BUG: fix Series.plot label setting

[TomAugspurger]

Closes #10119, a regression in 0.16.1

Basically Series.plot(label='foo') isn't actually setting the label.

I want to look at one more thing. With this fix you'd need

s.plot(label='foo', legend=True

to get the legend to actually show up. I can't remember what the old behavior was. At least for seaborn, setting a label implies that the legend should be drawn (e.g. sns.kdeplot(s, label='foo'))

[sinhrks]

I can't remember what the old behavior was.

Because of the default value of plot_series, no legend should appear otherwise specified in prior versions.

https://github.com/pydata/pandas/blob/master/pandas/tools/plotting.py#L2492

[TomAugspurger]

That's what I thought. Ok, I suppose we'll keep it like that for now. So ts.plot(label='series') will attach the legend information, but not actually draw the legend.

[jorisvandenbossche]

Can you add here a test for the case where the series has a name, but you provide a label (so add ´s.plot(label='LABEL', legend=True)´ and then check the legend label is LABEL and not NAME)?

[TomAugspurger]

Good call. Updated.

[TomAugspurger]

I think this is ready then. Will merge once Travis is green.

[shoyer]

Thanks @TomAugspurger !

[schmohlio]

@TomAugspurger Thanks, Tom.

I'm trying to tie back to my commit - #9574 . I don't have that .pop() in _compute_plot_data() within my commit. Was that added somewhere else? In fact I deleted a .pop() line in my commit.

[jorisvandenbossche]

@schmohlio This specific line that @TomAugspurger changed in this PR was in any case not introduced by your PR (https://github.com/pydata/pandas/blame/95dc0506f1a9237b0a34a1d26eba1e1a8d357758/pandas/tools/plotting.py#L1001), so certainly not necessary a regression by your PR! (and anyway, who introduced it does not really matter, as it is reviewed by multiple people and we should improve the test suite to catch such things)
But it was a case we could have thought of to test as well .. (easy to say afterwards of course), we only tested dataframes and not series I think.

[schmohlio]

@jorisvandenbossche both thoughts makes a lot of sense. Just wanted to see if I could help at any point. Thanks again for you and your team's great work.