I am using xgboost version 2.1.0
When converting a pandas dataframe containing category columns to a DMatrix using xgboost.DMatrix() with ‘enable_categorical’=True, all behaves as expected unless the dataframe is one returned by sklearn train_test_split(), despite the datatypes of all columns remaining category.

The following code produces expected behavior:

import pandas as pd 
import xgboost as xgb
import seaborn as sns

tips = sns.load_dataset('tips')

X, y = tips.drop('tip', axis=1), tips['tip']
print(X.dtypes)

# convert to DMatrix
dm = xgb.DMatrix(X, y, enable_categorical=True)
dm

---------------------------

total_bill     float64
sex           category
smoker        category
day           category
time          category
size             int64
dtype: object

<xgboost.core.DMatrix at 0x23153d7cf10>

The following code throws an error:

from sklearn.model_selection import train_test_split

X_train, y_train, X_test, y_test = train_test_split(X, y, random_state=1)
print(X_train.dtypes)

dtrain = xgb.DMatrix(X_train, y_train, enable_categorical=True)

-----------------------------

total_bill     float64
sex           category
smoker        category
day           category
time          category
size             int64
dtype: object

ValueError: DataFrame.dtypes for data must be int, float, bool or category. When categorical type is supplied, the experimental DMatrix parameter`enable_categorical` must be set to `True`.  Invalid columns:sex: category, smoker: category, day: category, time: category

I am confused.
Thank you.