Index without 0 in xerr/yerr causes KeyError

[diazona]

When creating an errorbar plot, if the data object being used for the errors (xerr or yerr) doesn't have 0 in its index, it produces a KeyError from matplotlib code. I can produce this with the following code sample:

import pandas as pd, numpy as np
i = np.array([1,2,3])
a = pd.DataFrame(i, index=i)
a.plot(yerr=a)

The underlying bug (if it is a bug) in matplotlib is responsible for several other issues, including #4493 and #6127, but this case is different because it uses only Pandas API methods, rather than passing a Pandas object to a matplotlib method. So it's a little harder to justify passing this off on matplotlib to fix, as was done in e.g. #6127.

If the "proper" fix ever is implemented in matplotlib code, it should solve this as well as #4493 and #6127 and all the others of that nature, but until that point, Pandas can work around it by converting the error object (if it is a Series or DataFrame) to an ndarray. I'm working on this in 4b04f80 but I'm not sure if there's a better way to fix it, or if this breaks something. (If so, the tests don't indicate it.)

[jorisvandenbossche]

@diazona This also seems a regression (in pandas or matplotlib), as I don't get an error with pandas 0.16.2 and matplotlib 1.4.3, but do get one with 0.17.1 and 1.5.0

[TomAugspurger]

Haven't looked, but sounds similar to matplotlib/matplotlib#5550

@diazona I'm assuming you're using matplotlib 1.5? Can you try with 1.4.3?

[diazona]

Yeah, matplotlib/matplotlib#5550 seems like the exact same issue. That didn't come up in my search, so I hadn't realized the matplotlib developers were open to fixing it. If they do implement that fix, it renders my patch obsolete (though I think it's still a semantic improvement).

Let me take some time to get a previous version of matplotlib and test with that. I am indeed using 1.5.0. Here's the complete version info:

>>> pd.show_versions()

INSTALLED VERSIONS
------------------
commit: 4b04f80c41684ba9ab05ce8c87b17b96fde87290
python: 3.4.3.final.0
python-bits: 64
OS: Darwin
OS-release: 15.2.0
machine: x86_64
processor: i386
byteorder: little
LC_ALL: None
LANG: None

pandas: 0.17.1+84.g4b04f80
nose: 1.3.7
pip: 7.1.2
setuptools: 18.5
Cython: 0.23.4
numpy: 1.10.1
scipy: 0.16.1
statsmodels: 0.6.1
IPython: 4.0.1
sphinx: None
patsy: 0.4.1
dateutil: 2.4.2
pytz: 2015.7
blosc: None
bottleneck: 1.0.0
tables: 3.2.2
numexpr: 2.4.6
matplotlib: 1.5.0
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: 2.8

[diazona]

Update: matplotlib 1.4.3 indeed does not exhibit the error. After an enthralling git bisect run, I traced the KeyError to code introduced in matplotlib/matplotlib@5f33e7d, on the 1.5.0 development branch.

For what it's worth, the current master HEAD matplotlib/matplotlib@9f6bb90 does raise the KeyError.

I haven't tested with other versions of Pandas.

[tacaswell]

Can someone re-check if this is still a problem? I think we fixed this is v1.5.1.

[andres-aguilar]

yes, here v. 1.5.1 and the same problem