xgboost 3.2.0 crashes with cudf 26.02 when there are categorical features #12138
Description
Latest stable version of xgboost does not work with cudf 26.02 when there are categorical features. It is fine with cudf 25.12.
Code to replicate:
import numpy as np
import cudf, xgboost
from xgboost import XGBRegressor
print(F'{cudf.__version__=}')
print(F'{xgboost.__version__=}')
rng = np.random.default_rng(0)
X = cudf.DataFrame(rng.random((100,10))).astype(str).astype('category')
y = cudf.Series(rng.random(100))
model = XGBRegressor(enable_categorical=True, device='cuda').fit(X, y)Expected output:
cudf.__version__='26.02.01'
xgboost.__version__='3.2.0'
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
/tmp/ipykernel_55/3028577614.py in <cell line: 0>()
10 y = cudf.Series(rng.random(100))
11
---> 12 model = XGBRegressor(enable_categorical=True, device='cuda').fit(X, y)
/usr/local/lib/python3.12/dist-packages/xgboost/core.py in inner_f(*args, **kwargs)
749 for k, arg in zip(sig.parameters, args):
750 kwargs[k] = arg
--> 751 return func(**kwargs)
752
753 return inner_f
/usr/local/lib/python3.12/dist-packages/xgboost/sklearn.py in fit(self, X, y, sample_weight, base_margin, eval_set, verbose, xgb_model, sample_weight_eval_set, base_margin_eval_set, feature_weights)
1341
1342 evals_result: EvalsLog = {}
-> 1343 train_dmatrix, evals = _wrap_evaluation_matrices(
1344 missing=self.missing,
1345 X=X,
/usr/local/lib/python3.12/dist-packages/xgboost/sklearn.py in _wrap_evaluation_matrices(missing, X, y, group, qid, sample_weight, base_margin, feature_weights, eval_set, sample_weight_eval_set, base_margin_eval_set, eval_group, eval_qid, create_dmatrix, enable_categorical, feature_types)
698 """
699 # Feature_types contains the optional reference categories from the booster object.
--> 700 train_dmatrix = create_dmatrix(
701 data=X,
702 label=y,
/usr/local/lib/python3.12/dist-packages/xgboost/sklearn.py in _create_dmatrix(self, ref, **kwargs)
1260 except TypeError: # `QuantileDMatrix` supports lesser types than DMatrix
1261 pass
-> 1262 return DMatrix(**kwargs, nthread=self.n_jobs)
1263
1264 def _set_evaluation_result(self, evals_result: EvalsLog) -> None:
/usr/local/lib/python3.12/dist-packages/xgboost/core.py in inner_f(*args, **kwargs)
749 for k, arg in zip(sig.parameters, args):
750 kwargs[k] = arg
--> 751 return func(**kwargs)
752
753 return inner_f
/usr/local/lib/python3.12/dist-packages/xgboost/core.py in __init__(self, data, label, weight, base_margin, missing, silent, feature_names, feature_types, nthread, group, qid, label_lower_bound, label_upper_bound, feature_weights, enable_categorical, data_split_mode)
974 return
975
--> 976 handle, feature_names, feature_types = dispatch_data_backend(
977 data=data,
978 missing=self.missing,
/usr/local/lib/python3.12/dist-packages/xgboost/data.py in dispatch_data_backend(data, missing, threads, feature_names, feature_types, enable_categorical, data_split_mode)
1437 )
1438 if _is_cudf_df(data) or _is_cudf_ser(data):
-> 1439 return _from_cudf_df(
1440 data=data,
1441 missing=missing,
/usr/local/lib/python3.12/dist-packages/xgboost/data.py in _from_cudf_df(data, missing, nthread, feature_names, feature_types, enable_categorical)
1130 enable_categorical: bool,
1131 ) -> DispatchedDataBackendReturnType:
-> 1132 df, feature_names, feature_types = _transform_cudf_df(
1133 data, feature_names, feature_types, enable_categorical
1134 )
/usr/local/lib/python3.12/dist-packages/xgboost/data.py in _transform_cudf_df(data, feature_names, feature_types, enable_categorical)
1115
1116 return (
-> 1117 CudfTransformed(result, ref_categories=ref_categories),
1118 feature_names,
1119 feature_types,
/usr/local/lib/python3.12/dist-packages/xgboost/data.py in __init__(self, columns, ref_categories)
1026
1027 for col in self.columns:
-> 1028 push_series(col)
1029
1030 super().__init__(
/usr/local/lib/python3.12/dist-packages/xgboost/data.py in push_series(ser)
1017 if _is_df_cat(ser):
1018 cats, codes = ser.categories, ser.codes
-> 1019 cats_ainf, codes_ainf, buf = cudf_cat_inf(cats, codes)
1020 temporary_buffers.append(buf)
1021 aitfs.append((cats_ainf, codes_ainf))
/usr/local/lib/python3.12/dist-packages/xgboost/_data_utils.py in cudf_cat_inf(cats, codes)
596
597 # pylint: disable=protected-access
--> 598 arrow_col = cats._column.to_pylibcudf(mode="read")
599 # Tuple[types.CapsuleType, types.CapsuleType]
600 schema, array = arrow_col.__arrow_c_device_array__()
TypeError: ColumnBase.to_pylibcudf() got an unexpected keyword argument 'mode'
Metadata
Metadata
Assignees
Labels
No labels