Setting type to category

[MagBuchSB1]

Hi

In many cases I need to set the data type of object-columns to "category", like so;

df[<object_cols>] = df[<object_cols>].astype("category")

How would I go about doing this in pdpipe? I've tried different approaches using ColumnDtypeEnforcer, but to no avail...

[shaypal5]

OK. ColumnDtypeEnforcer should actually support this.

For a static column list, this works fine (at least for me), assuming 'a' and 'b' are the labels of you object columns:

dt_enforcer = pdp. ColumnDtypeEnforcer({'a': 'category', 'b': 'category'})
post_df = dt_enforcer(df)

If you want the pipeline stage to convert all object columns to "category", to be determined on runtime, do this:

dt_enforcer = pdp. ColumnDtypeEnforcer({pdp.cq.OfDtypes('object'): 'category'})
post_df = dt_enforcer(df)

This also works fine for me locally. Notice that if you use the same pipeline object several times, and you do not want this column list to have a fit/predict logic (meaning, to be determined once on fit time and retain this list on transform), use fittable=False when initializing the column qualifier object, like so:

dt_enforcer = pdp. ColumnDtypeEnforcer({pdp.cq.OfDtypes('object', fittable=False): 'category'})
post_df = dt_enforcer(df)

That's it. Let me know if this does not answer your need.
Cheers!