Make 1.4.2 patch release. (#6962)

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

.github/workflows/main.yml

@@ -31,7 +31,10 @@ jobs:
     - name: Run gtest binary
       run: |
         cd build
-        ctest --exclude-regex AllTestsInDMLCUnitTests --extra-verbose
+        # libomp internal error:
+        #   OMP: Error #131: Thread identifier invalid.
+        ./testxgboost  --gtest_filter="-HistIndexCreationWithExternalMemory.Test"
+        ctest -R TestXGBoostCLI --extra-verbose
 
   gtest-cpu-nonomp:
     name: Test Google C++ unittest (CPU Non-OMP)

.travis.yml

@@ -19,18 +19,10 @@ jobs:
       env: TASK=java_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+# the dependencies from homebrew is installed manually from setup script due to outdated image from travis.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

CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.13)
-project(xgboost LANGUAGES CXX C VERSION 1.4.1)
+project(xgboost LANGUAGES CXX C VERSION 1.4.2)
 include(cmake/Utils.cmake)
 list(APPEND CMAKE_MODULE_PATH "${xgboost_SOURCE_DIR}/cmake/modules")
 cmake_policy(SET CMP0022 NEW)

Jenkinsfile

@@ -179,10 +179,10 @@ 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'
@@ -231,6 +231,7 @@ 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"

R-package/DESCRIPTION

@@ -1,7 +1,7 @@
 Package: xgboost
 Type: Package
 Title: Extreme Gradient Boosting
-Version: 1.4.1.1
+Version: 1.4.2.1
 Date: 2020-08-28
 Authors@R: c(
   person("Tianqi", "Chen", role = c("aut"),

dev/release-pypi.py

@@ -49,7 +49,7 @@ def download_wheels(
     dir_URL: str,
     src_filename_prefix: str,
     target_filename_prefix: str,
-) -> None:
+) -> List[str]:
     """Download all binary wheels. dir_URL is the URL for remote directory storing the release
     wheels
 
@@ -72,6 +72,8 @@ def download_wheels(
         assert stderr.find("warning") == -1, "Unresolved warnings:\n" + stderr
         assert stdout.find("warning") == -1, "Unresolved warnings:\n" + stdout
 
+    return filenames
+
 
 def check_path():
     root = os.path.abspath(os.path.curdir)

include/xgboost/version_config.h

@@ -6,6 +6,6 @@
 
 #define XGBOOST_VER_MAJOR 1
 #define XGBOOST_VER_MINOR 4
-#define XGBOOST_VER_PATCH 1
+#define XGBOOST_VER_PATCH 2
 
 #endif  // XGBOOST_VERSION_CONFIG_H_

jvm-packages/create_jni.py

@@ -84,8 +84,9 @@ if __name__ == "__main__":
 
     print("building Java wrapper")
     with cd(".."):
-        maybe_makedirs("build")
-        with cd("build"):
+        build_dir = 'build-gpu' if cli_args.use_cuda == 'ON' else 'build'
+        maybe_makedirs(build_dir)
+        with cd(build_dir):
             if sys.platform == "win32":
                 # Force x64 build on Windows.
                 maybe_generator = ' -A x64'
@@ -114,6 +115,9 @@ if __name__ == "__main__":
             if gpu_arch_flag is not None:
                 args.append("%s" % gpu_arch_flag)
 
+            lib_dir = os.path.join(os.pardir, 'lib')
+            if os.path.exists(lib_dir):
+                shutil.rmtree(lib_dir)
             run("cmake .. " + " ".join(args) + maybe_generator)
             run("cmake --build . --config Release" + maybe_parallel_build)
 

jvm-packages/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>ml.dmlc</groupId>
     <artifactId>xgboost-jvm_2.12</artifactId>
-    <version>1.4.1</version>
+    <version>1.4.2</version>
     <packaging>pom</packaging>
     <name>XGBoost JVM Package</name>
     <description>JVM Package for XGBoost</description>

jvm-packages/xgboost4j-example/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.4.1</version>
+        <version>1.4.2</version>
     </parent>
     <artifactId>xgboost4j-example_2.12</artifactId>
-    <version>1.4.1</version>
+    <version>1.4.2</version>
     <packaging>jar</packaging>
     <build>
         <plugins>
@@ -26,7 +26,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-spark_${scala.binary.version}</artifactId>
-            <version>1.4.1</version>
+            <version>1.4.2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>
@@ -37,7 +37,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-flink_${scala.binary.version}</artifactId>
-            <version>1.4.1</version>
+            <version>1.4.2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.commons</groupId>

jvm-packages/xgboost4j-flink/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.4.1</version>
+        <version>1.4.2</version>
     </parent>
     <artifactId>xgboost4j-flink_2.12</artifactId>
-    <version>1.4.1</version>
+    <version>1.4.2</version>
     <build>
         <plugins>
             <plugin>
@@ -26,7 +26,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j_${scala.binary.version}</artifactId>
-            <version>1.4.1</version>
+            <version>1.4.2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.commons</groupId>

jvm-packages/xgboost4j-gpu/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.4.1</version>
+        <version>1.4.2</version>
     </parent>
     <artifactId>xgboost4j-gpu_2.12</artifactId>
-    <version>1.4.1</version>
+    <version>1.4.2</version>
     <packaging>jar</packaging>
 
     <dependencies>

jvm-packages/xgboost4j-gpu/src

@@ -1 +0,0 @@
-../xgboost4j/src/

jvm-packages/xgboost4j-gpu/src/main/java

@@ -0,0 +1 @@
+../../../xgboost4j/src/main/java/

jvm-packages/xgboost4j-gpu/src/main/resources/xgboost4j-version.properties

@@ -0,0 +1 @@
+../../../../xgboost4j/src/main/resources/xgboost4j-version.properties

jvm-packages/xgboost4j-gpu/src/main/scala

@@ -0,0 +1 @@
+../../../xgboost4j/src/main/scala/

jvm-packages/xgboost4j-gpu/src/native

@@ -0,0 +1 @@
+../../xgboost4j/src/native

jvm-packages/xgboost4j-gpu/src/test

@@ -0,0 +1 @@
+../../xgboost4j/src/test

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

@@ -6,7 +6,7 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.4.1</version>
+        <version>1.4.2</version>
     </parent>
     <artifactId>xgboost4j-spark-gpu_2.12</artifactId>
     <build>
@@ -24,7 +24,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j-gpu_${scala.binary.version}</artifactId>
-            <version>1.4.1</version>
+            <version>1.4.2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>

jvm-packages/xgboost4j-spark-gpu/src

@@ -1 +0,0 @@
-../xgboost4j-spark/src/

jvm-packages/xgboost4j-spark-gpu/src/main/scala

@@ -0,0 +1 @@
+../../../xgboost4j-spark/src/main/scala

jvm-packages/xgboost4j-spark-gpu/src/test

@@ -0,0 +1 @@
+../../xgboost4j-spark/src/test

jvm-packages/xgboost4j-spark/pom.xml

@@ -6,7 +6,7 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.4.1</version>
+        <version>1.4.2</version>
     </parent>
     <artifactId>xgboost4j-spark_2.12</artifactId>
     <build>
@@ -24,7 +24,7 @@
         <dependency>
             <groupId>ml.dmlc</groupId>
             <artifactId>xgboost4j_${scala.binary.version}</artifactId>
-            <version>1.4.1</version>
+            <version>1.4.2</version>
         </dependency>
         <dependency>
             <groupId>org.apache.spark</groupId>

jvm-packages/xgboost4j/pom.xml

@@ -6,10 +6,10 @@
     <parent>
         <groupId>ml.dmlc</groupId>
         <artifactId>xgboost-jvm_2.12</artifactId>
-        <version>1.4.1</version>
+        <version>1.4.2</version>
     </parent>
     <artifactId>xgboost4j_2.12</artifactId>
-    <version>1.4.1</version>
+    <version>1.4.2</version>
     <packaging>jar</packaging>
 
     <dependencies>

python-package/xgboost/VERSION

@@ -1 +1 @@
-1.4.1
+1.4.2

python-package/xgboost/core.py

@@ -229,6 +229,9 @@ def _numpy2ctypes_type(dtype):
 
 
 def _array_interface(data: np.ndarray) -> bytes:
+    assert (
+        data.dtype.hasobject is False
+    ), "Input data contains `object` dtype.  Expecting numeric data."
     interface = data.__array_interface__
     if "mask" in interface:
         interface["mask"] = interface["mask"].__array_interface__
@@ -1841,8 +1844,8 @@ class Booster(object):
                 )
 
         if isinstance(data, np.ndarray):
-            from .data import _maybe_np_slice
-            data = _maybe_np_slice(data, data.dtype)
+            from .data import _ensure_np_dtype
+            data, _ = _ensure_np_dtype(data, data.dtype)
             _check_call(
                 _LIB.XGBoosterPredictFromDense(
                     self.handle,
@@ -1872,7 +1875,9 @@ class Booster(object):
                 )
             )
             return _prediction_output(shape, dims, preds, False)
-        if lazy_isinstance(data, "cupy.core.core", "ndarray"):
+        if lazy_isinstance(data, "cupy.core.core", "ndarray") or lazy_isinstance(
+            data, "cupy._core.core", "ndarray"
+        ):
             from .data import _transform_cupy_array
             data = _transform_cupy_array(data)
             interface = data.__cuda_array_interface__
@@ -2027,7 +2032,7 @@ class Booster(object):
         """
         if isinstance(fout, (STRING_TYPES, os.PathLike)):
             fout = os.fspath(os.path.expanduser(fout))
-            fout = open(fout, 'w')
+            fout = open(fout, 'w')  # pylint: disable=consider-using-with
             need_close = True
         else:
             need_close = False

python-package/xgboost/data.py

@@ -104,6 +104,13 @@ def _is_numpy_array(data):
     return isinstance(data, (np.ndarray, np.matrix))
 
 
+def _ensure_np_dtype(data, dtype):
+    if data.dtype.hasobject:
+        data = data.astype(np.float32, copy=False)
+        dtype = np.float32
+    return data, dtype
+
+
 def _maybe_np_slice(data, dtype):
     '''Handle numpy slice.  This can be removed if we use __array_interface__.
     '''
@@ -118,6 +125,7 @@ def _maybe_np_slice(data, dtype):
             data = np.array(data, copy=False, dtype=dtype)
     except AttributeError:
         data = np.array(data, copy=False, dtype=dtype)
+    data, dtype = _ensure_np_dtype(data, dtype)
     return data
 
 

src/c_api/c_api.cc

@@ -734,8 +734,9 @@ XGB_DLL int XGBoosterPredictFromCSR(BoosterHandle handle, char const *indptr,
   API_BEGIN();
   CHECK_HANDLE();
   std::shared_ptr<xgboost::data::CSRArrayAdapter> x{
-      new xgboost::data::CSRArrayAdapter{
-          StringView{indptr}, StringView{indices}, StringView{data}, cols}};
+      new xgboost::data::CSRArrayAdapter{StringView{indptr},
+                                         StringView{indices}, StringView{data},
+                                         static_cast<size_t>(cols)}};
   std::shared_ptr<DMatrix> p_m {nullptr};
   if (m) {
     p_m = *static_cast<std::shared_ptr<DMatrix> *>(m);

src/c_api/c_api_utils.h

@@ -98,6 +98,10 @@ inline void CalcPredictShape(bool strict_shape, PredictionType type, size_t rows
       forest = std::max(static_cast<decltype(forest)>(1), forest);
       shape[3] = forest;
       *out_dim = shape.size();
+    } else if (chunksize == 1) {
+      *out_dim = 1;
+      shape.resize(*out_dim);
+      shape.front() = rows;
     } else {
       *out_dim = 2;
       shape.resize(*out_dim);

src/data/array_interface.h

@@ -229,8 +229,11 @@ class ArrayInterfaceHandler {
       }
       strides[1] = n;
     }
-    auto valid = (rows - 1) * strides[0] + (cols - 1) * strides[1] == (rows * cols) - 1;
-    CHECK(valid) << "Invalid strides in array.";
+
+    auto valid = rows * strides[0] + cols * strides[1] >= (rows * cols);
+    CHECK(valid) << "Invalid strides in array."
+                 << "  strides: (" << strides[0] << "," << strides[1]
+                 << "), shape: (" << rows << ", " << cols << ")";
   }
 
   static void* ExtractData(std::map<std::string, Json> const &column,

src/metric/auc.cc

@@ -87,8 +87,7 @@ std::tuple<float, float, float> BinaryAUC(std::vector<float> const &predts,
  * - Kleiman, Ross and Page, David. $AUC_{\mu}$: A Performance Metric for Multi-Class
  *   Machine Learning Models
  */
-float MultiClassOVR(std::vector<float> const& predts, MetaInfo const& info) {
-  auto n_classes = predts.size() / info.labels_.Size();
+float MultiClassOVR(std::vector<float> const& predts, MetaInfo const& info, size_t n_classes) {
   CHECK_NE(n_classes, 0);
   auto const& labels = info.labels_.ConstHostVector();
 
@@ -230,6 +229,10 @@ class EvalAUC : public Metric {
       info.labels_.SetDevice(tparam_->gpu_id);
       info.weights_.SetDevice(tparam_->gpu_id);
     }
+    //  We use the global size to handle empty dataset.
+    std::array<size_t, 2> meta{info.labels_.Size(), preds.Size()};
+    rabit::Allreduce<rabit::op::Max>(meta.data(), meta.size());
+
     if (!info.group_ptr_.empty()) {
       /**
        * learning to rank
@@ -261,16 +264,17 @@ class EvalAUC : public Metric {
         CHECK_LE(auc, 1) << "Total AUC across groups: " << auc * valid_groups
                          << ", valid groups: " << valid_groups;
       }
-    } else if (info.labels_.Size() != preds.Size() &&
-               preds.Size() % info.labels_.Size() == 0) {
+    } else if (meta[0] != meta[1] && meta[1] % meta[0] == 0) {
       /**
        * multi class
        */
+      size_t n_classes = meta[1] / meta[0];
+      CHECK_NE(n_classes, 0);
       if (tparam_->gpu_id == GenericParameter::kCpuId) {
-        auc = MultiClassOVR(preds.ConstHostVector(), info);
+        auc = MultiClassOVR(preds.ConstHostVector(), info, n_classes);
       } else {
         auc = GPUMultiClassAUCOVR(preds.ConstDeviceSpan(), info, tparam_->gpu_id,
-                                  &this->d_cache_);
+                                  &this->d_cache_, n_classes);
       }
     } else {
       /**
@@ -323,7 +327,8 @@ GPUBinaryAUC(common::Span<float const> predts, MetaInfo const &info,
 }
 
 float GPUMultiClassAUCOVR(common::Span<float const> predts, MetaInfo const &info,
-                          int32_t device, std::shared_ptr<DeviceAUCCache>* cache) {
+                          int32_t device, std::shared_ptr<DeviceAUCCache>* cache,
+                          size_t n_classes) {
   common::AssertGPUSupport();
   return 0;
 }

src/metric/auc.cu

@@ -61,10 +61,12 @@ struct DeviceAUCCache {
       neg_pos.resize(sorted_idx.size());
       if (is_multi) {
         predts_t.resize(sorted_idx.size());
-        reducer.reset(new dh::AllReducer);
-        reducer->Init(rabit::GetRank());
       }
     }
+    if (is_multi && !reducer) {
+      reducer.reset(new dh::AllReducer);
+      reducer->Init(device);
+    }
   }
 };
 
@@ -197,12 +199,48 @@ XGBOOST_DEVICE size_t LastOf(size_t group, common::Span<Idx> indptr) {
   return indptr[group + 1] - 1;
 }
 
+
+float ScaleClasses(common::Span<float> results, common::Span<float> local_area,
+                   common::Span<float> fp, common::Span<float> tp,
+                   common::Span<float> auc, std::shared_ptr<DeviceAUCCache> cache,
+                   size_t n_classes) {
+  dh::XGBDeviceAllocator<char> alloc;
+  if (rabit::IsDistributed()) {
+    CHECK_EQ(dh::CudaGetPointerDevice(results.data()), dh::CurrentDevice());
+    cache->reducer->AllReduceSum(results.data(), results.data(), results.size());
+  }
+  auto reduce_in = dh::MakeTransformIterator<thrust::pair<float, float>>(
+      thrust::make_counting_iterator(0), [=] __device__(size_t i) {
+        if (local_area[i] > 0) {
+          return thrust::make_pair(auc[i] / local_area[i] * tp[i], tp[i]);
+        }
+        return thrust::make_pair(std::numeric_limits<float>::quiet_NaN(), 0.0f);
+      });
+
+  float tp_sum;
+  float auc_sum;
+  thrust::tie(auc_sum, tp_sum) = thrust::reduce(
+      thrust::cuda::par(alloc), reduce_in, reduce_in + n_classes,
+      thrust::make_pair(0.0f, 0.0f),
+      [=] __device__(auto const &l, auto const &r) {
+        return thrust::make_pair(l.first + r.first, l.second + r.second);
+      });
+  if (tp_sum != 0 && !std::isnan(auc_sum)) {
+    auc_sum /= tp_sum;
+  } else {
+    return std::numeric_limits<float>::quiet_NaN();
+  }
+  return auc_sum;
+}
+
 /**
  * MultiClass implementation is similar to binary classification, except we need to split
  * up each class in all kernels.
  */
 float GPUMultiClassAUCOVR(common::Span<float const> predts, MetaInfo const &info,
-                          int32_t device, std::shared_ptr<DeviceAUCCache>* p_cache) {
+                          int32_t device, std::shared_ptr<DeviceAUCCache>* p_cache,
+                          size_t n_classes) {
+  dh::safe_cuda(cudaSetDevice(device));
   auto& cache = *p_cache;
   if (!cache) {
     cache.reset(new DeviceAUCCache);
@@ -213,8 +251,19 @@ float GPUMultiClassAUCOVR(common::Span<float const> predts, MetaInfo const &info
   auto weights = info.weights_.ConstDeviceSpan();
 
   size_t n_samples = labels.size();
-  size_t n_classes = predts.size() / labels.size();
-  CHECK_NE(n_classes, 0);
+
+  if (n_samples == 0) {
+    dh::TemporaryArray<float> resutls(n_classes * 4, 0.0f);
+    auto d_results = dh::ToSpan(resutls);
+    dh::LaunchN(device, n_classes * 4, [=]__device__(size_t i) {
+      d_results[i] = 0.0f;
+    });
+    auto local_area = d_results.subspan(0, n_classes);
+    auto fp = d_results.subspan(n_classes, n_classes);
+    auto tp = d_results.subspan(2 * n_classes, n_classes);
+    auto auc = d_results.subspan(3 * n_classes, n_classes);
+    return ScaleClasses(d_results, local_area, fp, tp, auc, cache, n_classes);
+  }
 
   /**
    * Create sorted index for each class
@@ -377,32 +426,7 @@ float GPUMultiClassAUCOVR(common::Span<float const> predts, MetaInfo const &info
     tp[c] = last.second;
     local_area[c] = last.first * last.second;
   });
-  if (rabit::IsDistributed()) {
-    cache->reducer->AllReduceSum(resutls.data().get(), resutls.data().get(),
-                                 resutls.size());
-  }
-  auto reduce_in = dh::MakeTransformIterator<thrust::pair<float, float>>(
-      thrust::make_counting_iterator(0), [=] __device__(size_t i) {
-        if (local_area[i] > 0) {
-          return thrust::make_pair(auc[i] / local_area[i] * tp[i], tp[i]);
-        }
-        return thrust::make_pair(std::numeric_limits<float>::quiet_NaN(), 0.0f);
-      });
-
-  float tp_sum;
-  float auc_sum;
-  thrust::tie(auc_sum, tp_sum) = thrust::reduce(
-      thrust::cuda::par(alloc), reduce_in, reduce_in + n_classes,
-      thrust::make_pair(0.0f, 0.0f),
-      [=] __device__(auto const &l, auto const &r) {
-        return thrust::make_pair(l.first + r.first, l.second + r.second);
-      });
-  if (tp_sum != 0 && !std::isnan(auc_sum)) {
-    auc_sum /= tp_sum;
-  } else {
-    return std::numeric_limits<float>::quiet_NaN();
-  }
-  return auc_sum;
+  return ScaleClasses(d_results, local_area, fp, tp, auc, cache, n_classes);
 }
 
 namespace {

src/metric/auc.h

@@ -26,7 +26,8 @@ GPUBinaryAUC(common::Span<float const> predts, MetaInfo const &info,
              int32_t device, std::shared_ptr<DeviceAUCCache> *p_cache);
 
 float GPUMultiClassAUCOVR(common::Span<float const> predts, MetaInfo const &info,
-                          int32_t device, std::shared_ptr<DeviceAUCCache>* cache);
+                          int32_t device, std::shared_ptr<DeviceAUCCache>* cache,
+                          size_t n_classes);
 
 std::pair<float, uint32_t>
 GPURankingAUC(common::Span<float const> predts, MetaInfo const &info,

tests/ci_build/conda_env/macos_cpu_test.yml

@@ -15,6 +15,7 @@ dependencies:
 - matplotlib
 - dask
 - distributed
+- graphviz
 - python-graphviz
 - hypothesis
 - astroid

tests/ci_build/insert_vcomp140.py

@@ -11,7 +11,7 @@ vcomp140_path = 'C:\\Windows\\System32\\vcomp140.dll'
 
 for wheel_path in sorted(glob.glob(sys.argv[1])):
     m = re.search(r'xgboost-(.*)-py3', wheel_path)
-    assert m
+    assert m, f'wheel_path = {wheel_path}'
     version = m.group(1)
 
     with zipfile.ZipFile(wheel_path, 'a') as f:

tests/ci_build/rename_whl.py

@@ -26,8 +26,9 @@ dirname, basename = os.path.dirname(whl_path), os.path.basename(whl_path)
 with cd(dirname):
     tokens = basename.split('-')
     assert len(tokens) == 5
+    version = tokens[1].split('+')[0]
     keywords = {'pkg_name': tokens[0],
-                'version': tokens[1],
+                'version': version,
                 'commit_id': commit_id,
                 'platform_tag': platform_tag}
     new_name = '{pkg_name}-{version}+{commit_id}-py3-none-{platform_tag}.whl'.format(**keywords)

tests/python-gpu/test_gpu_with_dask.py

@@ -277,7 +277,7 @@ class TestDistributedGPU:
             X = dask_cudf.from_dask_dataframe(dd.from_dask_array(X_))
             y = dask_cudf.from_dask_dataframe(dd.from_dask_array(y_))
             w = dask_cudf.from_dask_dataframe(dd.from_dask_array(w_))
-            run_dask_classifier(X, y, w, model, client, 10)
+            run_dask_classifier(X, y, w, model, "gpu_hist", client, 10)
 
     @pytest.mark.skipif(**tm.no_dask())
     @pytest.mark.skipif(**tm.no_dask_cuda())

tests/python/test_linear.py

@@ -57,15 +57,13 @@ class TestLinear:
         param['updater'] = 'shotgun'
         param = dataset.set_params(param)
         result = train_result(param, dataset.get_dmat(), num_rounds)['train'][dataset.metric]
-        # shotgun is non-deterministic, so we relax the test by sampling
-        # result.
+        # shotgun is non-deterministic, so we relax the test by only using first and last
+        # iteration.
         if len(result) > 2:
-            sampled_result = [score for i, score in enumerate(result)
-                              if i % 2 == 0]
-            sampled_result[-1] = result[-1]  # make sure the last one is used
+            sampled_result = (result[0], result[-1])
         else:
             sampled_result = result
-        assert tm.non_increasing(sampled_result, 1e-3)
+        assert tm.non_increasing(sampled_result)
 
     @given(parameter_strategy, strategies.integers(10, 50),
            tm.dataset_strategy, strategies.floats(1e-5, 2.0),

tests/python/test_predict.py

@@ -75,6 +75,11 @@ def run_predict_leaf(predictor):
     first = sliced[0, ...]
 
     assert np.prod(first.shape) == classes * num_parallel_tree * ntree_limit
+
+    # When there's only 1 tree, the output is a 1 dim vector
+    booster = xgb.train({"tree_method": "hist"}, num_boost_round=1, dtrain=m)
+    assert booster.predict(m, pred_leaf=True).shape == (rows, )
+
     return leaf
 
 
@@ -150,6 +155,14 @@ class TestInplacePredict:
         predt_from_array = booster.inplace_predict(X[:10, ...], missing=self.missing)
         predt_from_dmatrix = booster.predict(test)
 
+        X_obj = X.copy().astype(object)
+
+        assert X_obj.dtype.hasobject is True
+        assert X.dtype.hasobject is False
+        np.testing.assert_allclose(
+            booster.inplace_predict(X_obj), booster.inplace_predict(X)
+        )
+
         np.testing.assert_allclose(predt_from_dmatrix, predt_from_array)
 
         predt_from_array = booster.inplace_predict(
@@ -187,8 +200,13 @@ class TestInplacePredict:
         arr_predt = booster.inplace_predict(X)
         dmat_predt = booster.predict(xgb.DMatrix(X))
 
+        X = df.values
+        X = np.asfortranarray(X)
+        fort_predt = booster.inplace_predict(X)
+
         np.testing.assert_allclose(dmat_predt, arr_predt)
         np.testing.assert_allclose(df_predt, arr_predt)
+        np.testing.assert_allclose(fort_predt, arr_predt)
 
     def test_base_margin(self):
         booster = self.booster

tests/python/test_with_dask.py

@@ -317,6 +317,7 @@ def run_dask_classifier(
     y: xgb.dask._DaskCollection,
     w: xgb.dask._DaskCollection,
     model: str,
+    tree_method: Optional[str],
     client: "Client",
     n_classes,
 ) -> None:
@@ -324,11 +325,11 @@ def run_dask_classifier(
 
     if model == "boosting":
         classifier = xgb.dask.DaskXGBClassifier(
-            verbosity=1, n_estimators=2, eval_metric=metric
+            verbosity=1, n_estimators=2, eval_metric=metric, tree_method=tree_method
         )
     else:
         classifier = xgb.dask.DaskXGBRFClassifier(
-            verbosity=1, n_estimators=2, eval_metric=metric
+            verbosity=1, n_estimators=2, eval_metric=metric, tree_method=tree_method
         )
 
     assert classifier._estimator_type == "classifier"
@@ -397,12 +398,12 @@ def run_dask_classifier(
 def test_dask_classifier(model: str, client: "Client") -> None:
     X, y, w = generate_array(with_weights=True)
     y = (y * 10).astype(np.int32)
-    run_dask_classifier(X, y, w, model, client, 10)
+    run_dask_classifier(X, y, w, model, None, client, 10)
 
     y_bin = y.copy()
     y_bin[y > 5] = 1.0
     y_bin[y <= 5] = 0.0
-    run_dask_classifier(X, y_bin, w, model, client, 2)
+    run_dask_classifier(X, y_bin, w, model, None, client, 2)
 
 
 @pytest.mark.skipif(**tm.no_sklearn())
@@ -568,22 +569,26 @@ def run_empty_dmatrix_auc(client: "Client", tree_method: str, n_workers: int) ->
     # multiclass
     X_, y_ = make_classification(
         n_samples=n_samples,
-        n_classes=10,
+        n_classes=n_workers,
         n_informative=n_features,
         n_redundant=0,
         n_repeated=0
     )
+    for i in range(y_.shape[0]):
+        y_[i] = i % n_workers
     X = dd.from_array(X_, chunksize=10)
     y = dd.from_array(y_, chunksize=10)
 
     n_samples = n_workers - 1
     valid_X_, valid_y_ = make_classification(
         n_samples=n_samples,
-        n_classes=10,
+        n_classes=n_workers,
         n_informative=n_features,
         n_redundant=0,
         n_repeated=0
     )
+    for i in range(valid_y_.shape[0]):
+        valid_y_[i] = i % n_workers
     valid_X = dd.from_array(valid_X_, chunksize=n_samples)
     valid_y = dd.from_array(valid_y_, chunksize=n_samples)
 
@@ -594,9 +599,9 @@ def run_empty_dmatrix_auc(client: "Client", tree_method: str, n_workers: int) ->
 
 
 def test_empty_dmatrix_auc() -> None:
-    with LocalCluster(n_workers=2) as cluster:
+    with LocalCluster(n_workers=8) as cluster:
         with Client(cluster) as client:
-            run_empty_dmatrix_auc(client, "hist", 2)
+            run_empty_dmatrix_auc(client, "hist", 8)
 
 
 def run_auc(client: "Client", tree_method: str) -> None:
@@ -1023,7 +1028,17 @@ class TestWithDask:
                                  evals=[(m, 'train')])['history']
         note(history)
         history = history['train'][dataset.metric]
-        assert tm.non_increasing(history)
+
+        def is_stump():
+            return params["max_depth"] == 1 or params["max_leaves"] == 1
+
+        def minimum_bin():
+            return "max_bin" in params and params["max_bin"] == 2
+
+        if minimum_bin() and is_stump():
+            assert tm.non_increasing(history, tolerance=1e-3)
+        else:
+            assert tm.non_increasing(history)
         # Make sure that it's decreasing
         assert history[-1] < history[0]
 

tests/python/testing.py

@@ -272,6 +272,8 @@ def eval_error_metric(predt, dtrain: xgb.DMatrix):
     label = dtrain.get_label()
     r = np.zeros(predt.shape)
     gt = predt > 0.5
+    if predt.size == 0:
+        return "CustomErr", 0
     r[gt] = 1 - label[gt]
     le = predt <= 0.5
     r[le] = label[le]

tests/travis/run_test.sh

@@ -1,7 +1,5 @@
 #!/bin/bash
 
-make -f dmlc-core/scripts/packages.mk lz4
-
 source $HOME/miniconda/bin/activate
 
 if [ ${TASK} == "python_sdist_test" ]; then

tests/travis/setup.sh

@@ -1,5 +1,9 @@
 #!/bin/bash
 
+# https://travis-ci.community/t/macos-build-fails-because-of-homebrew-bundle-unknown-command/7296/27
+brew install cmake libomp lz4
+
+
 if [ ${TASK} == "python_test" ] || [ ${TASK} == "python_sdist_test" ]; then
     if [ ${TRAVIS_OS_NAME} == "osx" ]; then
         wget --no-verbose -O conda.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-x86_64.sh