[PYTHON] Scikit-learn DecisionTreeClassifier with datetime type values

I've been struggling for almost an hour to solve the error when running DecisionTreeClassifier.

TypeError                                 Traceback (most recent call last)
<ipython-input-42-dbd33597c073> in <module>()
      7 learner = DecisionTreeClassifier(random_state = 2)
----> 8 learner = learner.fit(features_train[:int(sample_size)], outcome_train[:int(sample_size)])
********* omitted **********
TypeError: float() argument must be a string or a number

Some people say that is because of NaN value so I checked but there were no NaN value. https://stackoverflow.com/questions/43709854/python-naive-bayes-fit-function-gives-typeerror-float-argument-must-be-a-str

print features_train.isnull().values.any()
 -> False

Is there anything wrong in the code lines where values are converted to datetime? But it seems right.

features_train['date_x'] = pd.to_datetime(features_train['date_x'], format='%Y-%m-%d')

Then, finally I found the link says you just need to datetime into categorical vaue. https://stackoverflow.com/questions/16453644/regression-with-date-variable-using-scikit-learn I just intentionally didn't convert datetime into categorical date when encoding the data set because I was not sure if it is mathematically right.

After including date columns in data encording process, DecisionTreeClassifier worked as expected and returned its prediction.

columns = ['activity_category', 'people_id', 'activity_id', 'date_x', 
****** omitted *****
          ]

#Set concoder
encoder = LabelEncoder()

#Encode data frame
encoded_people_act_train_df = people_act_train_df.copy()
for col in columns:
    encoded_people_act_train_df[col]=  encoder.fit_transform(people_act_train_df[col])