Fix for DataFrame.hist() with by- and weights-keyword

[Twizzledrizzle]

closes #9540

for example:

import pandas as pd
d = {'one' : ['A', 'A', 'B', 'C'],
     'two' : [4., 3., 2., 1.],
     'three' : [10., 8., 5., 7.]}     
df = pd.DataFrame(d)
df.hist('two', by='one', weights='three', bins=range(0, 10))

does not seem to break anything, but this is my first meddling in the pandas library, so a review would be nice

[TomAugspurger]

I suspect this will fail if the missing values in the weights column don't align perfectly with the missing values in the group column. It might be cleaner to refactor this to drop rows missing either group or weight earlier on, so that plot_group only has to deal with valid observations.

[Twizzledrizzle]

I added this because on the next line, group drops na values as well?
ax.hist(group.dropna().values

[Twizzledrizzle]

ah, issues when na is different on weights and group arrays, did not think of this, will think a little bit more

[TomAugspurger]

Just gave a quick look through here. This is a good idea that we should support.

We'll need tests for this. Put them in pandas/tests/test_graphics.py. Make sure to cover cases where

• by is None or a column
• weights is an array of weights or a string column
• values has missing data and / or weights has missing data

The plotting stuff is being refactored a bunch currently, so I've tagged this for the 0.18. You might want to hold off on making more changes, but in the meantime you can write tests for this.

[jreback]

FYI the checking is quite similar to how weights are checked for DataFrame.sample, so would want to make this a common function (could be a private function on a dataframe)

[Twizzledrizzle]

Thanks jreback, I will try my best to add some tests. I will also check what can be done with synching dropna with the weighs.

I was thinking in the lines of
drow rows in weighs, that are na in group
drop rows na in group
fillna in weighs with zero, so they do not count to anything

would this be ok?

[Twizzledrizzle]

Or more logical to drop all rows that are NA in group or in weights?

[TomAugspurger]

I'd say your lest method. Something like data.dropna(subset=['group', 'weight'])

[jreback]

@TomAugspurger can you review

[Twizzledrizzle]

Sorry I have not had time to continue with the tests, my limited git knowledge made it tough interacting with the repo. I planned to look how my patch worked with the new release

[Twizzledrizzle]

Added a new pull request here: #11441

for the new version, sorry for not using git correctly :(

[TomAugspurger]

Superseded by #11441