BUG: plot with kind=scatter fails when checking if an array is in the DataFrame

[TomAugspurger]

df = pd.DataFrame({'A': [1, 2], 'B': [3, 4]})
c = np.array([1, 0])
df.plot(kind='scatter', x='A', y='B', c=c, cmap='spring')

fails with a TypeError.

we check elif c in self.data.columns which tries to hash the array.

(I know I can just add the array as a column to the DataFrame)

[jorisvandenbossche]

How should this supposed to work in this case? The length of the array is not equal to the number of elements in the dataframe?

[shoyer]

@jorisvandenbossche c = np.array([1, 0]) also raises.

[jorisvandenbossche]

@shoyer yes, OK, I just found it a bit strange example (as this would have failed with matplotlib also)

I suppose this is related to #7780, and apparantly this case wasn't catched by the tests? So it seems a regression from 0.14

[shoyer]

@jorisvandenbossche Yes, this is a regression (and I introduced it in #7780). If nobody else gets to it first I can write a fix (should be pretty easy).

[TomAugspurger]

@jorisvandenbossche sorry about the bad example! I meant to have c=[1, 0]. My excuse is that I'm on a Windows PC right now and copy-pasting from the interpreter is a pain.

[jorisvandenbossche]

@TomAugspurger right-clicking with the mouse?

[TomAugspurger]

Wow, that is not at all obvious that that puts it on your clipboard :) Thanks.

[jorisvandenbossche]

Yep, indeed, it also took me a while to know this.. But now rather essential when I am working on Windows

[jorisvandenbossche]

@jreback What is actually the best way to check if something is allowed to be a column name?

elif c in self.data.columns gives an error with np.arrays eg, but how to check for this?

[TomAugspurger]

I'm having trouble coming up with anything better than checking if it's an iterable.

There may be something in core.common

[jorisvandenbossche]

@TomAugspurger there are some people working here at the sprint on this issue

[TomAugspurger]

Great! I'll be offline for the next hour or so but after that I'll stick around the computer. Have them post if there are any questions.

[jorisvandenbossche]

problem is that eg tuples are also iterables, but allowed as column names

[TomAugspurger]

Mmm. That's true.

So let's be explicit in the documentation and implementation. With

In [81]: df = pd.DataFrame({(1, 2): [1, 5], 3: [0, 1], 4: [1, 2]})

In [82]: df
Out[82]: 
   (1, 2)  3  4
0       1  0  1
1       5  1  2

A df.plot(kind='scatter', x=3, y=4, c=(1, 2)) should get the color (or size if s=(1, 2)) from the column (1, 2).

[TomAugspurger]

As far as implementing that, the best may be a try / except?

[jreback]

@jorisvandenbossche

if col in self.columns

will work but put in a try except
also if the columns are a MultiIndex its a bit tricky