Lightgbm: 'categorical_feature in param dict is overridden.'

[JYLFamily]

Why this warning？lgb.Dataset params is None.
Thank you.

Environment info

Operating System、CPU:Linux Server
C++/Python/R version:Python lightgbm 2.1.1

Error Message:

warnings.warn('categorical_feature in param dict is overridden.')

Reproducible examples

   def train(self, train_feature, train_label):
    train = self.__lgb.Dataset(
        data=train_feature,
        label=train_label,
        feature_name=train_feature.columns.tolist(),
        categorical_feature=self.__categorical_feature.tolist()
    )
    self.__clf = self.__lgb.train(
        self.__train_params,
        train,
        num_boost_round=self.__num_boost_round,
        categorical_feature=self.__categorical_feature.tolist()
    )

[StrikerRUS]

@JYLFamily This warning is raised because you're passing categorical features twice: when constuct Dataset and when call train.

The correct way is to pass them only while constructing Dataset via categorical_feature argument.

[soerendip]

This warning appears also when the categorical features are only passed to train and not specified anywhere else.

[StrikerRUS]

The correct way is to pass them only while constructing Dataset via categorical_feature argument.

import lightgbm as lgb

from sklearn.datasets import load_boston

X, y = load_boston(True)
lgb_data = lgb.Dataset(X, y, categorical_feature=[1, 2])
lgb.train({}, lgb_data)

D:\Miniconda3\lib\site-packages\lightgbm\basic.py:1247: UserWarning: Using categorical_feature in Dataset.
  warnings.warn('Using categorical_feature in Dataset.')

This warning is expected as per #792.

[guolinke]

@StrikerRUS
maybe we can discard that warning when the data source is not pandas? I think the auto mode only works for pandas.

[StrikerRUS]

@guolinke

I think the auto mode only works for pandas.

Yeah, I think so.

maybe we can discard that warning when the data source is not pandas?

OK, I'll take a look.

[elzizyel]

@StrikerRUS Warning still exsits.