dmlc/xgboost@bea26c9

Updating to dmlc/xgboost#2543 (02/08/2017)

CMakeLists.txt

@@ -38,27 +38,42 @@ include_directories (
 file(GLOB_RECURSE SOURCES 
     src/*.cc
     src/*.h
+    include/*.h
 )
 # Only add main function for executable target
 list(REMOVE_ITEM SOURCES ${PROJECT_SOURCE_DIR}/src/cli_main.cc)
 
+
+# rabit
 # TODO: Create rabit cmakelists.txt
 set(RABIT_SOURCES
     rabit/src/allreduce_base.cc
     rabit/src/allreduce_robust.cc
     rabit/src/engine.cc
     rabit/src/c_api.cc
 )
-
-file(GLOB_RECURSE CUDA_SOURCES
-    plugin/updater_gpu/src/*.cu
-    plugin/updater_gpu/src/*.cuh
+set(RABIT_EMPTY_SOURCES
+    rabit/src/engine_empty.cc
+    rabit/src/c_api.cc
 )
+if(MINGW)
+  # build a dummy rabit library
+  add_library(rabit STATIC ${RABIT_EMPTY_SOURCES})
+else()
+  add_library(rabit STATIC ${RABIT_SOURCES})
+endif()
 
+
+# dmlc-core
 add_subdirectory(dmlc-core)
 set(LINK_LIBRARIES dmlccore rabit)
 
+
 # GPU Plugin
+file(GLOB_RECURSE CUDA_SOURCES
+    plugin/updater_gpu/src/*.cu
+    plugin/updater_gpu/src/*.cuh
+)
 if(PLUGIN_UPDATER_GPU)
   find_package(CUDA 7.5 REQUIRED)
   cmake_minimum_required(VERSION 3.5)
@@ -82,8 +97,6 @@ if(PLUGIN_UPDATER_GPU)
   list(APPEND SOURCES plugin/updater_gpu/src/register_updater_gpu.cc) 
 endif()
 
-add_library(rabit STATIC ${RABIT_SOURCES})
-
 add_library(objxgboost OBJECT ${SOURCES})
 
 # Executable
@@ -98,6 +111,10 @@ target_link_libraries(runxgboost ${LINK_LIBRARIES})
 add_library(xgboost SHARED $<TARGET_OBJECTS:objxgboost>)
 target_link_libraries(xgboost ${LINK_LIBRARIES})
 set_output_directory(xgboost ${PROJECT_SOURCE_DIR}/lib)
+if(MINGW)
+  # remove the 'lib' prefix to conform to windows convention for shared library names
+  set_target_properties(xgboost PROPERTIES PREFIX "")
+endif()
 
 #Ensure these two targets do not build simultaneously, as they produce outputs with conflicting names
 add_dependencies(xgboost runxgboost)

Makefile

@@ -73,7 +73,7 @@ ifndef LINT_LANG
 endif
 
 ifeq ($(UNAME), Windows)
-	XGBOOST_DYLIB = lib/libxgboost.dll
+	XGBOOST_DYLIB = lib/xgboost.dll
 	JAVAINCFLAGS += -I${JAVA_HOME}/include/win32
 else
 ifeq ($(UNAME), Darwin)
@@ -176,7 +176,7 @@ lib/libxgboost.a: $(ALL_DEP)
 	@mkdir -p $(@D)
 	ar crv $@ $(filter %.o, $?)
 
-lib/libxgboost.dll lib/libxgboost.so lib/libxgboost.dylib: $(ALL_DEP)
+lib/xgboost.dll lib/libxgboost.so lib/libxgboost.dylib: $(ALL_DEP)
 	@mkdir -p $(@D)
 	$(CXX) $(CFLAGS) -shared -o $@ $(filter %.o %a,  $^) $(LDFLAGS)
 
@@ -213,7 +213,7 @@ cover: check
 endif
 
 clean:
-	$(RM) -rf build build_plugin lib bin *~ */*~ */*/*~ */*/*/*~ */*.o */*/*.o */*/*/*.o xgboost
+	$(RM) -rf build build_plugin lib bin *~ */*~ */*/*~ */*/*/*~ */*.o */*/*.o */*/*/*.o #xgboost
 	$(RM) -rf build_tests *.gcov tests/cpp/xgboost_test
 
 clean_all: clean

amalgamation/xgboost-all0.cc

@@ -32,6 +32,10 @@
 #include "../src/data/simple_dmatrix.cc"
 #include "../src/data/sparse_page_raw_format.cc"
 
+// prediction
+#include "../src/predictor/predictor.cc"
+#include "../src/predictor/cpu_predictor.cc"
+
 #if DMLC_ENABLE_STD_THREAD
 #include "../src/data/sparse_page_source.cc"
 #include "../src/data/sparse_page_dmatrix.cc"

appveyor.yml

@@ -1,30 +1,60 @@
 environment:
-        matrix:
-                - target: native
-                  solution_name: C:/projects/xgboost/build2013/xgboost.sln
-                - target: native
-                  solution_name: C:/projects/xgboost/build2015/xgboost.sln
-                - target: jvm
-platform:
-        - x64
+    matrix:
+        - target: msvc
+          ver: 2013
+          generator: "Visual Studio 12 2013 Win64"
+          configuration: Release
+        - 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"
+        - target: jvm
 
-configuration:
-        - Debug
-        - Release
+platform:
+    - x64
 
 install:
-        - SET PATH=;%PATH%
-        - git submodule update --init --recursive
-
-before_build:
-        - mkdir build2013
-        - mkdir build2015
-        - cd build2013
-        - cmake .. -G"Visual Studio 12 2013 Win64" -DCMAKE_CONFIGURATION_TYPES="Release;Debug;" 
-        - cd ../build2015
-        - cmake .. -G"Visual Studio 14 2015 Win64" -DCMAKE_CONFIGURATION_TYPES="Release;Debug;" 
+    - git submodule update --init --recursive
+    # MinGW
+    - set PATH=C:\msys64\mingw64\bin;C:\msys64\usr\bin;%PATH%
+    - gcc -v
+    - ls -l C:\
+    # Miniconda2
+    - set PATH=;C:\Miniconda-x64;C:\Miniconda-x64\Scripts;%PATH%
+    - 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 install -y numpy scipy pandas matplotlib nose scikit-learn graphviz python-graphviz
 
 build_script:
-        - cd %APPVEYOR_BUILD_FOLDER%
-        - if "%target%" == "native" msbuild %solution_name%
-        - if "%target%" == "jvm" cd jvm-packages && mvn test -pl :xgboost4j
+    - 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
+      )
+    - if /i "%DO_PYTHON%" == "on" (
+        cd %APPVEYOR_BUILD_FOLDER%\python-package &&
+        python setup.py install
+      )
+    - if /i "%target%" == "jvm" cd jvm-packages && mvn test -pl :xgboost4j
+
+test_script:
+    - cd %APPVEYOR_BUILD_FOLDER%
+    - if /i "%DO_PYTHON%" == "on" python -m nose tests/python

doc/parameter.md

@@ -107,23 +107,6 @@ Parameters for Tree Booster
   - This is only used if 'hist' is specified as `tree_method`.
   - Maximum number of discrete bins to bucket continuous features.
   - Increasing this number improves the optimality of splits at the cost of higher computation time.
-* use_columnar_access, [default=1]
-  - This is only used if 'hist' is specified as `tree_method`.
-  - If greater than zero, store a transposed copy of input matrix for fast columnar access. May increase memory usage and initial setup time.
-* sparse_threshold, [default=0.2]
-  - range: [0.0, 1.0]
-  - This is only used if 'hist' is specified as `tree_method`.
-  - Percentage threshold for treating a feature as sparse. For instance, 0.2 indicates that any feature with fewer than 20% nonzero rows will be considered sparse. May impact computation time slightly.
-* enable_feature_grouping, [default=0]
-  - This is only used if 'hist' is specified as `tree_method`.
-  - If greater than zero, group complementary features together so as to improve work balance for parallel histogram aggregation. May increase memory usage and initial setup time.
-* max_conflict_rate, [default=0]
-  - range: [0.0, 1.0]
-  - Only relevant when `enable_feature_grouping=1` is specified.
-  - Specifies criterion for "complementary" features. By default, only features with no common nonzero rows are considered complementary. Increase this number to encourage larger feature groups.
-* max_search_group, [default=100]
-  - Only relevant when `enable_feature_grouping=1` is specified.
-  - Increasing this number will result in better feature grouping, at the cost of greater initial setup time.
 
 Additional parameters for Dart Booster
 --------------------------------------

include/xgboost/base.h

@@ -43,13 +43,13 @@
 /*!
  * \brief Check if alignas(*) keyword is supported. (g++ 4.8 or higher)
  */
-#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ >= 8
+#if defined(__GNUC__) && ((__GNUC__ == 4 && __GNUC_MINOR__ >= 8) || __GNUC__ > 4)
 #define XGBOOST_ALIGNAS(X) alignas(X)
 #else
 #define XGBOOST_ALIGNAS(X)
 #endif
 
-#if defined(__GNUC__) && __GNUC__ == 4 && __GNUC_MINOR__ >= 8 && \
+#if defined(__GNUC__) && ((__GNUC__ == 4 && __GNUC_MINOR__ >= 8) || __GNUC__ > 4) && \
     !defined(__CUDACC__)
 #include <parallel/algorithm>
 #define XGBOOST_PARALLEL_SORT(X, Y, Z) __gnu_parallel::sort((X), (Y), (Z))

include/xgboost/gbm.h

@@ -77,7 +77,7 @@ class GradientBooster {
    * \param ntree_limit limit the number of trees used in prediction, when it equals 0, this means
    *    we do not limit number of trees, this parameter is only valid for gbtree, but not for gblinear
    */
-  virtual void Predict(DMatrix* dmat,
+  virtual void PredictBatch(DMatrix* dmat,
                        std::vector<bst_float>* out_preds,
                        unsigned ntree_limit = 0) = 0;
   /*!
@@ -92,7 +92,7 @@ class GradientBooster {
    * \param root_index the root index
    * \sa Predict
    */
-  virtual void Predict(const SparseBatch::Inst& inst,
+  virtual void PredictInstance(const SparseBatch::Inst& inst,
                        std::vector<bst_float>* out_preds,
                        unsigned ntree_limit = 0,
                        unsigned root_index = 0) = 0;

include/xgboost/learner.h

@@ -189,7 +189,7 @@ inline void Learner::Predict(const SparseBatch::Inst& inst,
                              bool output_margin,
                              std::vector<bst_float>* out_preds,
                              unsigned ntree_limit) const {
-  gbm_->Predict(inst, out_preds, ntree_limit);
+  gbm_->PredictInstance(inst, out_preds, ntree_limit);
   if (!output_margin) {
     obj_->PredTransform(out_preds);
   }