Merge branch 'nccl-dev' of https://github.com/Microsoft/LightGBM into nccl-dev

Author: shiyu1994

.vsts-ci.yml

@@ -254,6 +254,9 @@ jobs:
       - script: |
           git clean -d -f -x
         displayName: 'Clean source directory'
+      # LGBM_SKIP_DASK_TESTS=true is set below only because running the tests under emulation is so slow...
+      # in theory, 'lightgbm.dask' should work without issue on aarch64 Linux systems.
+      # That could probably be removed as part of https://github.com/microsoft/LightGBM/issues/6788
       - script: |
           cat > docker-script.sh <<EOF
           export CONDA=\$HOME/miniforge
@@ -273,6 +276,7 @@ jobs:
             --env BUILD_ARTIFACTSTAGINGDIRECTORY=$BUILD_ARTIFACTSTAGINGDIRECTORY \
             --env BUILD_DIRECTORY=$BUILD_DIRECTORY \
             --env COMPILER=$COMPILER \
+            --env LGBM_SKIP_DASK_TESTS=true \
             --env METHOD=$METHOD \
             --env OS_NAME=linux \
             --env PRODUCES_ARTIFACTS=$PRODUCES_ARTIFACTS \

R-package/R/lightgbm.R

@@ -69,6 +69,9 @@
 #'          in \code{params}, that metric will be considered the "first" one. If you omit \code{metric},
 #'          a default metric will be used based on your choice for the parameter \code{obj} (keyword argument)
 #'          or \code{objective} (passed into \code{params}).
+#'
+#'          \bold{NOTE:} if using \code{boosting_type="dart"}, any early stopping configuration will be ignored
+#'          and early stopping will not be performed.
 #' @section Model serialization:
 #'
 #'          LightGBM model objects can be serialized and de-serialized through functions such as \code{save}

R-package/man/lgb.cv.Rd

@@ -64,7 +64,7 @@ LightGBM's Python-package supports distributed learning via `Dask`_. This integr
 
 .. warning::
 
-    Dask integration is only tested on Linux.
+    Dask integration is only tested on macOS and Linux.
 
 Dask Examples
 '''''''''''''

R-package/man/lgb.train.Rd

@@ -468,6 +468,11 @@ def early_stopping(
     To check only the first metric set ``first_metric_only`` to True.
     The index of iteration that has the best performance will be saved in the ``best_iteration`` attribute of a model.
 
+    .. note::
+
+        If using ``boosting_type="dart"``, this callback has no effect and early stopping
+        will not be performed.
+
     Parameters
     ----------
     stopping_rounds : int

R-package/man/lgb_shared_params.Rd

@@ -5,7 +5,6 @@
 import socket
 from itertools import groupby
 from os import getenv
-from platform import machine
 from sys import platform
 from urllib.parse import urlparse
 
@@ -16,10 +15,8 @@
 
 from .utils import sklearn_multiclass_custom_objective
 
-if not platform.startswith("linux"):
-    pytest.skip("lightgbm.dask is currently supported in Linux environments", allow_module_level=True)
-if machine() != "x86_64":
-    pytest.skip("lightgbm.dask tests are currently skipped on some architectures like arm64", allow_module_level=True)
+if platform in {"cygwin", "win32"}:
+    pytest.skip("lightgbm.dask is not currently supported on Windows", allow_module_level=True)
 if not lgb.compat.DASK_INSTALLED:
     pytest.skip("Dask is not installed", allow_module_level=True)
 
@@ -58,6 +55,10 @@
     pytest.mark.skipif(getenv("TASK", "") == "mpi", reason="Fails to run with MPI interface"),
     pytest.mark.skipif(getenv("TASK", "") == "gpu", reason="Fails to run with GPU interface"),
     pytest.mark.skipif(getenv("TASK", "") == "cuda", reason="Fails to run with CUDA interface"),
+    pytest.mark.skipif(
+        getenv("LGBM_SKIP_DASK_TESTS", "") == "true",
+        reason="Skipping lightgbm.dask tests (found env variable LGBM_SKIP_DASK_TESTS=true)",
+    ),
 ]