enable_categorical=True results in error code 0xc0000409

[ernesto-dreier]

Fitting a XGBRegressor with enable_categorical=True on a Windows 10 machine results in an error. See below for a minimal example:

from random import choice
from string import ascii_lowercase

import pandas as pd
from xgboost.sklearn import XGBRegressor


n = 5
n_cat  = 100


X = pd.Series(
    [''.join(choice(ascii_lowercase) for _ in range(3)) for _ in range(n_cat)],
    dtype="category"
)[:n].to_frame()
y = pd.Series(range(n))

model = XGBRegressor(
    enable_categorical=True,
    tree_method="approx",
)

model.fit(X=X, y=y)

Error:
Process finished with exit code -1073740791 (0xC0000409)

conda list:

# Name                    Version                   Build  Channel
ca-certificates           2021.10.8            h5b45459_0    conda-forge
joblib                    1.1.0                    pypi_0    pypi
numpy                     1.22.3                   pypi_0    pypi
openssl                   1.1.1n               h8ffe710_0    conda-forge
pandas                    1.4.2                    pypi_0    pypi
pip                       22.0.4             pyhd8ed1ab_0    conda-forge
python                    3.9.1           h7840368_5_cpython    conda-forge
python-dateutil           2.8.2                    pypi_0    pypi
python_abi                3.9                      2_cp39    conda-forge
pytz                      2022.1                   pypi_0    pypi
scikit-learn              1.0.2                    pypi_0    pypi
scipy                     1.8.0                    pypi_0    pypi
setuptools                62.1.0           py39hcbf5309_0    conda-forge
six                       1.16.0                   pypi_0    pypi
sklearn                   0.0                      pypi_0    pypi
sqlite                    3.38.3               h8ffe710_0    conda-forge
threadpoolctl             3.1.0                    pypi_0    pypi
tzdata                    2022a                h191b570_0    conda-forge
ucrt                      10.0.20348.0         h57928b3_0    conda-forge
vc                        14.2                 hb210afc_6    conda-forge
vs2015_runtime            14.29.30037          h902a5da_6    conda-forge
wheel                     0.37.1             pyhd8ed1ab_0    conda-forge
xgboost                   1.6.0                    pypi_0    pypi

[trivialfis]

Emm .. we calculate the number of categories as the number of discrete values, which couldn't handle the case when the number of categories is greater than the total number of entries. Should find a better way to handle this.

[ernesto-dreier]

Thanks for the incredibly quick fix.