IndexError: tuple index out of range with pandas 0.25.

[Foadsf]

Bug report

Bug summary

I have reported the issue in details here
basically after an upgrade to 3.1.1 it drops the error:

IndexError: tuple index out of range

all other details inclduing the CSV file and code are in the above link.

[jamespreed]

Looks like this commit changed how the indices were assigned by removing the _xy_from_xy method. This works as of 3.1.0, but not 3.1.1 (on Windows).

[Foadsf]

Downgrading to the previous versions doesn't help! There might be more issues IMHO ¯\_(ツ)_/¯

[anntzer]

This is actually due to the upgrade to pandas 0.25:

import pandas as pd
df = pd.DataFrame({"a": [1, 2, 3]})
print(df.index.shape, df.index[:, None].shape)

This now prints (3,) (3,), but with pandas 0.24 used to print (3,) (3, 1) which we relied on to convert input to 2D.

[Foadsf]

@anntzer downgrading to Pandas 0.24 solved the problem!

[Foadsf]

I reported an issue on pandas repo here

[jorisvandenbossche]

@story645 I am not fully sure we can provide an upstream fix (or at least not long term).

df.index is a one-dimensional object, so df.index[:, None] actually giving an Index that is 2D, could have been considered a bug in pandas.

[tacaswell]

@Foadsf Can you please paste a minimal copy-pastable example into this issue? SO links rot over time.

[story645]

@jorisvandenbossche wondering then if pandas should be throwing a warning or something since
array[:,None] syntax is the numpy recommended way of expanding dims and so the expectation there is to return a 2D item.

[jklymak]

What does np.newaxis give if we use that instead?

[Foadsf]

Dear @tacaswell the Pandas guys were kind to edit my post here and add clear and concise examples demonstrating the issue.

[jorisvandenbossche]

@Foadsf that's a reproducible example for just the pandas side, but I think @tacaswell is looking for an example that shows the error with matplotlib.

Something like

s = pd.Series([1, 2, 3])
plt.plot(s.index, s.values)

[tacaswell]

Thanks @jorisvandenbossche

@Foadsf in general, you will get much better responses from projects if you provide an easily understandable and testable examples (see http://www.sscce.org/ for a longer discussion as to why). The information we ask for in the issue template to guide you so we can help you as easily as possible.

[QuLogic]

@jklymak np.newaxis is None; there is no difference between the two other than programmer clarity.

[ImportanceOfBeingErnest]

df = pd.DataFrame({"XX" : [4,5,6], "YY" : [7,1,2]})
print(df.index[:,None].shape)
print(df["YY"][:,None].shape)

prints

(3,)
(3, 1)

This seems rather inconsistent.

Marking as release critical, because something needs to be done here.

[jorisvandenbossche]

This seems rather inconsistent.

@ImportanceOfBeingErnest yes, that is certainly inconsistent, and "kind of" a regression in pandas (which I also propose to fix in pandas itself for 0.25.1, see pandas-dev/pandas#27775 (comment))
I am saying "kind of" because we should decide what to do when you end up with a 2D Index (which is invalid, but currently you can get that with the indexing operation as idx[:, None]): return a 2D array or raise an error.
series[:, None] does return a 2D numpy array, hence you get the correct shape.

But in any case, I think it would be good for matplotlib to not rely on this indexing aspect to convert a pandas Index into a 2D array-like.