Future of colwise

[bkamins]

Following #1590 (comment) I would like to discuss here what we should do with colwise function.

Now its basic form colwise(f, df) is the same as map(f, columns(df)), but we have also two additional methods:

• one producing a matrix when multiple functions are passed as f argument (works only for a data frame);
• one producing an array of arrays when a single function is applied to a grouped data frame.

How useful do you think those are and is is worth to keep colwise defined or should we deprecate it.

[nalimilan]

I'm not sure these two methods are very useful. Since they return arrays, they don't include any information about the names of applied functions or about the groups. Anyway this behavior is easy to get using array comprehensions.

Typically, when applying a function on each group, you want the grouping column to be able to interpret the result, so you'd use map(df -> map(f, columns(df)), gd), or map(gd) do f.(columns(df)) end. It could be useful to provide a convenience function to do that, maybe via map(f, columns(gd)) and f.(columns(gd))? The similarity with the DataFrame syntax is appealing.

[bkamins]

I would deprecate colwise then. OK?

Users can use map or mapcols. I guess in the long run when @nalimilan redesigns GroupedDataFrame API some additional convenience methods for this type can be implemented.

[nalimilan]

Yes, let's deprecate it!