Error when handling categoricals with LightGBM about powershap HOT 2 CLOSED

Thank you for the issue. I have taken a look at your code and tested it and it appears this error is because of the argument requirements for LightGBM. According to the documentation of LightGBM for the categorical_feature documentation argument, the model only accepts categorical features if they use a int format when using Pandas.

Using the following snippet to convert all string categories to ints for preprocessing will solve your issue!

cats = ['Cabin', 'Embarked', 'Gender', 'Name', 'Parch', 'Pclass', 'SibSp', 'Ticket']
for col in cats:
    X[col] = X[col].factorize()[0]

The PowerShap API is used correctly in your snippet!

from powershap.

@JarneVerhaeghe thanks for looking into this. However, I may have an alternate explanation for the documentation on lightgbm. If the columns in the dataframe are of CategoricalDtype() and are nominal, one could use the name representation of the cat_features. e.g., simply running the following line verifies that:

lgb_cl = LGBMClassifier(random_state=42, n_estimators=10, cat_features=cats)
lgb_cl.fit(X, y)

from powershap.