ENH/VIS: Dataframe bar plot can now accept width and pos keywords

[sinhrks]

Following keywords are now acceptable for dataframe.plot(kind='bar') and dataframe.plot(kind='barh').
width: Specify the bar width. In previous versions, static value 0.5 was passed to matplotlib and it cannot be overwritten.
pos: Specify relative alignments for bar plot layout. From 0 (left/bottom-end) to 1(right/top-end). Default is 0.5 (center).

Examples:

[TomAugspurger]

Thanks. We'll need an entry under the API section in release.rst. We have this PR for error bars: #5638. Need to check if these play well together. I'll take a look early next week.

[jreback]

@TomAugspurger iirc some convention for simply passing thru keywords on plot directly to matplotlob
isn't that right?

[sinhrks]

Thank you for your confirmation. I've added release note description. I'll recheck once #5638 is merged to master.

[TomAugspurger]

@jreback We do some stuff on the pandas side before passing things through to matplotlib's ax.bar(). Right now if you pass a width= in the keyword args you get a TypeError with multiple values for width.

@sinhrks What do you think about position instead of pos? position seems to be more common in the matplotlib code. Also could you get Travis-CI hooked up so that the tests get run?

[sinhrks]

OK, agreed. I've changed the keyword to position and hooked to Travis-CI.

Also, I found some bar plot related dataframe tests are under TestSeriesPlot class, thus moved it to TestDataFramePlots class.

[sinhrks]

Travis test should have passed, but no icon seems to be displayed. Is it OK or any other configuration required?
https://travis-ci.org/sinhrks/pandas/builds/20881935

[jreback]

pls add to v0.14.0.txt as well

[sinhrks]

Thanks. Got it.

[jreback]

pls rebase and then squash as well

[sinhrks]

Yes, did it.

[TomAugspurger]

Thanks, I added your note about the position argument to the plot* docstrings. Width didn't need one since it's a regular matplotlib bar argument.

[jreback]

@TomAugspurger did you mean to close?

[TomAugspurger]

Hmm. I merged via command line since I had to move the release note down a line. Did I break something?

[TomAugspurger]

It is merged. 8d87a29

[jreback]

@TomAugspurger I see..no problem...git closes the issue automatically, but I see you just put a reference (which is correct!)

@sinhrks thanks!

[TomAugspurger]

Thanks

[TomAugspurger]

@jreback I see that this failed on Travis: https://travis-ci.org/pydata/pandas/jobs/21098475

Looking into it.

[TomAugspurger]

Hmm strange. Here's the failing test:

kinds = 'line', 'bar', 'barh', 'kde', 'density'
df = DataFrame(randn(10, 2), dtype=object)
df[np.random.rand(df.shape[0]) > 0.5] = 'a'
for kind in kinds:
    with tm.assertRaises(TypeError):
        df.plot(kind=kind)

I can't reliably reproduce the failure. What does reproduce it is if I don't do the df[np.random.rand(df.shape[0]) > 0.5] = 'a' line. In that case (all numeric data, just objects) we automatically convert types and plot.

Do you think it could have randomly drawn 10 values all greater than .5? The probability is .5 ** 10 = 0.0009765625, not impossible...

I'm going to set a seed and see if it fixes.

[TomAugspurger]

See #6670 for a hopeful fix on the last comment.