Panel.ffill ignores axis parameter and fill along axis=1

[Tux1]

A small, complete example of the issue

a = pd.Panel({'a':np.arange(15).reshape((5,3)), 'b':np.arange(15,30).reshape((5,3))})
a['a'].loc[1,1] = np.nan

a['a']
Out[36]:
    0     1   2
0   0   1.0   2
1   3   NaN   5
2   6   7.0   8
3   9  10.0  11
4  12  13.0  14

a.ffill(axis=2)['a']
Out[38]:
      0     1     2
0   0.0   1.0   2.0
1   3.0   1.0   5.0
2   6.0   7.0   8.0
3   9.0  10.0  11.0
4  12.0  13.0  14.0

Expected Output

a.at['a', 1, 1] == 3
If you ffill along axis=2, you should get 3 instead of 1

Output of pd.show_versions()

Pandas 0.19.0

[jorisvandenbossche]

@Tux1 Thanks for the report.

PRs with a fix are certainly welcome (should be rather easy, I think axis has to be passed here: https://github.com/pandas-dev/pandas/blob/v0.19.0/pandas/core/generic.py#L3238). But note that we are not giving Panels many attention anymore for new features: #13563

[Tux1]

@jorisvandenbossche Yes, it is quite easy to fix. Will do it as soon as i have free time
Yes I know, I am disappointed by this decision

[Tux1]

@jorisvandenbossche finally, it is not as easy as we thought.
I don't see how to handle the case axis=0 because I don't want to loop over major_axis or minor_axis. Any suggestion ?

[jorisvandenbossche]

@Tux1 In the PR, you now took the approach of raising a NotImplementedError for those cases? I think that is OK to start with

[Tux1]

@jorisvandenbossche yes but I will break compatibility with pct_change when axis=0 and probably some other function which use fillna
I could bypass forward filling in pct_change when axis=0 but it is a bit dirty

[Tux1]

@jorisvandenbossche Finally, it is fine. I handle the case axis=0, it is a bit ugly (I swap axes twice in case axis=0 to fall back on a trivial case) but it works fine.

[jreback]

dupe of #8251