[MRG] Fix aesthetic example roc crossval - #8232

@jnothman I slightly change the example to get similar plot as in SO.

If we're going to show the AUC for each fold in the legend, we need to mark each distinctively, still.

Can you add a legend entry for the standard deviation?

Can you also add an image of the new plot?
It is possible that upgrading to Matplotlib 2.0 changes drastically the looks of this plot.

FWIW, @NelleV: you can easily get to the CircleCI-rendered docs using this bookmarklet (although it seems GitHub keeps tweaking their HTML so I need to keep tweaking that script)

@jnothman we really need this for Matplotlib...

Does matplotlib have CI rendering docs and pushing to a public server as Circle CI does? Otherwise, the bookmarklet uses our "likely changed docs page" compiled by this code and incidentally exploits our web server having a URL shorthand for built docs

We build our documentation on travis-ci, but this might be a good opportunity to see what we can do with circle-ci.

@jnothman IMHO, I would keep only the mean/std AUC in the legend with the luck, implying that we would need a single colour. I made the changes such that you can compare with the SO aesthetic and make a choice.

@NelleV I am not sure if this is what you mean by adding images?

Matplotlib 1.5.3

Matplotlib 2.0.0

@jnothman Now that I am seeing the legend with matplotlib 2.0, I would probably keep the individual ROC.

When I said put the std in the legend, I meant show that that's what a grey mass means. Or else it can be noted in the example description.

Otherwise LGTM.

looks great.

thx @glemaitre