BUG/ER: Stricter testing of 'monotocity' when reindexing with ffill or bfill.

[prossahl]

closes #4484

Reversing an index and forward filling succeeds but produces misleading data:

>>> dr = pd.date_range('2013-08-01', periods=6, freq='B')
>>> df = pd.DataFrame(np.random.randn(6,1), index=dr, columns=list('A'))
>>> df['A'][3] = np.nan
>>> df.reindex(df.index[::-1], method='ffill')
                   A
2013-08-08  2.127302
2013-08-07       NaN
2013-08-06       NaN
2013-08-05       NaN
2013-08-02       NaN
2013-08-01       NaN

This change enforces "monotocity" (well actually weakly increasing) and causes a ValueError to be raised.

[jreback]

Can you also add an example of how to do this in the missing_data.rst ? (and maybe a pointer from the cookbook)?

also need release notes (put in the API section), reference the issue #4484

This is the solution (but yours is nice on error reporting!)

In [11]: >>> df.reindex(df.index[::-1]).reset_index().ffill().set_index('index')
Out[11]: 
                   A
index               
2013-08-08  1.112043
2013-08-07  0.250710
2013-08-06  0.250710
2013-08-05 -0.499677
2013-08-02 -0.327536
2013-08-01 -1.774846

[jreback]

also pls hook up travis to this (see contributing.md)

[jreback]

@prossahl ok...this looks good (ignore the failure on travis)...that's a weird error that is fixed in master

merge?

[jreback]

and I added the example I put above to the cookbook (will be there later today)

[prossahl]

I have moved this to a PR #4642.

[jreback]

ok...close this one then?

[prossahl]

Will do. Over to GH4642.