How to use categorical features?

[schachem]

I would be very thankful if you could provide a small example on how to use the categorical features parameter during fitting?
I am trying to put a string denoting the column of the dataframe, but that is not working.

Let's say we have dataframe with a column for protocol and we already encoded them to numerical values.
Now when calling fit, I would try it like clf.fit(train_df, train_target, categorical_columns=['protocol'], base_class=0)

However it tells me: ValueError: Specifying the columns using strings is only supported for pandas DataFrames
But I did pass a dataframe, i.e. train_df to the fit function.
The train_df dataframe does contain multiple columns (including the protocol column, where TCP and UDP are encoded to 30 and 31; arbitrary values).

[munak98]

Hi @schachem . Thanks for opening the issue, I hadn't noticed this problem before!

Indeed, the method is not accepting columns names. For now, categorical features must be passed as indexes.
This happens because inside the fit method there is an initial step of validation of the training set that transforms it from a pandas DataFrame to an array. So when it gets to the preprocessing step, the data is not a pandas DataFrame anymore and the preprocessor throws this error.

I'll think about whether to change this or not, because this initial validation is done by a standard scikit-learn function and I don't want to lose compatibility. For now, using the column indexes should work.