ENH: support coordinate access for indexing

[jreback]

df.iloc[[0,1],[0,1]] means return the square 0-1 rows and 0-1 columns

might be nice to have a way to mean return the coordinates (0,0),(0,1)

related #7522

http://stackoverflow.com/questions/23686561/slice-a-pandas-dataframe-by-an-array-of-indices-and-column-names/23686855#23686855

[immerrr]

panel.fancy[:, [x, y], [0, 1]] -> df, where index=[(x, 0), (y, 1)] and columns=[a, b, c]

I don't think another indexer is the answer

Well, there's not much I can fancy doing to overload getitem:

• add an overload wrapper to ndframe in the form of an indexer panel.coord[:, [x,y], [0,1]]
• add an overload wrapper to index expression panel.loc[pd.coord[:, [x,y], [0,1]]]
• add an overload wrapper to index expression components panel.loc[:, pd.coord[x,y], pd.coord[0,1]]

With the second option being a syntactic variation of the first, I take it you have something like the third one in mind. I would prefer something like that, too: it's lengthier yet a lot more flexible (e.g. would allow combining coord and non-coord lookups). In fact, I proposed something like that a while ago in #6328, but that didn't seem to get public appeal.

[jreback]

@immerrr I remember that issue, but I didn't look in detail hah!

so if we were to expand/add to pd.IndexSlice, maybe pd.Coord or something.

I am wondering how complicated this would get

[immerrr]

Converting spelled-out look ups to numpy flavour should not be, but supporting such numpy lookups in BlockManager code could.

[jreback]

I don't think we need to be a slave to numpy

[immerrr]

As long as pandas containers are built on top of np.ndarray it's simply a matter of convenience (and programmer's laziness).

[jreback]

no, what i mean is fundamentally pandas is much more flexible, so we have to have a more general soln to things like indexing (which means its more complicated). Since already support location, positional, boolean masking in single dimensions, we are contemplating adding a multi-dimensional indexer (that could be any of the above).

[jreback]

http://stackoverflow.com/questions/24833130/how-can-i-select-a-specific-column-from-each-row-in-a-pandas-dataframe

This is the .lookup method essentially.

maybe ought to just turn this into an indexer

df.lookup[(0,'a'),(2,'b')] ?