Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix external memory with gpu_hist and subsampling combination bug. (#7481)

Browse Source 
Instead of accessing data from the `original_page_`, access the data from the first page of the available batch.

fix #7476

Co-authored-by: jiamingy <jm.yuan@outlook.com>
This commit is contained in:
Ginko Balboa
2021-12-24 04:15:35 +01:00
committed by GitHub
parent 7f399eac8b
commit 29bfa94bb6
4 changed files with 61 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
tests/python-gpu/test_gpu_data_iterator.py
View File

@@ -17,16 +17,23 @@ def test_gpu_single_batch() -> None:
@pytest.mark.skipif(**no_cupy())
@given(
strategies.integers(0, 1024), strategies.integers(1, 7), strategies.integers(0, 13)
strategies.integers(0, 1024),
strategies.integers(1, 7),
strategies.integers(0, 13),
strategies.booleans(),
)
@settings(deadline=None)
def test_gpu_data_iterator(
n_samples_per_batch: int, n_features: int, n_batches: int
n_samples_per_batch: int, n_features: int, n_batches: int, subsample: bool
) -> None:
run_data_iterator(n_samples_per_batch, n_features, n_batches, "gpu_hist", True)
run_data_iterator(n_samples_per_batch, n_features, n_batches, "gpu_hist", False)
run_data_iterator(
n_samples_per_batch, n_features, n_batches, "gpu_hist", subsample, True
)
run_data_iterator(
n_samples_per_batch, n_features, n_batches, "gpu_hist", subsample, False
)
def test_cpu_data_iterator() -> None:
"""Make sure CPU algorithm can handle GPU inputs"""
run_data_iterator(1024, 2, 3, "approx", True)
run_data_iterator(1024, 2, 3, "approx", False, True)