control over memory allocation behavior.
+Users can override this behavior and force the use of rmm for all allocations by setting
+the global configuration ``use_rmm``:
+
+``` python
+with xgb.config_context(use_rmm=True):
+    clf = xgb.XGBClassifier(tree_method="gpu_hist")
+```
+
+Depending on the choice of memory pool size or type of allocator, this may have negative
+performance impact.
+
 * [Using RMM with a single GPU](./rmm_singlegpu.py)
 * [Using RMM with a local Dask cluster consisting of multiple GPUs](./rmm_mgpu_with_dask.py)

demo/rmm_plugin/rmm_mgpu_with_dask.py

@@ -4,11 +4,14 @@ import dask
 from dask.distributed import Client
 from dask_cuda import LocalCUDACluster
 
+
 def main(client):
-    # Inform XGBoost that RMM is used for GPU memory allocation
-    xgb.set_config(use_rmm=True)
+    # Optionally force XGBoost to use RMM for all GPU memory allocation, see ./README.md
+    # xgb.set_config(use_rmm=True)
 
     X, y = make_classification(n_samples=10000, n_informative=5, n_classes=3)
+    # In pratice one should prefer loading the data with dask collections instead of using
+    # `from_array`.
     X = dask.array.from_array(X)
     y = dask.array.from_array(y)
     dtrain = xgb.dask.DaskDMatrix(client, X, label=y)
@@ -22,6 +25,7 @@ def main(client):
     for i, e in enumerate(history['train']['merror']):
         print(f'[{i}] train-merror: {e}')
 
+
 if __name__ == '__main__':
     # To use RMM pool allocator with a GPU Dask cluster, just add rmm_pool_size option to
     # LocalCUDACluster constructor.

demo/rmm_plugin/rmm_singlegpu.py

@@ -4,13 +4,18 @@ from sklearn.datasets import make_classification
 
 # Initialize RMM pool allocator
 rmm.reinitialize(pool_allocator=True)
-# Inform XGBoost that RMM is used for GPU memory allocation
-xgb.set_config(use_rmm=True)
+# Optionally force XGBoost to use RMM for all GPU memory allocation, see ./README.md
+# xgb.set_config(use_rmm=True)
 
 X, y = make_classification(n_samples=10000, n_informative=5, n_classes=3)
 dtrain = xgb.DMatrix(X, label=y)
 
-params = {'max_depth': 8, 'eta': 0.01, 'objective': 'multi:softprob', 'num_class': 3,
-          'tree_method': 'gpu_hist'}
+params = {
+    "max_depth": 8,
+    "eta": 0.01,
+    "objective": "multi:softprob",
+    "num_class": 3,
+    "tree_method": "gpu_hist",
+}
 # XGBoost will automatically use the RMM pool allocator
-bst = xgb.train(params, dtrain, num_boost_round=100, evals=[(dtrain, 'train')])
+bst = xgb.train(params, dtrain, num_boost_round=100, evals=[(dtrain, "train")])

dev/release-pypi.py

@@ -36,10 +36,10 @@ def retrieve(url, filename=None):
     return urlretrieve(url, filename, reporthook=show_progress)
 
 
-def lastest_hash() -> str:
+def latest_hash() -> str:
     "Get latest commit hash."
     ret = subprocess.run(["git", "rev-parse", "HEAD"], capture_output=True)
-    assert ret.returncode == 0, "Failed to get lastest commit hash."
+    assert ret.returncode == 0, "Failed to get latest commit hash."
     commit_hash = ret.stdout.decode("utf-8").strip()
     return commit_hash
 
@@ -49,7 +49,7 @@ def download_wheels(
     dir_URL: str,
     src_filename_prefix: str,
     target_filename_prefix: str,
-) -> None:
+) -> List:
     """Download all binary wheels. dir_URL is the URL for remote directory storing the release
     wheels
 
@@ -71,6 +71,7 @@ def download_wheels(
         stdout = ret.stdout.decode("utf-8")
         assert stderr.find("warning") == -1, "Unresolved warnings:\n" + stderr
         assert stdout.find("warning") == -1, "Unresolved warnings:\n" + stdout
+    return filenames
 
 
 def check_path():
@@ -84,7 +85,7 @@ def main(args: argparse.Namespace) -> None:
     rel = version.StrictVersion(args.release)
     platforms = [
         "win_amd64",
-        "manylinux2010_x86_64",
+        "manylinux2014_x86_64",
         "manylinux2014_aarch64",
         "macosx_10_14_x86_64.macosx_10_15_x86_64.macosx_11_0_x86_64",
     ]
@@ -95,7 +96,7 @@ def main(args: argparse.Namespace) -> None:
     git.checkout(branch)
     git.pull("origin", branch)
     git.submodule("update")
-    commit_hash = lastest_hash()
+    commit_hash = latest_hash()
 
     dir_URL = PREFIX + str(major) + "." + str(minor) + ".0" + "/"
     src_filename_prefix = "xgboost-" + args.release + "%2B" + commit_hash + "-py3-none-"

doc/R-package/discoverYourData.md

@@ -5,9 +5,9 @@ Understand your dataset with XGBoost
 Introduction
 ------------
 
-The purpose of this Vignette is to show you how to use **Xgboost** to discover and understand your own dataset better.
+The purpose of this Vignette is to show you how to use **XGBoost** to discover and understand your own dataset better.
 
-This Vignette is not about predicting anything (see [Xgboost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)). We will explain how to use **Xgboost** to highlight the *link* between the *features* of your data and the *outcome*.
+This Vignette is not about predicting anything (see [XGBoost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)). We will explain how to use **XGBoost** to highlight the *link* between the *features* of your data and the *outcome*.
 
 Package loading:
 
@@ -27,7 +27,7 @@ Preparation of the dataset
 ### Numeric VS categorical variables
 
 
-**Xgboost** manages only `numeric` vectors.
+**XGBoost** manages only `numeric` vectors.
 
 What to do when you have *categorical* data?
 
@@ -55,7 +55,7 @@ data(Arthritis)
 df <- data.table(Arthritis, keep.rownames = FALSE)
 ```
 
-> `data.table` is 100% compliant with **R** `data.frame` but its syntax is more consistent and its performance for large dataset is [best in class](http://stackoverflow.com/questions/21435339/data-table-vs-dplyr-can-one-do-something-well-the-other-cant-or-does-poorly) (`dplyr` from **R** and `Pandas` from **Python** [included](https://github.com/Rdatatable/data.table/wiki/Benchmarks-%3A-Grouping)). Some parts of **Xgboost** **R** package use `data.table`.
+> `data.table` is 100% compliant with **R** `data.frame` but its syntax is more consistent and its performance for large dataset is [best in class](http://stackoverflow.com/questions/21435339/data-table-vs-dplyr-can-one-do-something-well-the-other-cant-or-does-poorly) (`dplyr` from **R** and `Pandas` from **Python** [included](https://github.com/Rdatatable/data.table/wiki/Benchmarks-%3A-Grouping)). Some parts of **XGBoost** **R** package use `data.table`.
 
 The first thing we want to do is to have a look to the first lines of the `data.table`:
 
@@ -217,7 +217,7 @@ output_vector = df[,Improved] == "Marked"
 Build the model
 ---------------
 
-The code below is very usual. For more information, you can look at the documentation of `xgboost` function (or at the vignette [Xgboost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)).
+The code below is very usual. For more information, you can look at the documentation of `xgboost` function (or at the vignette [XGBoost presentation](https://github.com/dmlc/xgboost/blob/master/R-package/vignettes/xgboostPresentation.Rmd)).
 
 
 ```r
@@ -422,19 +422,19 @@ Linear models may not be that smart in this scenario.
 Special Note: What about Random Forests™?
 -----------------------------------------
 
-As you may know, [Random Forests™](http://en.wikipedia.org/wiki/Random_forest) algorithm is cousin with boosting and both are part of the [ensemble learning](http://en.wikipedia.org/wiki/Ensemble_learning) family.
+As you may know, [Random Forests](http://en.wikipedia.org/wiki/Random_forest) algorithm is cousin with boosting and both are part of the [ensemble learning](http://en.wikipedia.org/wiki/Ensemble_learning) family.
 
-Both train several decision trees for one dataset. The *main* difference is that in Random Forests™, trees are independent and in boosting, the tree `N+1` focus its learning on the loss (<=> what has not been well modeled by the tree `N`).
+Both train several decision trees for one dataset. The *main* difference is that in Random Forests, trees are independent and in boosting, the tree `N+1` focus its learning on the loss (<=> what has not been well modeled by the tree `N`).
 
 This difference have an impact on a corner case in feature importance analysis: the *correlated features*.
 
-Imagine two features perfectly correlated, feature `A` and feature `B`. For one specific tree, if the algorithm needs one of them, it will choose randomly (true in both boosting and Random Forests™).
+Imagine two features perfectly correlated, feature `A` and feature `B`. For one specific tree, if the algorithm needs one of them, it will choose randomly (true in both boosting and Random Forests).
 
-However, in Random Forests™ this random choice will be done for each tree, because each tree is independent from the others. Therefore, approximatively, depending of your parameters, 50% of the trees will choose feature `A` and the other 50% will choose feature `B`. So the *importance* of the information contained in `A` and `B` (which is the same, because they are perfectly correlated) is diluted in `A` and `B`. So you won't easily know this information is important to predict what you want to predict! It is even worse when you have 10 correlated features...
+However, in Random Forests this random choice will be done for each tree, because each tree is independent from the others. Therefore, approximatively, depending of your parameters, 50% of the trees will choose feature `A` and the other 50% will choose feature `B`. So the *importance* of the information contained in `A` and `B` (which is the same, because they are perfectly correlated) is diluted in `A` and `B`. So you won't easily know this information is important to predict what you want to predict! It is even worse when you have 10 correlated features...
 
 In boosting, when a specific link between feature and outcome have been learned by the algorithm, it will try to not refocus on it (in theory it is what happens, reality is not always that simple). Therefore, all the importance will be on feature `A` or on feature `B` (but not both). You will know that one feature have an important role in the link between the observations and the label. It is still up to you to search for the correlated features to the one detected as important if you need to know all of them.
 
-If you want to try Random Forests™ algorithm, you can tweak Xgboost parameters!
+If you want to try Random Forests algorithm, you can tweak XGBoost parameters!
 
 **Warning**: this is still an experimental parameter.
 
@@ -447,7 +447,7 @@ data(agaricus.test, package='xgboost')
 train <- agaricus.train
 test <- agaricus.test
 
-#Random Forest™ - 1000 trees
+#Random Forest - 1000 trees
 bst <- xgboost(data = train$data, label = train$label, max.depth = 4, num_parallel_tree = 1000, subsample = 0.5, colsample_bytree =0.5, nrounds = 1, objective = "binary:logistic")
 ```
 
@@ -468,4 +468,4 @@ bst <- xgboost(data = train$data, label = train$label, max.depth = 4, nrounds =
 
 > Note that the parameter `round` is set to `1`.
 
-> [**Random Forests™**](https://www.stat.berkeley.edu/~breiman/RandomForests/cc_papers.htm) is a trademark of Leo Breiman and Adele Cutler and is licensed exclusively to Salford Systems for the commercial release of the software.
+> [**Random Forests**](https://www.stat.berkeley.edu/~breiman/RandomForests/cc_papers.htm) is a trademark of Leo Breiman and Adele Cutler and is licensed exclusively to Salford Systems for the commercial release of the software.

doc/R-package/index.rst

@@ -12,7 +12,7 @@ You have found the XGBoost R Package!
 ***********
 Get Started
 ***********
-* Checkout the :doc:`Installation Guide </build>` contains instructions to install xgboost, and :doc:`Tutorials </tutorials/index>` for examples on how to use XGBoost for various tasks.
+* Checkout the :doc:`Installation Guide </install>` contains instructions to install xgboost, and :doc:`Tutorials </tutorials/index>` for examples on how to use XGBoost for various tasks.
 * Read the `API documentation <https://cran.r-project.org/web/packages/xgboost/xgboost.pdf>`_.
 * Please visit `Walk-through Examples <https://github.com/dmlc/xgboost/tree/master/R-package/demo>`_.
 
@@ -23,6 +23,6 @@ Tutorials
 .. toctree::
   :maxdepth: 2
   :titlesonly:
-  
+
   Introduction to XGBoost in R <xgboostPresentation>
   Understanding your dataset with XGBoost <discoverYourData>