ENH: drop function now has errors keyword for non-existing column handling

[sinhrks]

Closes #5300.

Currently drop raises ValueError when non-existing label is passed. I think it is useful if drop has an option to suppress error and drop existing labels only.

For example, I sometimes process lots of files which has slightly different columns, and want to drop if data has unnecessary columns. Previously, I have to prepare different drop arguments checking columns existing each data.

[jreback]

this could be added on after #6599
@hayd

[jtratner]

let's give at least one of these an errors='raise' argument (and add it to an existing test that doesn't raise) just for completeness.

[jtratner]

This all looks fine to me - thanks for the patch.

We might want to reduce the number of test cases (just because we don't necessarily need to duplicate), but aside from that, quite good. - somebody else want to take a look too? I may be a bit rusty :P

[hayd]

Looks all good to me. But let's wait for #6599. :)

[jreback]

yep....so @hayd will revisit after that #6599 merge

[jreback]

@sinhrks this idea is ok, can you rebase

[sinhrks]

@jreback Yes, rebased.

[jreback]

hmm, shouldn't this raise a KeyError? @jorisvandenbossche @shoyer to be consistent with say .loc and other indexers? @hayd

[shoyer]

This is kind of a gray area to me. We're not doing actual indexing here. I would probably stick with ValueError, especially to remain consistent with the existing implementation.

[jreback]

have to think about this

[jreback]

@sinhrks can you rebase this.

[sinhrks]

Sure, rebased.

[jreback]

@sinhrks thanks!