Why does categorical!(df::DataFrame, ...) exist?

[matthieugomez]

I don't understand why categorical(df, cols) exists.

If it is about applying a function (categorical) to multiple columns of a dataframes, there should be a way to do it for any function, not just categorical (I guess that's what transform is about?).

I'm not following the package closely, so sorry if this was discussed somewhere.

[bkamins]

Thank you for raising this point (we have a similar case with allowmissing and disallowmissing). Simply in the past our API was much less flexible.

It has one minor functionality that:

1. by default it converts string column
2. it can take type as a column selector

But apart from this it is true that roughly (I am dropping some corner cases):

@deprecate categorical(df, cols) transform(df, cols .=> categorical .=> cols)

and

@deprecate categorical!(df, cols) transform!(df, cols .=> categorical .=> cols)

Having said this, I do not see a huge problem to keep categorical and categorical! as convenience functions. We define such functions for most common transformations, so that users have an easier life. @nalimilan - what do you think?

[nalimilan]

Yes I think it makes sense to have some convenience functions. That's the clearest for allowmissing as it's a very common need, but categorical is already exported since it exists in CategoricalArrays, so adding a new method isn't very costly.

For other functions, mapcols can also be used instead of transform. I guess we could add mapcols! for in-place operation.

[matthieugomez]

Thanks for the explanation! I think it'd be better to have only one way to do things, but that's your choice! I appreciate all the improvement happening these days, by the way. Thanks for your work!