ENH: Implement Panel pct_change

[TomAugspurger]

Closes #6904

There's just a bit of extra index handling that needs to be done before moving on to generic.pct_change(). I had to adjust that to use the .div and .sub ops instead of / and - to work with panels.

I wasn't sure why axis wasn't included as an actual names keyword arg. generic just looks for it in **kwargs. I did the same in panel.

A related issue was the Panel.shift() has a different argument signature than generic.shift(). I can make those consistent and put in a deprecation warning in this issue or in a new one.

[TomAugspurger]

@jreback any idea why this isn't allowed? wp.shift(0) doesn't raise, and seems to produce the correct answer.

[jreback]

their is a perf issue currently because of the recent change of panel.shift to generic.shift

see #6826

I think wp.shift(0) is technically ok (though not intuitive)

prob should have a panel.shift that enforces kwargs (and doesn't have positional args at all)

you can use some of the machinery in reindex to detect the correct signatures

[TomAugspurger]

Ok. I'm going to hold off on touching Panel.shift() until #6826 is settled.

[TomAugspurger]

I added a vbench, but maybe it isn't worth it since really all that's happing is a panel.shift() and a panel.div(). Should I take it out?

[jreback]

it's fine

[jreback]

instead of specifying the items axis use info_axis for ndim > 2

maybe this should also be explict for ndim > 2
iow you must specify an axis ?
not sure if we do that in other functions

[TomAugspurger]

I changed the default axis to major to be more consistent with panel.shift(). Wasn't sure what you meant about making it explicit. Do you mean adding the axis as a real kwarg?

[jreback]

axis should default to stat_axis (which is 1 for a panel)

[TomAugspurger]

I finally understand what you meant about the axes (I've never really understood what info_axis and stat_axis were before now). This let me move everything to generic. Should be good now. Haven't tested with Panel4D and up, but in principle there's no reason it shouldn't work.

[jreback]

gr8 - generic is good! makes maint easier

u can simply put in the Panel4D notimementrd list (as this is prob not intuitive), look in panelnd for a list of functions that it disables

[TomAugspurger]

Actually this will already raise NotImplemented since shift is on the NotImplemented list, and that's used in pct_change.

[jreback]

ok gr8