read_csv arguments: can we have skipcols and userows?

[pseth]

Is there a reason why read_csv has a usecols and skiprows as arguments, but not skipcols and userows? Is this to avoid parameter checks or something more fundamental than that?

It would be nice to have all four options to avoid clunky inversions of the type usecols = columns.remove(unwanted_col).

[jreback]

this is essentially a duplicate of the now closed: #10882

usecols accepts a callable, to allow arbitrary evaluation of which columns to use.

in a similar vein, skiprows accepts a callable as well.

The defaults make the most sense here, e.g. generally you want to keep columns (out of a larger set) and skip (a small subset of rows).

I am not anti the counter parts, but this is just one more keyword and added complexity.

[gfyoung]

Yep, I'm in agreement with @jreback here. Especially since we can accept callables for both inputs you can emulate skipcols and userows as follows:

skipcols = [...]
userows = [...]
read_csv(..., usecols=lambda x: x not in skipcols,
              skiprows=lambda x: x not in userows])

I think this should resolve your concern about "clunkiness" as you put it, so if there are no other concerns, I think this is safe to close.

[pseth]

@gfyoung Ah, I did not realise that was possible, the online documentation for read_csv doesn't seem to be up to date. That is indeed a more elegant solution over possibly conflicting arguments.

[jreback]

This feature is in 0.20.0 which is not released yet, docs are in the dev-docs: http://pandas-docs.github.io/pandas-docs-travis/generated/pandas.read_csv.html?highlight=read_csv#pandas.read_csv