multi-index panel to xrarray dataarray index selection

[toniatop]

I am playing with xarray (moving away from pandas panel) but I cannot get my head around the following example. I create a multiindex panel and then convert it to a dataarray. I would like then to slice as I used to do in pandas , but trying the equivalent it fails with TypeError: '(slice(None, None, None), [3])' is an invalid key

Similarly I cannot get x.sel(major_axis=?) to work trying to achieve the same result.

import numpy as np
import pandas as pd
import xarray as xr
columns=['A','B','C']
index=pd.MultiIndex.from_product([['a','b','c','d'],[1,2,3]])
items=['x','y','z']
z=pd.Panel.from_dict({i:pd.DataFrame(data=np.random.rand(12,3),index=index,columns=columns) for i in items})
z.loc[:,(slice(None),[3]),'A'] #<-- That's what I want to do in xarray

x=z.to_xarray()
x.loc[:,(slice(None),[3]),'A'] #<- It fails

Do you have any insight?

[shoyer]

Indeed, I'm afraid this currently doesn't work in xarray -- we need to use the get_loc_level method.

I believe @benbovy's fix in #802 should solve this (are you still working on that patch, Benoit?).

[benbovy]

Sorry for the delay. I put #802 aside for some time but now I'm working on it again and hopefully I'll commit tests and docs in a couple of days.

#802 fixes the case above, although it currently requires to pass a dictionary with explicit index level names or numbers as keys:

x.loc[:, {0: slice(None), 1: 3}, 'A']  #<- It works with #802

[benbovy]

After some additional commits in #802:

In [1]: x.loc[:, (slice(None), [3]), 'A']
Out[1]:
<xarray.DataArray (items: 3, major_axis: 4)>
array([[ 0.23589318,  0.28298667,  0.84166271,  0.22935685],
       [ 0.22939828,  0.47265628,  0.89810054,  0.17696009],
       [ 0.65109315,  0.7532165 ,  0.12492445,  0.12908149]])
Coordinates:
  * items       (items) object 'x' 'y' 'z'
  * major_axis  (major_axis) object ('a', 3) ('b', 3) ('c', 3) ('d', 3)
    minor_axis  <U1 'A'

[shoyer]

Thanks to #802, this works now in xarray v0.8.