Use sys.base_prefix instead of sys.prefix (#9711)

* Use sys.base_prefix instead of sys.prefix

* Update libpath.py too

.github/workflows/jvm_tests.yml

@@ -51,14 +51,14 @@ jobs:
       id: extract_branch
       if: |
         (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
-        matrix.os == 'windows-latest'
+        (matrix.os == 'windows-latest' || matrix.os == 'macos-11')
 
     - 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/${{ steps.extract_branch.outputs.branch }}/ --acl public-read
+        python -m awscli s3 cp xgboost4j_${{ github.sha }}.dll s3://xgboost-nightly-builds/${{ steps.extract_branch.outputs.branch }}/libxgboost4j/ --acl public-read
       if: |
         (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
         matrix.os == 'windows-latest'
@@ -66,6 +66,19 @@ jobs:
         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: Publish artifact libxgboost4j.dylib to S3
+      run: |
+        cd lib/
+        mv -v libxgboost4j.dylib libxgboost4j_${{ github.sha }}.dylib
+        ls
+        python -m awscli s3 cp libxgboost4j_${{ github.sha }}.dylib s3://xgboost-nightly-builds/${{ steps.extract_branch.outputs.branch }}/libxgboost4j/ --acl public-read
+      if: |
+        (github.ref == 'refs/heads/master' || contains(github.ref, 'refs/heads/release_')) &&
+        matrix.os == 'macos-11'
+      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 (Core, Spark, Examples)
       run: |

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
-project(xgboost LANGUAGES CXX C VERSION 2.0.0)
+project(xgboost LANGUAGES CXX C VERSION 2.0.1)
 include(cmake/Utils.cmake)
 list(APPEND CMAKE_MODULE_PATH "${xgboost_SOURCE_DIR}/cmake/modules")
 cmake_policy(SET CMP0022 NEW)
@@ -233,6 +233,11 @@ endif (RABIT_BUILD_MPI)
 add_subdirectory(${xgboost_SOURCE_DIR}/src)
 target_link_libraries(objxgboost PUBLIC dmlc)
 
+# Link -lstdc++fs for GCC 8.x
+if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU" AND CMAKE_CXX_COMPILER_VERSION VERSION_LESS "9.0")
+  target_link_libraries(objxgboost PUBLIC stdc++fs)
+endif()
+
 # Exports some R specific definitions and objects
 if (R_LIB)
   add_subdirectory(${xgboost_SOURCE_DIR}/R-package)

R-package/DESCRIPTION

@@ -1,8 +1,8 @@
 Package: xgboost
 Type: Package
 Title: Extreme Gradient Boosting
-Version: 2.0.0.1
-Date: 2023-09-11
+Version: 2.0.1.1
+Date: 2023-10-12
 Authors@R: c(
   person("Tianqi", "Chen", role = c("aut"),
          email = "tianqi.tchen@gmail.com"),

R-package/R/callbacks.R

@@ -70,7 +70,7 @@ cb.print.evaluation <- function(period = 1, showsd = TRUE) {
         i == env$begin_iteration ||
         i == env$end_iteration) {
       stdev <- if (showsd) env$bst_evaluation_err else NULL
-      msg <- format.eval.string(i, env$bst_evaluation, stdev)
+      msg <- .format_eval_string(i, env$bst_evaluation, stdev)
       cat(msg, '\n')
     }
   }
@@ -380,7 +380,9 @@ cb.early.stop <- function(stopping_rounds, maximize = FALSE,
     if ((maximize && score > best_score) ||
         (!maximize && score < best_score)) {
 
-      best_msg <<- format.eval.string(i, env$bst_evaluation, env$bst_evaluation_err)
+      best_msg <<- .format_eval_string(
+        i, env$bst_evaluation, env$bst_evaluation_err
+      )
       best_score <<- score
       best_iteration <<- i
       best_ntreelimit <<- best_iteration * env$num_parallel_tree
@@ -754,7 +756,7 @@ xgb.gblinear.history <- function(model, class_index = NULL) {
 #
 
 # Format the evaluation metric string
-format.eval.string <- function(iter, eval_res, eval_err = NULL) {
+.format_eval_string <- function(iter, eval_res, eval_err = NULL) {
   if (length(eval_res) == 0)
     stop('no evaluation results')
   enames <- names(eval_res)

R-package/configure

@@ -1,6 +1,6 @@
 #! /bin/sh
 # Guess values for system-dependent variables and create Makefiles.
-# Generated by GNU Autoconf 2.71 for xgboost 2.0.0.
+# Generated by GNU Autoconf 2.71 for xgboost 2.0.1.
 #
 #
 # Copyright (C) 1992-1996, 1998-2017, 2020-2021 Free Software Foundation,
@@ -607,8 +607,8 @@ MAKEFLAGS=
 # Identity of this package.
 PACKAGE_NAME='xgboost'
 PACKAGE_TARNAME='xgboost'
-PACKAGE_VERSION='2.0.0'
-PACKAGE_STRING='xgboost 2.0.0'
+PACKAGE_VERSION='2.0.1'
+PACKAGE_STRING='xgboost 2.0.1'
 PACKAGE_BUGREPORT=''
 PACKAGE_URL=''
 
@@ -1225,7 +1225,7 @@ if test "$ac_init_help" = "long"; then
   # Omit some internal or obsolete options to make the list less imposing.
   # This message is too long to be a string in the A/UX 3.1 sh.
   cat <<_ACEOF
-\`configure' configures xgboost 2.0.0 to adapt to many kinds of systems.
+\`configure' configures xgboost 2.0.1 to adapt to many kinds of systems.
 
 Usage: $0 [OPTION]... [VAR=VALUE]...
 
@@ -1287,7 +1287,7 @@ fi
 
 if test -n "$ac_init_help"; then
   case $ac_init_help in
-     short | recursive ) echo "Configuration of xgboost 2.0.0:";;
+     short | recursive ) echo "Configuration of xgboost 2.0.1:";;
    esac
   cat <<\_ACEOF
 
@@ -1367,7 +1367,7 @@ fi
 test -n "$ac_init_help" && exit $ac_status
 if $ac_init_version; then
   cat <<\_ACEOF
-xgboost configure 2.0.0
+xgboost configure 2.0.1
 generated by GNU Autoconf 2.71
 
 Copyright (C) 2021 Free Software Foundation, Inc.
@@ -1533,7 +1533,7 @@ cat >config.log <<_ACEOF
 This file contains any messages produced by compilers while
 running configure, to aid debugging if configure makes a mistake.
 
-It was created by xgboost $as_me 2.0.0, which was
+It was created by xgboost $as_me 2.0.1, which was
 generated by GNU Autoconf 2.71.  Invocation command line was
 
   $ $0$ac_configure_args_raw
@@ -3412,7 +3412,7 @@ cat >>$CONFIG_STATUS <<\_ACEOF || ac_write_fail=1
 # report actual input values of CONFIG_FILES etc. instead of their
 # values after options handling.
 ac_log="
-This file was extended by xgboost $as_me 2.0.0, which was
+This file was extended by xgboost $as_me 2.0.1, which was
 generated by GNU Autoconf 2.71.  Invocation command line was
 
   CONFIG_FILES    = $CONFIG_FILES
@@ -3467,7 +3467,7 @@ ac_cs_config_escaped=`printf "%s\n" "$ac_cs_config" | sed "s/^ //; s/'/'\\\\\\\\
 cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
 ac_cs_config='$ac_cs_config_escaped'
 ac_cs_version="\\
-xgboost config.status 2.0.0
+xgboost config.status 2.0.1
 configured by $0, generated by GNU Autoconf 2.71,
   with options \\"\$ac_cs_config\\"
 

R-package/configure.ac

@@ -2,7 +2,7 @@
 
 AC_PREREQ(2.69)
 
-AC_INIT([xgboost],[2.0.0],[],[xgboost],[])
+AC_INIT([xgboost],[2.0.1],[],[xgboost],[])
 
 : ${R_HOME=`R RHOME`}
 if test -z "${R_HOME}"; then

include/xgboost/version_config.h

@@ -6,6 +6,6 @@
 
 #define XGBOOST_VER_MAJOR 2  /* NOLINT */
 #define XGBOOST_VER_MINOR 0  /* NOLINT */
-#define XGBOOST_VER_PATCH 0  /* NOLINT */
+#define XGBOOST_VER_PATCH 1  /* NOLINT */
 
 #endif  // XGBOOST_VERSION_CONFIG_H_

jvm-packages/CMakeLists.txt

@@ -25,4 +25,3 @@ target_include_directories(xgboost4j
   ${PROJECT_SOURCE_DIR}/rabit/include)
 
 set_output_directory(xgboost4j ${PROJECT_SOURCE_DIR}/lib)
-target_link_libraries(xgboost4j PRIVATE ${JAVA_JVM_LIBRARY})

jvm-packages/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>ml.dmlc</groupId>
     <artifactId>xgboost-jvm</artifactId>
-    <version>2.0.0</version>
+    <version>2.0.1</version>
     <packaging>pom</packaging>
     <name>XGBoost JVM Package</name>
     <description>JVM Package for XGBoost</description>

jvm-packages/xgboost4j-example/pom.xml

@@ -6,11 +6,11 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.1</version>
     </parent>
     <name>xgboost4j-example</name>
     <artifactId>xgboost4j-example_${scala.binary.version}</artifactId>
-    <version>2.0.0</version>
+    <version>2.0.1</version>
     <packaging>jar</packaging>
     <build>
         <plugins>

jvm-packages/xgboost4j-flink/pom.xml

@@ -6,12 +6,12 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.1</version>
     </parent>
 
     <name>xgboost4j-flink</name>
     <artifactId>xgboost4j-flink_${scala.binary.version}</artifactId>
-    <version>2.0.0</version>
+    <version>2.0.1</version>
     <properties>
       <flink-ml.version>2.2.0</flink-ml.version>
     </properties>

jvm-packages/xgboost4j-gpu/pom.xml

@@ -6,11 +6,11 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.1</version>
     </parent>
     <artifactId>xgboost4j-gpu_${scala.binary.version}</artifactId>
     <name>xgboost4j-gpu</name>
-    <version>2.0.0</version>
+    <version>2.0.1</version>
     <packaging>jar</packaging>
 
     <dependencies>

jvm-packages/xgboost4j-spark-gpu/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.1</version>
     </parent>
     <name>xgboost4j-spark-gpu</name>
     <artifactId>xgboost4j-spark-gpu_${scala.binary.version}</artifactId>

jvm-packages/xgboost4j-spark/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.1</version>
     </parent>
     <name>xgboost4j-spark</name>
     <artifactId>xgboost4j-spark_${scala.binary.version}</artifactId>

jvm-packages/xgboost4j/pom.xml

@@ -6,11 +6,11 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm</artifactId>
-        <version>2.0.0</version>
+        <version>2.0.1</version>
     </parent>
     <name>xgboost4j</name>
     <artifactId>xgboost4j_${scala.binary.version}</artifactId>
-    <version>2.0.0</version>
+    <version>2.0.1</version>
     <packaging>jar</packaging>
 
     <dependencies>

python-package/packager/nativelib.py

@@ -132,16 +132,28 @@ def locate_or_build_libxgboost(
 
     if build_config.use_system_libxgboost:
         # Find libxgboost from system prefix
-        sys_base_prefix = pathlib.Path(sys.base_prefix).absolute().resolve()
-        libxgboost_sys = sys_base_prefix / "lib" / _lib_name()
-        if not libxgboost_sys.exists():
-            raise RuntimeError(
-                f"use_system_libxgboost was specified but {_lib_name()} is "
-                f"not found in {libxgboost_sys.parent}"
-            )
-
-        logger.info("Using system XGBoost: %s", str(libxgboost_sys))
-        return libxgboost_sys
+        sys_prefix = pathlib.Path(sys.base_prefix)
+        sys_prefix_candidates = [
+            sys_prefix / "lib",
+            # Paths possibly used on Windows
+            sys_prefix / "bin",
+            sys_prefix / "Library",
+            sys_prefix / "Library" / "bin",
+            sys_prefix / "Library" / "lib",
+        ]
+        sys_prefix_candidates = [
+            p.expanduser().resolve() for p in sys_prefix_candidates
+        ]
+        for candidate_dir in sys_prefix_candidates:
+            libtreelite_sys = candidate_dir / _lib_name()
+            if libtreelite_sys.exists():
+                logger.info("Using system XGBoost: %s", str(libtreelite_sys))
+                return libtreelite_sys
+        raise RuntimeError(
+            f"use_system_libxgboost was specified but {_lib_name()} is "
+            f"not found. Paths searched (in order): \n"
+            + "\n".join([f"* {str(p)}" for p in sys_prefix_candidates])
+        )
 
     libxgboost = locate_local_libxgboost(toplevel_dir, logger=logger)
     if libxgboost is not None:

python-package/pyproject.toml

@@ -7,7 +7,7 @@ build-backend = "packager.pep517"
 
 [project]
 name = "xgboost"
-version = "2.0.0"
+version = "2.0.1"
 authors = [
     { name = "Hyunsu Cho", email = "chohyu01@cs.washington.edu" },
     { name = "Jiaming Yuan", email = "jm.yuan@outlook.com" }

python-package/xgboost/VERSION

@@ -1 +1 @@
-2.0.0
+2.0.1

python-package/xgboost/core.py

@@ -2399,6 +2399,7 @@ class Booster:
             _is_cudf_df,
             _is_cupy_array,
             _is_list,
+            _is_np_array_like,
             _is_pandas_df,
             _is_pandas_series,
             _is_tuple,
@@ -2428,7 +2429,7 @@ class Booster:
                     f"got {data.shape[1]}"
                 )
 
-        if isinstance(data, np.ndarray):
+        if _is_np_array_like(data):
             from .data import _ensure_np_dtype
 
             data, _ = _ensure_np_dtype(data, data.dtype)

python-package/xgboost/data.py

@@ -164,8 +164,8 @@ def _is_scipy_coo(data: DataType) -> bool:
     return isinstance(data, scipy.sparse.coo_matrix)
 
 
-def _is_numpy_array(data: DataType) -> bool:
-    return isinstance(data, (np.ndarray, np.matrix))
+def _is_np_array_like(data: DataType) -> bool:
+    return hasattr(data, "__array_interface__")
 
 
 def _ensure_np_dtype(
@@ -317,7 +317,6 @@ def pandas_feature_info(
 ) -> Tuple[Optional[FeatureNames], Optional[FeatureTypes]]:
     """Handle feature info for pandas dataframe."""
     import pandas as pd
-    from pandas.api.types import is_categorical_dtype, is_sparse
 
     # handle feature names
     if feature_names is None and meta is None:
@@ -332,10 +331,10 @@ def pandas_feature_info(
     if feature_types is None and meta is None:
         feature_types = []
         for dtype in data.dtypes:
-            if is_sparse(dtype):
+            if is_pd_sparse_dtype(dtype):
                 feature_types.append(_pandas_dtype_mapper[dtype.subtype.name])
             elif (
-                is_categorical_dtype(dtype) or is_pa_ext_categorical_dtype(dtype)
+                is_pd_cat_dtype(dtype) or is_pa_ext_categorical_dtype(dtype)
             ) and enable_categorical:
                 feature_types.append(CAT_T)
             else:
@@ -345,18 +344,13 @@ def pandas_feature_info(
 
 def is_nullable_dtype(dtype: PandasDType) -> bool:
     """Whether dtype is a pandas nullable type."""
-    from pandas.api.types import (
-        is_bool_dtype,
-        is_categorical_dtype,
-        is_float_dtype,
-        is_integer_dtype,
-    )
+    from pandas.api.types import is_bool_dtype, is_float_dtype, is_integer_dtype
 
     is_int = is_integer_dtype(dtype) and dtype.name in pandas_nullable_mapper
     # np.bool has alias `bool`, while pd.BooleanDtype has `boolean`.
     is_bool = is_bool_dtype(dtype) and dtype.name == "boolean"
     is_float = is_float_dtype(dtype) and dtype.name in pandas_nullable_mapper
-    return is_int or is_bool or is_float or is_categorical_dtype(dtype)
+    return is_int or is_bool or is_float or is_pd_cat_dtype(dtype)
 
 
 def is_pa_ext_dtype(dtype: Any) -> bool:
@@ -371,17 +365,48 @@ def is_pa_ext_categorical_dtype(dtype: Any) -> bool:
     )
 
 
+def is_pd_cat_dtype(dtype: PandasDType) -> bool:
+    """Wrapper for testing pandas category type."""
+    import pandas as pd
+
+    if hasattr(pd.util, "version") and hasattr(pd.util.version, "Version"):
+        Version = pd.util.version.Version
+        if Version(pd.__version__) >= Version("2.1.0"):
+            from pandas import CategoricalDtype
+
+            return isinstance(dtype, CategoricalDtype)
+
+    from pandas.api.types import is_categorical_dtype
+
+    return is_categorical_dtype(dtype)
+
+
+def is_pd_sparse_dtype(dtype: PandasDType) -> bool:
+    """Wrapper for testing pandas sparse type."""
+    import pandas as pd
+
+    if hasattr(pd.util, "version") and hasattr(pd.util.version, "Version"):
+        Version = pd.util.version.Version
+        if Version(pd.__version__) >= Version("2.1.0"):
+            from pandas import SparseDtype
+
+            return isinstance(dtype, SparseDtype)
+
+    from pandas.api.types import is_sparse
+
+    return is_sparse(dtype)
+
+
 def pandas_cat_null(data: DataFrame) -> DataFrame:
     """Handle categorical dtype and nullable extension types from pandas."""
     import pandas as pd
-    from pandas.api.types import is_categorical_dtype
 
     # handle category codes and nullable.
     cat_columns = []
     nul_columns = []
     # avoid an unnecessary conversion if possible
     for col, dtype in zip(data.columns, data.dtypes):
-        if is_categorical_dtype(dtype):
+        if is_pd_cat_dtype(dtype):
             cat_columns.append(col)
         elif is_pa_ext_categorical_dtype(dtype):
             raise ValueError(
@@ -398,7 +423,7 @@ def pandas_cat_null(data: DataFrame) -> DataFrame:
         transformed = data
 
     def cat_codes(ser: pd.Series) -> pd.Series:
-        if is_categorical_dtype(ser.dtype):
+        if is_pd_cat_dtype(ser.dtype):
             return ser.cat.codes
         assert is_pa_ext_categorical_dtype(ser.dtype)
         # Not yet supported, the index is not ordered for some reason. Alternately:
@@ -454,14 +479,12 @@ def _transform_pandas_df(
     meta: Optional[str] = None,
     meta_type: Optional[NumpyDType] = None,
 ) -> Tuple[np.ndarray, Optional[FeatureNames], Optional[FeatureTypes]]:
-    from pandas.api.types import is_categorical_dtype, is_sparse
-
     pyarrow_extension = False
     for dtype in data.dtypes:
         if not (
             (dtype.name in _pandas_dtype_mapper)
-            or is_sparse(dtype)
-            or (is_categorical_dtype(dtype) and enable_categorical)
+            or is_pd_sparse_dtype(dtype)
+            or (is_pd_cat_dtype(dtype) and enable_categorical)
             or is_pa_ext_dtype(dtype)
         ):
             _invalid_dataframe_dtype(data)
@@ -515,9 +538,8 @@ def _meta_from_pandas_series(
 ) -> None:
     """Help transform pandas series for meta data like labels"""
     data = data.values.astype("float")
-    from pandas.api.types import is_sparse
 
-    if is_sparse(data):
+    if is_pd_sparse_dtype(getattr(data, "dtype", data)):
         data = data.to_dense()  # type: ignore
     assert len(data.shape) == 1 or data.shape[1] == 0 or data.shape[1] == 1
     _meta_from_numpy(data, name, dtype, handle)
@@ -539,13 +561,11 @@ def _from_pandas_series(
     feature_names: Optional[FeatureNames],
     feature_types: Optional[FeatureTypes],
 ) -> DispatchedDataBackendReturnType:
-    from pandas.api.types import is_categorical_dtype
-
     if (data.dtype.name not in _pandas_dtype_mapper) and not (
-        is_categorical_dtype(data.dtype) and enable_categorical
+        is_pd_cat_dtype(data.dtype) and enable_categorical
     ):
         _invalid_dataframe_dtype(data)
-    if enable_categorical and is_categorical_dtype(data.dtype):
+    if enable_categorical and is_pd_cat_dtype(data.dtype):
         data = data.cat.codes
     return _from_numpy_array(
         data.values.reshape(data.shape[0], 1).astype("float"),
@@ -1051,7 +1071,7 @@ def dispatch_data_backend(
         return _from_scipy_csr(
             data.tocsr(), missing, threads, feature_names, feature_types
         )
-    if _is_numpy_array(data):
+    if _is_np_array_like(data):
         return _from_numpy_array(
             data, missing, threads, feature_names, feature_types, data_split_mode
         )
@@ -1194,7 +1214,7 @@ def dispatch_meta_backend(
     if _is_tuple(data):
         _meta_from_tuple(data, name, dtype, handle)
         return
-    if _is_numpy_array(data):
+    if _is_np_array_like(data):
         _meta_from_numpy(data, name, dtype, handle)
         return
     if _is_pandas_df(data):
@@ -1281,7 +1301,7 @@ def _proxy_transform(
         return _transform_dlpack(data), None, feature_names, feature_types
     if _is_list(data) or _is_tuple(data):
         data = np.array(data)
-    if _is_numpy_array(data):
+    if _is_np_array_like(data):
         data, _ = _ensure_np_dtype(data, data.dtype)
         return data, None, feature_names, feature_types
     if _is_scipy_csr(data):
@@ -1331,7 +1351,7 @@ def dispatch_proxy_set_data(
     if not allow_host:
         raise err
 
-    if _is_numpy_array(data):
+    if _is_np_array_like(data):
         _check_data_shape(data)
         proxy._set_data_from_array(data)  # pylint: disable=W0212
         return

python-package/xgboost/libpath.py

@@ -31,16 +31,15 @@ def find_lib_path() -> List[str]:
     ]
 
     if sys.platform == "win32":
-        if platform.architecture()[0] == "64bit":
-            dll_path.append(os.path.join(curr_path, "../../windows/x64/Release/"))
-            # hack for pip installation when copy all parent source
-            # directory here
-            dll_path.append(os.path.join(curr_path, "./windows/x64/Release/"))
-        else:
-            dll_path.append(os.path.join(curr_path, "../../windows/Release/"))
-            # hack for pip installation when copy all parent source
-            # directory here
-            dll_path.append(os.path.join(curr_path, "./windows/Release/"))
+        # On Windows, Conda may install libs in different paths
+        dll_path.extend(
+            [
+                os.path.join(sys.base_prefix, "bin"),
+                os.path.join(sys.base_prefix, "Library"),
+                os.path.join(sys.base_prefix, "Library", "bin"),
+                os.path.join(sys.base_prefix, "Library", "lib"),
+            ]
+        )
         dll_path = [os.path.join(p, "xgboost.dll") for p in dll_path]
     elif sys.platform.startswith(("linux", "freebsd", "emscripten")):
         dll_path = [os.path.join(p, "libxgboost.so") for p in dll_path]

python-package/xgboost/spark/core.py

@@ -22,7 +22,7 @@ from typing import (
 
 import numpy as np
 import pandas as pd
-from pyspark import SparkContext, cloudpickle
+from pyspark import RDD, SparkContext, cloudpickle
 from pyspark.ml import Estimator, Model
 from pyspark.ml.functions import array_to_vector, vector_to_array
 from pyspark.ml.linalg import VectorUDT
@@ -44,6 +44,7 @@ from pyspark.ml.util import (
     MLWritable,
     MLWriter,
 )
+from pyspark.resource import ResourceProfileBuilder, TaskResourceRequests
 from pyspark.sql import Column, DataFrame
 from pyspark.sql.functions import col, countDistinct, pandas_udf, rand, struct
 from pyspark.sql.types import (
@@ -88,6 +89,7 @@ from .utils import (
     _get_rabit_args,
     _get_spark_session,
     _is_local,
+    _is_standalone_or_localcluster,
     deserialize_booster,
     deserialize_xgb_model,
     get_class_name,
@@ -342,6 +344,54 @@ class _SparkXGBParams(
                 predict_params[param.name] = self.getOrDefault(param)
         return predict_params
 
+    def _validate_gpu_params(self) -> None:
+        """Validate the gpu parameters and gpu configurations"""
+
+        if use_cuda(self.getOrDefault(self.device)) or self.getOrDefault(self.use_gpu):
+            ss = _get_spark_session()
+            sc = ss.sparkContext
+
+            if _is_local(sc):
+                # Support GPU training in Spark local mode is just for debugging
+                # purposes, so it's okay for printing the below warning instead of
+                # checking the real gpu numbers and raising the exception.
+                get_logger(self.__class__.__name__).warning(
+                    "You have enabled GPU in spark local mode. Please make sure your"
+                    " local node has at least %d GPUs",
+                    self.getOrDefault(self.num_workers),
+                )
+            else:
+                executor_gpus = sc.getConf().get("spark.executor.resource.gpu.amount")
+                if executor_gpus is None:
+                    raise ValueError(
+                        "The `spark.executor.resource.gpu.amount` is required for training"
+                        " on GPU."
+                    )
+
+                if not (ss.version >= "3.4.0" and _is_standalone_or_localcluster(sc)):
+                    # We will enable stage-level scheduling in spark 3.4.0+ which doesn't
+                    # require spark.task.resource.gpu.amount to be set explicitly
+                    gpu_per_task = sc.getConf().get("spark.task.resource.gpu.amount")
+                    if gpu_per_task is not None:
+                        if float(gpu_per_task) < 1.0:
+                            raise ValueError(
+                                "XGBoost doesn't support GPU fractional configurations. "
+                                "Please set `spark.task.resource.gpu.amount=spark.executor"
+                                ".resource.gpu.amount`"
+                            )
+
+                        if float(gpu_per_task) > 1.0:
+                            get_logger(self.__class__.__name__).warning(
+                                "%s GPUs for each Spark task is configured, but each "
+                                "XGBoost training task uses only 1 GPU.",
+                                gpu_per_task,
+                            )
+                    else:
+                        raise ValueError(
+                            "The `spark.task.resource.gpu.amount` is required for training"
+                            " on GPU."
+                        )
+
     def _validate_params(self) -> None:
         # pylint: disable=too-many-branches
         init_model = self.getOrDefault("xgb_model")
@@ -421,53 +471,7 @@ class _SparkXGBParams(
                     "`pyspark.ml.linalg.Vector` type."
                 )
 
-        if use_cuda(self.getOrDefault(self.device)) or self.getOrDefault(self.use_gpu):
-            gpu_per_task = (
-                _get_spark_session()
-                .sparkContext.getConf()
-                .get("spark.task.resource.gpu.amount")
-            )
-
-            is_local = _is_local(_get_spark_session().sparkContext)
-
-            if is_local:
-                # checking spark local mode.
-                if gpu_per_task is not None:
-                    raise RuntimeError(
-                        "The spark local mode does not support gpu configuration."
-                        "Please remove spark.executor.resource.gpu.amount and "
-                        "spark.task.resource.gpu.amount"
-                    )
-
-                # Support GPU training in Spark local mode is just for debugging
-                # purposes, so it's okay for printing the below warning instead of
-                # checking the real gpu numbers and raising the exception.
-                get_logger(self.__class__.__name__).warning(
-                    "You have enabled GPU in spark local mode. Please make sure your"
-                    " local node has at least %d GPUs",
-                    self.getOrDefault(self.num_workers),
-                )
-            else:
-                # checking spark non-local mode.
-                if gpu_per_task is not None:
-                    if float(gpu_per_task) < 1.0:
-                        raise ValueError(
-                            "XGBoost doesn't support GPU fractional configurations. "
-                            "Please set `spark.task.resource.gpu.amount=spark.executor"
-                            ".resource.gpu.amount`"
-                        )
-
-                    if float(gpu_per_task) > 1.0:
-                        get_logger(self.__class__.__name__).warning(
-                            "%s GPUs for each Spark task is configured, but each "
-                            "XGBoost training task uses only 1 GPU.",
-                            gpu_per_task,
-                        )
-                else:
-                    raise ValueError(
-                        "The `spark.task.resource.gpu.amount` is required for training"
-                        " on GPU."
-                    )
+        self._validate_gpu_params()
 
 
 def _validate_and_convert_feature_col_as_float_col_list(
@@ -592,6 +596,8 @@ class _SparkXGBEstimator(Estimator, _SparkXGBParams, MLReadable, MLWritable):
             arbitrary_params_dict={},
         )
 
+        self.logger = get_logger(self.__class__.__name__)
+
     def setParams(self, **kwargs: Any) -> None:  # pylint: disable=invalid-name
         """
         Set params for the estimator.
@@ -894,6 +900,116 @@ class _SparkXGBEstimator(Estimator, _SparkXGBParams, MLReadable, MLWritable):
 
         return booster_params, train_call_kwargs_params, dmatrix_kwargs
 
+    def _skip_stage_level_scheduling(self) -> bool:
+        # pylint: disable=too-many-return-statements
+        """Check if stage-level scheduling is not needed,
+        return true to skip stage-level scheduling"""
+
+        if use_cuda(self.getOrDefault(self.device)) or self.getOrDefault(self.use_gpu):
+            ss = _get_spark_session()
+            sc = ss.sparkContext
+
+            if ss.version < "3.4.0":
+                self.logger.info(
+                    "Stage-level scheduling in xgboost requires spark version 3.4.0+"
+                )
+                return True
+
+            if not _is_standalone_or_localcluster(sc):
+                self.logger.info(
+                    "Stage-level scheduling in xgboost requires spark standalone or "
+                    "local-cluster mode"
+                )
+                return True
+
+            executor_cores = sc.getConf().get("spark.executor.cores")
+            executor_gpus = sc.getConf().get("spark.executor.resource.gpu.amount")
+            if executor_cores is None or executor_gpus is None:
+                self.logger.info(
+                    "Stage-level scheduling in xgboost requires spark.executor.cores, "
+                    "spark.executor.resource.gpu.amount to be set."
+                )
+                return True
+
+            if int(executor_cores) == 1:
+                # there will be only 1 task running at any time.
+                self.logger.info(
+                    "Stage-level scheduling in xgboost requires spark.executor.cores > 1 "
+                )
+                return True
+
+            if int(executor_gpus) > 1:
+                # For spark.executor.resource.gpu.amount > 1, we suppose user knows how to configure
+                # to make xgboost run successfully.
+                #
+                self.logger.info(
+                    "Stage-level scheduling in xgboost will not work "
+                    "when spark.executor.resource.gpu.amount>1"
+                )
+                return True
+
+            task_gpu_amount = sc.getConf().get("spark.task.resource.gpu.amount")
+
+            if task_gpu_amount is None:
+                # The ETL tasks will not grab a gpu when spark.task.resource.gpu.amount is not set,
+                # but with stage-level scheduling, we can make training task grab the gpu.
+                return False
+
+            if float(task_gpu_amount) == float(executor_gpus):
+                # spark.executor.resource.gpu.amount=spark.task.resource.gpu.amount "
+                # results in only 1 task running at a time, which may cause perf issue.
+                return True
+
+            # We can enable stage-level scheduling
+            return False
+
+        # CPU training doesn't require stage-level scheduling
+        return True
+
+    def _try_stage_level_scheduling(self, rdd: RDD) -> RDD:
+        """Try to enable stage-level scheduling"""
+
+        if self._skip_stage_level_scheduling():
+            return rdd
+
+        ss = _get_spark_session()
+
+        # executor_cores will not be None
+        executor_cores = ss.sparkContext.getConf().get("spark.executor.cores")
+        assert executor_cores is not None
+
+        # Spark-rapids is a project to leverage GPUs to accelerate spark SQL.
+        # If spark-rapids is enabled, to avoid GPU OOM, we don't allow other
+        # ETL gpu tasks running alongside training tasks.
+        spark_plugins = ss.conf.get("spark.plugins", " ")
+        assert spark_plugins is not None
+        spark_rapids_sql_enabled = ss.conf.get("spark.rapids.sql.enabled", "true")
+        assert spark_rapids_sql_enabled is not None
+
+        task_cores = (
+            int(executor_cores)
+            if "com.nvidia.spark.SQLPlugin" in spark_plugins
+            and "true" == spark_rapids_sql_enabled.lower()
+            else (int(executor_cores) // 2) + 1
+        )
+
+        # Each training task requires cpu cores > total executor cores//2 + 1 which can
+        # make sure the tasks be sent to different executors.
+        #
+        # Please note that we can't use GPU to limit the concurrent tasks because of
+        # https://issues.apache.org/jira/browse/SPARK-45527.
+
+        task_gpus = 1.0
+        treqs = TaskResourceRequests().cpus(task_cores).resource("gpu", task_gpus)
+        rp = ResourceProfileBuilder().require(treqs).build
+
+        self.logger.info(
+            "XGBoost training tasks require the resource(cores=%s, gpu=%s).",
+            task_cores,
+            task_gpus,
+        )
+        return rdd.withResources(rp)
+
     def _fit(self, dataset: DataFrame) -> "_SparkXGBModel":
         # pylint: disable=too-many-statements, too-many-locals
         self._validate_params()
@@ -994,14 +1110,16 @@ class _SparkXGBEstimator(Estimator, _SparkXGBParams, MLReadable, MLWritable):
                 )
 
         def _run_job() -> Tuple[str, str]:
-            ret = (
+            rdd = (
                 dataset.mapInPandas(
-                    _train_booster, schema="config string, booster string"  # type: ignore
+                    _train_booster,  # type: ignore
+                    schema="config string, booster string",
                 )
                 .rdd.barrier()
                 .mapPartitions(lambda x: x)
-                .collect()[0]
             )
+            rdd_with_resource = self._try_stage_level_scheduling(rdd)
+            ret = rdd_with_resource.collect()[0]
             return ret[0], ret[1]
 
         get_logger("XGBoost-PySpark").info(

python-package/xgboost/spark/utils.py

@@ -129,6 +129,13 @@ def _is_local(spark_context: SparkContext) -> bool:
     return spark_context._jsc.sc().isLocal()
 
 
+def _is_standalone_or_localcluster(spark_context: SparkContext) -> bool:
+    master = spark_context.getConf().get("spark.master")
+    return master is not None and (
+        master.startswith("spark://") or master.startswith("local-cluster")
+    )
+
+
 def _get_gpu_id(task_context: TaskContext) -> int:
     """Get the gpu id from the task resources"""
     if task_context is None:

src/common/io.h

@@ -384,7 +384,8 @@ class PrivateMmapConstStream : public AlignedResourceReadStream {
    * @param length    See the `length` parameter of `mmap` for details.
    */
   explicit PrivateMmapConstStream(std::string path, std::size_t offset, std::size_t length)
-      : AlignedResourceReadStream{std::make_shared<MmapResource>(path, offset, length)} {}
+      : AlignedResourceReadStream{std::shared_ptr<MmapResource>{  // NOLINT
+            new MmapResource{std::move(path), offset, length}}} {}
   ~PrivateMmapConstStream() noexcept(false) override;
 };
 

src/common/ref_resource_view.h

@@ -76,7 +76,7 @@ class RefResourceView {
 
   [[nodiscard]] size_type size() const { return size_; }  // NOLINT
   [[nodiscard]] size_type size_bytes() const {            // NOLINT
-    return Span{data(), size()}.size_bytes();
+    return Span<const value_type>{data(), size()}.size_bytes();
   }
   [[nodiscard]] value_type* data() { return ptr_; };              // NOLINT
   [[nodiscard]] value_type const* data() const { return ptr_; };  // NOLINT

src/common/threading_utils.cc

@@ -3,14 +3,23 @@
  */
 #include "threading_utils.h"
 
-#include <fstream>
-#include <string>
+#include <algorithm>   // for max
+#include <exception>   // for exception
+#include <filesystem>  // for path, exists
+#include <fstream>     // for ifstream
+#include <string>      // for string
 
-#include "xgboost/logging.h"
+#include "common.h"  // for DivRoundUp
 
-namespace xgboost {
-namespace common {
-int32_t GetCfsCPUCount() noexcept {
+namespace xgboost::common {
+/**
+ * Modified from
+ * github.com/psiha/sweater/blob/master/include/boost/sweater/hardware_concurrency.hpp
+ *
+ * MIT License: Copyright (c) 2016 Domagoj Šarić
+ */
+std::int32_t GetCGroupV1Count(std::filesystem::path const& quota_path,
+                              std::filesystem::path const& peroid_path) {
 #if defined(__linux__)
   // https://bugs.openjdk.java.net/browse/JDK-8146115
   // http://hg.openjdk.java.net/jdk/hs/rev/7f22774a5f42
@@ -31,8 +40,8 @@ int32_t GetCfsCPUCount() noexcept {
     }
   };
   // complete fair scheduler from Linux
-  auto const cfs_quota(read_int("/sys/fs/cgroup/cpu/cpu.cfs_quota_us"));
-  auto const cfs_period(read_int("/sys/fs/cgroup/cpu/cpu.cfs_period_us"));
+  auto const cfs_quota(read_int(quota_path.c_str()));
+  auto const cfs_period(read_int(peroid_path.c_str()));
   if ((cfs_quota > 0) && (cfs_period > 0)) {
     return std::max(cfs_quota / cfs_period, 1);
   }
@@ -40,6 +49,47 @@ int32_t GetCfsCPUCount() noexcept {
   return -1;
 }
 
+std::int32_t GetCGroupV2Count(std::filesystem::path const& bandwidth_path) noexcept(true) {
+  std::int32_t cnt{-1};
+#if defined(__linux__)
+  namespace fs = std::filesystem;
+
+  std::int32_t a{0}, b{0};
+
+  auto warn = [] { LOG(WARNING) << "Invalid cgroupv2 file."; };
+  try {
+    std::ifstream fin{bandwidth_path, std::ios::in};
+    fin >> a;
+    fin >> b;
+  } catch (std::exception const&) {
+    warn();
+    return cnt;
+  }
+  if (a > 0 && b > 0) {
+    cnt = std::max(common::DivRoundUp(a, b), 1);
+  }
+#endif  //  defined(__linux__)
+  return cnt;
+}
+
+std::int32_t GetCfsCPUCount() noexcept {
+  namespace fs = std::filesystem;
+  fs::path const bandwidth_path{"/sys/fs/cgroup/cpu.max"};
+  auto has_v2 = fs::exists(bandwidth_path);
+  if (has_v2) {
+    return GetCGroupV2Count(bandwidth_path);
+  }
+
+  fs::path const quota_path{"/sys/fs/cgroup/cpu/cpu.cfs_quota_us"};
+  fs::path const peroid_path{"/sys/fs/cgroup/cpu/cpu.cfs_period_us"};
+  auto has_v1 = fs::exists(quota_path) && fs::exists(peroid_path);
+  if (has_v1) {
+    return GetCGroupV1Count(quota_path, peroid_path);
+  }
+
+  return -1;
+}
+
 std::int32_t OmpGetNumThreads(std::int32_t n_threads) {
   // Don't use parallel if we are in a parallel region.
   if (omp_in_parallel()) {
@@ -54,5 +104,4 @@ std::int32_t OmpGetNumThreads(std::int32_t n_threads) {
   n_threads = std::max(n_threads, 1);
   return n_threads;
 }
-}  // namespace common
-}  // namespace xgboost
+}  // namespace xgboost::common

src/common/threading_utils.h

@@ -253,11 +253,6 @@ inline std::int32_t OmpGetThreadLimit() {
  * \brief Get thread limit from CFS.
  *
  *   This function has non-trivial overhead and should not be called repeatly.
- *
- * Modified from
- * github.com/psiha/sweater/blob/master/include/boost/sweater/hardware_concurrency.hpp
- *
- * MIT License: Copyright (c) 2016 Domagoj Šarić
  */
 std::int32_t GetCfsCPUCount() noexcept;
 

src/learner.cc

@@ -1317,7 +1317,9 @@ class LearnerImpl : public LearnerIO {
     if (metrics_.empty() && tparam_.disable_default_eval_metric <= 0) {
       metrics_.emplace_back(Metric::Create(obj_->DefaultEvalMetric(), &ctx_));
       auto config = obj_->DefaultMetricConfig();
-      metrics_.back()->LoadConfig(config);
+      if (!IsA<Null>(config)) {
+        metrics_.back()->LoadConfig(config);
+      }
       metrics_.back()->Configure({cfg_.begin(), cfg_.end()});
     }
 

src/objective/regression_obj.cu

@@ -268,6 +268,13 @@ class PseudoHuberRegression : public FitIntercept {
     }
     FromJson(in["pseudo_huber_param"], &param_);
   }
+  [[nodiscard]] Json DefaultMetricConfig() const override {
+    CHECK(param_.GetInitialised());
+    Json config{Object{}};
+    config["name"] = String{this->DefaultEvalMetric()};
+    config["pseudo_huber_param"] = ToJson(param_);
+    return config;
+  }
 };
 
 XGBOOST_REGISTER_OBJECTIVE(PseudoHuberRegression, "reg:pseudohubererror")

tests/ci_build/Dockerfile.gpu_build_centos7

@@ -1,5 +1,5 @@
 ARG CUDA_VERSION_ARG
-FROM nvidia/cuda:$CUDA_VERSION_ARG-devel-centos7
+FROM nvcr.io/nvidia/cuda:$CUDA_VERSION_ARG-devel-centos7
 ARG CUDA_VERSION_ARG
 ARG NCCL_VERSION_ARG
 ARG RAPIDS_VERSION_ARG

tests/ci_build/Dockerfile.gpu_build_r_centos7

@@ -1,5 +1,5 @@
 ARG CUDA_VERSION_ARG
-FROM nvidia/cuda:$CUDA_VERSION_ARG-devel-centos7
+FROM nvcr.io/nvidia/cuda:$CUDA_VERSION_ARG-devel-centos7
 ARG CUDA_VERSION_ARG
 
 # Install all basic requirements

tests/ci_build/Dockerfile.jvm_gpu_build

@@ -1,5 +1,5 @@
 ARG CUDA_VERSION_ARG
-FROM nvidia/cuda:$CUDA_VERSION_ARG-devel-centos7
+FROM nvcr.io/nvidia/cuda:$CUDA_VERSION_ARG-devel-centos7
 ARG CUDA_VERSION_ARG
 ARG NCCL_VERSION_ARG
 

tests/cpp/common/test_io.cc

@@ -148,7 +148,8 @@ TEST(IO, Resource) {
     fout << 1.0 << std::endl;
     fout.close();
 
-    auto resource = std::make_shared<MmapResource>(path, 0, sizeof(double));
+    auto resource = std::shared_ptr<MmapResource>{
+      new MmapResource{path, 0, sizeof(double)}};
     ASSERT_EQ(resource->Size(), sizeof(double));
     ASSERT_EQ(resource->Type(), ResourceHandler::kMmap);
     ASSERT_EQ(resource->DataAs<double>()[0], val);

tests/cpp/objective/test_objective.cc

@@ -6,6 +6,7 @@
 #include <xgboost/objective.h>
 
 #include "../helpers.h"
+#include "../objective_helpers.h"
 
 TEST(Objective, UnknownFunction) {
   xgboost::ObjFunction* obj = nullptr;
@@ -43,4 +44,61 @@ TEST(Objective, PredTransform) {
     ASSERT_TRUE(predts.HostCanWrite());
   }
 }
+
+class TestDefaultObjConfig : public ::testing::TestWithParam<std::string> {
+  Context ctx_;
+
+ public:
+  void Run(std::string objective) {
+    auto Xy = MakeFmatForObjTest(objective);
+    std::unique_ptr<Learner> learner{Learner::Create({Xy})};
+    std::unique_ptr<ObjFunction> objfn{ObjFunction::Create(objective, &ctx_)};
+
+    learner->SetParam("objective", objective);
+    if (objective.find("multi") != std::string::npos) {
+      learner->SetParam("num_class", "3");
+      objfn->Configure(Args{{"num_class", "3"}});
+    } else if (objective.find("quantile") != std::string::npos) {
+      learner->SetParam("quantile_alpha", "0.5");
+      objfn->Configure(Args{{"quantile_alpha", "0.5"}});
+    } else {
+      objfn->Configure(Args{});
+    }
+    learner->Configure();
+    learner->UpdateOneIter(0, Xy);
+    learner->EvalOneIter(0, {Xy}, {"train"});
+    Json config{Object{}};
+    learner->SaveConfig(&config);
+    auto jobj = get<Object const>(config["learner"]["objective"]);
+
+    ASSERT_TRUE(jobj.find("name") != jobj.cend());
+    // FIXME(jiamingy): We should have the following check, but some legacy parameter like
+    // "pos_weight", "delta_step" in objectives are not in metrics.
+
+    // if (jobj.size() > 1) {
+    //   ASSERT_FALSE(IsA<Null>(objfn->DefaultMetricConfig()));
+    // }
+    auto mconfig = objfn->DefaultMetricConfig();
+    if (!IsA<Null>(mconfig)) {
+      // make sure metric can handle it
+      std::unique_ptr<Metric> metricfn{Metric::Create(get<String const>(mconfig["name"]), &ctx_)};
+      metricfn->LoadConfig(mconfig);
+      Json loaded(Object{});
+      metricfn->SaveConfig(&loaded);
+      metricfn->Configure(Args{});
+      ASSERT_EQ(mconfig, loaded);
+    }
+  }
+};
+
+TEST_P(TestDefaultObjConfig, Objective) {
+  std::string objective = GetParam();
+  this->Run(objective);
+}
+
+INSTANTIATE_TEST_SUITE_P(Objective, TestDefaultObjConfig,
+                         ::testing::ValuesIn(MakeObjNamesForTest()),
+                         [](const ::testing::TestParamInfo<TestDefaultObjConfig::ParamType>& info) {
+                           return ObjTestNameGenerator(info);
+                         });
 } // namespace xgboost

tests/cpp/objective_helpers.cc

@@ -0,0 +1,31 @@
+/**
+ * Copyright (c) 2023, XGBoost contributors
+ */
+#include "objective_helpers.h"
+
+#include "../../src/common/linalg_op.h"  // for begin, end
+#include "helpers.h"                     // for RandomDataGenerator
+
+namespace xgboost {
+std::shared_ptr<DMatrix> MakeFmatForObjTest(std::string const& obj) {
+  auto constexpr kRows = 10, kCols = 10;
+  auto p_fmat = RandomDataGenerator{kRows, kCols, 0}.GenerateDMatrix(true);
+  auto& h_upper = p_fmat->Info().labels_upper_bound_.HostVector();
+  auto& h_lower = p_fmat->Info().labels_lower_bound_.HostVector();
+  h_lower.resize(kRows);
+  h_upper.resize(kRows);
+  for (size_t i = 0; i < kRows; ++i) {
+    h_lower[i] = 1;
+    h_upper[i] = 10;
+  }
+  if (obj.find("rank:") != std::string::npos) {
+    auto h_label = p_fmat->Info().labels.HostView();
+    std::size_t k = 0;
+    for (auto& v : h_label) {
+      v = k % 2 == 0;
+      ++k;
+    }
+  }
+  return p_fmat;
+};
+}  // namespace xgboost

tests/cpp/objective_helpers.h

@@ -1,6 +1,8 @@
 /**
  * Copyright (c) 2023, XGBoost contributors
  */
+#pragma once
+
 #include <dmlc/registry.h>  // for Registry
 #include <gtest/gtest.h>
 #include <xgboost/objective.h>  // for ObjFunctionReg
@@ -29,4 +31,6 @@ inline std::string ObjTestNameGenerator(const ::testing::TestParamInfo<ParamType
   }
   return name;
 };
+
+std::shared_ptr<DMatrix> MakeFmatForObjTest(std::string const& obj);
 }  // namespace xgboost

tests/cpp/test_learner.cc

@@ -655,33 +655,11 @@ TEST_F(InitBaseScore, InitWithPredict) { this->TestInitWithPredt(); }
 TEST_F(InitBaseScore, UpdateProcess) { this->TestUpdateProcess(); }
 
 class TestColumnSplit : public ::testing::TestWithParam<std::string> {
-  static auto MakeFmat(std::string const& obj) {
-    auto constexpr kRows = 10, kCols = 10;
-    auto p_fmat = RandomDataGenerator{kRows, kCols, 0}.GenerateDMatrix(true);
-    auto& h_upper = p_fmat->Info().labels_upper_bound_.HostVector();
-    auto& h_lower = p_fmat->Info().labels_lower_bound_.HostVector();
-    h_lower.resize(kRows);
-    h_upper.resize(kRows);
-    for (size_t i = 0; i < kRows; ++i) {
-      h_lower[i] = 1;
-      h_upper[i] = 10;
-    }
-    if (obj.find("rank:") != std::string::npos) {
-      auto h_label = p_fmat->Info().labels.HostView();
-      std::size_t k = 0;
-      for (auto& v : h_label) {
-        v = k % 2 == 0;
-        ++k;
-      }
-    }
-    return p_fmat;
-  };
-
   void TestBaseScore(std::string objective, float expected_base_score, Json expected_model) {
     auto const world_size = collective::GetWorldSize();
     auto const rank = collective::GetRank();
 
-    auto p_fmat = MakeFmat(objective);
+    auto p_fmat = MakeFmatForObjTest(objective);
     std::shared_ptr<DMatrix> sliced{p_fmat->SliceCol(world_size, rank)};
     std::unique_ptr<Learner> learner{Learner::Create({sliced})};
     learner->SetParam("tree_method", "approx");
@@ -705,7 +683,7 @@ class TestColumnSplit : public ::testing::TestWithParam<std::string> {
 
  public:
   void Run(std::string objective) {
-    auto p_fmat = MakeFmat(objective);
+    auto p_fmat = MakeFmatForObjTest(objective);
     std::unique_ptr<Learner> learner{Learner::Create({p_fmat})};
     learner->SetParam("tree_method", "approx");
     learner->SetParam("objective", objective);

tests/python/test_with_pandas.py

@@ -211,7 +211,7 @@ class TestPandas:
         y = np.random.randn(kRows)
         w = np.random.uniform(size=kRows).astype(np.float32)
         w_pd = pd.DataFrame(w)
-        data = xgb.DMatrix(X, y, w_pd)
+        data = xgb.DMatrix(X, y, weight=w_pd)
 
         assert data.num_row() == kRows
         assert data.num_col() == kCols
@@ -301,14 +301,14 @@ class TestPandas:
 
     @pytest.mark.parametrize("DMatrixT", [xgb.DMatrix, xgb.QuantileDMatrix])
     def test_nullable_type(self, DMatrixT) -> None:
-        from pandas.api.types import is_categorical_dtype
+        from xgboost.data import is_pd_cat_dtype
 
         for orig, df in pd_dtypes():
             if hasattr(df.dtypes, "__iter__"):
-                enable_categorical = any(is_categorical_dtype for dtype in df.dtypes)
+                enable_categorical = any(is_pd_cat_dtype(dtype) for dtype in df.dtypes)
             else:
                 # series
-                enable_categorical = is_categorical_dtype(df.dtype)
+                enable_categorical = is_pd_cat_dtype(df.dtype)
 
             f0_orig = orig[orig.columns[0]] if isinstance(orig, pd.DataFrame) else orig
             f0 = df[df.columns[0]] if isinstance(df, pd.DataFrame) else df