Method for df

[gerigk]

See http://stackoverflow.com/questions/11941492/selecting-rows-from-a-pandas-dataframe-with-a-compound-hierarchical-index

let's say I want to filter a DataFrame such that level[k] only contains the values x, y or z.
A vectorized possibility is:

df[np.in1d(df.index.labels[k], match([x, y, z], df.index.levels[k]))]

but I guess it would be nice to have some pandas method wrapping it or implementing it more efficiently.

[lodagro]

What about putting this in df.ix[]? And when indexing a MultiIndex a tuple needs to be used. For each of the tuple values the same rules apply as for a single level index (including slicing and boolean indexing). When no tuple is used to index a MultiIndex it must be an int and its a position (always) or a boolean index.

suppose df is a frame:
- with a four level MultiIndex on both axis
- level2 of axis0 has five distinct values.

df.ix[('a', slice[None), [True, True, False, False, False]), slice(3, 7, 2)]
from axis0 select rows with:
- level value = 'a'
- all level 1 values
- only two first values of level 2
- all level 3 values
from axis 1 select positions 3 and 5

Now one can already use a tuple but it is not enforced, without tuple indexing a Multiindex works, but not always (try indexing without tuple on more than two levels) Also when the first level is integer, and you want to select a single row one needs to specify all levels, which can be a lot of typing.

[jreback]

closing in favor of #2425