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

[backport] Fix arrow compatibility, hypothesis tests. (#7979)

Browse Source
This commit is contained in:
Jiaming Yuan 2022-06-07 01:47:45 +08:00 committed by GitHub
parent eefa1ddd8a
commit 645855e8b1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 3 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
python-package/xgboost/data.py
View File

@ -506,7 +506,6 @@ def record_batch_data_iter(data_iter: Iterator) -> Callable:
"""Data iterator used to ingest Arrow columnar record batches. We are not using """Data iterator used to ingest Arrow columnar record batches. We are not using
class DataIter because it is only intended for building Device DMatrix and external class DataIter because it is only intended for building Device DMatrix and external
memory DMatrix. memory DMatrix.
""" """
from pyarrow.cffi import ffi from pyarrow.cffi import ffi
@ -557,13 +556,7 @@ def _from_arrow(
if enable_categorical: if enable_categorical:
raise ValueError("categorical data in arrow is not supported yet.") raise ValueError("categorical data in arrow is not supported yet.")
major, _, _ = version.StrictVersion(pa.__version__).version
if major == 4:
rb_iter = iter(data.to_batches()) rb_iter = iter(data.to_batches())
else:
# use_async=True to workaround pyarrow 6.0.1 hang,
# see Modin-3982 and ARROW-15362
rb_iter = iter(data.to_batches(use_async=True))
it = record_batch_data_iter(rb_iter) it = record_batch_data_iter(rb_iter)
next_callback = ctypes.CFUNCTYPE(ctypes.c_int, ctypes.c_void_p)(it) next_callback = ctypes.CFUNCTYPE(ctypes.c_int, ctypes.c_void_p)(it)
handle = ctypes.c_void_p() handle = ctypes.c_void_p()

4
tests/python/test_updaters.py
View File

@ -10,10 +10,10 @@ exact_parameter_strategy = strategies.fixed_dictionaries({
'nthread': strategies.integers(1, 4), 'nthread': strategies.integers(1, 4),
'max_depth': strategies.integers(1, 11), 'max_depth': strategies.integers(1, 11),
'min_child_weight': strategies.floats(0.5, 2.0), 'min_child_weight': strategies.floats(0.5, 2.0),
'alpha': strategies.floats(0.0, 2.0), 'alpha': strategies.floats(1e-5, 2.0),
'lambda': strategies.floats(1e-5, 2.0), 'lambda': strategies.floats(1e-5, 2.0),
'eta': strategies.floats(0.01, 0.5), 'eta': strategies.floats(0.01, 0.5),
'gamma': strategies.floats(0.0, 2.0), 'gamma': strategies.floats(1e-5, 2.0),
'seed': strategies.integers(0, 10), 'seed': strategies.integers(0, 10),
# We cannot enable subsampling as the training loss can increase # We cannot enable subsampling as the training loss can increase
# 'subsample': strategies.floats(0.5, 1.0), # 'subsample': strategies.floats(0.5, 1.0),