Hi @orbitfold I won't accept this PR until we decide how to manage plotting code of this variety between statsmodels (e.g. the more mature qqplot in https://github.com/statsmodels/statsmodels/blob/master/statsmodels/graphics/gofplots.py) and pandas. Can you please engage the statsmodels developers on the mailing list to get the conversation started?

Hi @orbitfold just pinging you about this. Can you e-mail the statsmodels mailing list to bring up the overlap in functionality so we can make some decisions about how to proceed?

Done.

<CC'ing myself>

Would be good to finish the conversation on the statsmodels mailing list.

Bump

I've updated the function as per suggestions on the mailing list.

Is there something else I can/should do about this?

I had expected that discussion to end with a conclusion, preferably one a bit broader than just for this function. Something like:

• it is OK to duplicate some plotting functions between pandas and statsmodels, general idea is to use the same API where possible; keeping the one in pandas basic and the ones in statsmodels with more options.
• all the plot functions in pandas should return a figure like in statsmodels. It's a bit mixed now, sometimes an axis (or list of axes?), sometimes a figure.

I thought the conversation was concluded on the mailing list with your and @josef-pkt's stamp of approval "[pystatsmodels] probability plot"?

Your points seem to make sense to me.

Some worries that I'm not really going to champion since I first brought them up months ago. It's going to cause some confusion for users, and I fear there will inevitably be PRs for pandas to enhance functionality for things like qqplot into areas we already cover. Ideally, I would think pandas would be used for visualization of data - not things that require estimation and inference - but, alas, it doesn't make sense to fight plots going into pandas and there's really not a clear cut line except that pandas can't have a reverse dependence on statsmodels.

@jseabold yes for this function, so the PR looks fine as is. However, what didn't get finished is (quoting Wes) "decide how to manage plotting code of this variety between statsmodels ... and pandas".

Besides cross-referencing in the See Also sections, I don't see much that can be done to avoid duplicate effort.

Oh, sure. I agree with your points then. Always return figs (this was from JDH advice). Also using intersphinx and See Also sections in the docs should hopefully help. (We just need to watch as new plotting functionality is added.)

@orbitfold closing this, if you want to restart this, pls reopen/do a new PR (as master has changed a lot)..thanks