Misc fixes 05

[josef-pkt]

a few fixes

closes #109 TestProbitCG failure on Ubuntu
closes #639 test folders not on python path ( #639 not in commit message)
closes #565 omni_normtest doesn't propagate the axis argument
closes #563 error in doc generation for AR.fit
closes #552 influence outliers breaks when regressing on constant

plus refactoring victims in ex_regression_plots

[josef-pkt]

I'm done for now, will merge tomorrow.

[dengemann]

Hi folks, I just noticed that some other things seem broken as well. Among those the interaction_plot examples (apparently due to py3k compatibility measures) but also the tsa arima example. Shall I open a separate issue for those?

[josef-pkt]

Denis, yes, please open an issue, It's better to have a record when examples for the official documentation break.

We haven't started to convert the examples for py3 compatibility yet, so I have no idea yet why these examples are broken.

[dengemann]

Ok, so let's then open separate examples for each of them. I'll look into it later.

[dengemann]

... Sorry, the arima / py3k was a false alarm. I'm used to running stuff using ipython -- pylab. Invoking plt.show() instead of show() works flawlessly. I'm not sure whether this difference is worth an issue. Might be good to add a note somewhere that some examples depend on pyplot rather than pylab.

[josef-pkt]

github ate our comments :(

I force pushed after a rebase, and some comments disappeared, at least I don't see them anymore.

merging

[josef-pkt]

9e6194d is the change that removes the super docs in fit method, that was overwriting the docstring of the specific models.fit methods

see #563

main discussion:
Josef: links are enough and we don't need to present the same information for each model.
Skipper: in the interpreter it's more useful to have the full docstring with all options.

[josef-pkt]

looks like the website didn't update, e.g
http://statsmodels.sourceforge.net/devel/generated/statsmodels.tsa.ar_model.AR.fit.html#statsmodels.tsa.ar_model.AR.fit

[jseabold]

I see that it didn't update, but I also don't see what the problem with having this information is.

[josef-pkt]

my problem is information overflow:

doc strings are too long
I want to know what the specific information is, and a toc/links to auxilliary generic information.

extreme case scipy.stats.distribution: I never read any docstrings, because it's 95% templated generic information and it's difficult to find the distribution specific information.

the help in the interpreter (which I almost never use) has several pages of docstring.
In the html and htmlhelp the links work very well.
I was thinking that we could add a full path to the object with the generic docstring in the see also, so that when we use the interpreter, then we can just call a second help with the copied object path

?generic.fit

[jseabold]

The scipy.stats.distribution documentation and the code in general is notoriously difficult and not really relevant here. I've given relevant examples for cases when having hidden extra (common) information is extremely helpful (matplotlib, pandas).

I do use the interpreter. Exclusively. I also read documentation. That's what it's there for. I like having the help there, and I know others do to -- likely more than who use htmlhelp or html. The information is at the bottom. It's noted in the docstring above where it is and clearly what it is. Length has nothing to do with clarity. The special case here is the html. If you want to write a template for the html/htmlhelp that scrapes this information then do that. This is possible.

https://raw.github.com/statsmodels/statsmodels/master/docs/source/_templates/autosummary/glmfamilies.rst

I don't want to import something else to see the docstring. It's the reason I added this information in the first place. It's a pain at best when you know the code works with inheritance and it's confusing at worst if you have no idea.