Hi @solegalli,

This looks interesting!

Are you envisioning that OHE allows the following functionality?

• init has a param called variables_w_new_category that represents which variables may contain the new category(ies).
• new_categories - a list of categories that should be encoded in the variables included in the variables_w_new_category
• the new_categories values would be added as values to the respective keys listed in variables_w_new_category in self.encoder_dict_

@solegalli, should this task be closed? It seems like task #403 resolved this issue

#403 is in essence asking for the same functionality. That's probably why I closed it. I flagged it as duped now. Still open.

@solegalli, resurrecting this issue ;)

When someone selects this functionality, do we want to limit the user to one variable?

I imagine that the user will select values that are specific to one variable. It seems odd for multiple categorical variables to have the same values.

I think the most straight forward would be to add a new parameter, or perhaps even better, extend top_categories to take a dictionary with the variable as key and the categories to encode as values. Then, for each variable, the transformer will create dummies only for the categories indicated by the user.

Will you pick this one up?

@solegalli,

I like the idea of using a dictionary. However, I'm unsure if the dictionary should be accepted by top_categories.

Would it be cleaner to have a separate param called custom_categories? We would check that both top_categories and custom_categories do not have values.

Sounds good to me!