Need a way to speciefy the names of coordinates from the indices which droped by DataArray.reset_index.

[weipeng1999]

When I try to use some different coordinates as the index of a dim, I notice the new API on v0.9 provided by DataArray.set_index:

>>> import numpy as np
>>> import xarray as xr
>>> arr = xr.DataArray(np.r_[:4].reshape(4,4),dims=('t','x'))
>>> arr['x_str'] = "x",["a","b","c","d"]
>>> arr["x_num"] = "x",[1,2,3,4]; arr
<xarray.DataArray (t: 4, x: 4)>
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15]])
Coordinates:
    x_str    (x) <U1 'a' 'b' 'c' 'd'
    x_num    (x) int64 1 2 3 4
Dimensions without coordinates: t, x
>>> arr.set_index(x='x_str'); arr
<xarray.DataArray (t: 4, x: 4)>
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15]])
Coordinates:
    x_num    (x) int64 1 2 3 4
  * x        (x) object 'a' 'b' 'c' 'd'
Dimensions without coordinates: t
>>> #... some code with "arr[{'x':[... some str index ...]}]" ...

That's really convenient, what a nice API.
But when I want to switch to another coordinate, I found that I can not recovery my arr to the version before I using :

>>> arr=arr.reset_index('x'); arr # why the croodinate to used as index now lose its name "x_str"? 
<xarray.DataArray (t: 4, x: 4)>
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15]])
Coordinates:
    x_num    (x) int64 1 2 3 4
    x_       (x) object 'a' 'b' 'c' 'd'
Dimensions without coordinates: t, x
>>> arr=arr.set_index(x="x_num");arr # anyway, continue going to the code use coordinate "x_num" as index
>>> #... some code with "arr[{'x':[... some int index ...]}]" ...
>>> arr=arr.reset_index('x');arr # now I need "x_str" coordinate as index, here we go
<xarray.DataArray (t: 4, x: 4)>
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15]])
Coordinates:
    x_       (x) int64 1 2 3 4
Dimensions without coordinates: t, x
>>> #NOOP!!! the "x_int" coordinate COVER the "x_str" coordinate, I can't access the later any more :(

To solve this problem, I get following ways:

• Add new function to do this.
  • may be most directly way.
  • add new function to implement and new API to design.
• stop the DataArray.set_index to change coordinate name
  • The benefit is that without new function, its the most convenient one.
  • But it may change the calling-promise of DataArray.set_index, so it may be useless.
  • Here are some example to show how it can be used:

>>> arr = arr.set_index({'x':'x_str'})
>>> #... some code with "arr[{'x':[... some str index ...]}]" ...
>>> arr = arr.set_index({'x':'x_int'})
>>> #... some code with "arr[{'x':[... some int index ...]}]" ...
>>> arr = arr.set_index({'x':'x_str'}) ...

• store the coordinate name in DataArray when calling "DataArray.set_index" and recovery them when calling "DataArray.reset_index"
  • Just a bit more complex than previews one.
  • But it need to add redundant data inner DataArray, useless too.
  • Example:

>>> arr = arr.set_index({'x':'x_str'})
>>> #... some code with "arr[{'x':[... some str index ...]}]" ...
>>> arr = arr.reset_index('x').set_index({'x':'x_int'})
>>> #... some code with "arr[{'x':[... some int index ...]}]" ...
>>> arr = arr.reset_index('x').set_index({'x':'x_str'}) ...

• let DataArray.reset_index support Mapping as names parameters, while use the keys as dims to reset indices and the value as the names of coordinates converted from those indices.
  • More complex.
  • Maybe the one cause least change, so I prefer it.
  • Example:

>>> arr = arr.set_index({'x':'x_str'})
>>> #... some code with "arr[{'x':[... some str index ...]}]" ...
>>> arr = arr.reset_index({'x':'x_str'}).set_index({'x':'x_int'})
>>> #... some code with "arr[{'x':[... some int index ...]}]" ...
>>> arr = arr.reset_index({'x':'x_int'}).set_index({'x':'x_str'}) ...

[shoyer]

You currently need reset_index because Xarray requires that indexes match a dimension.

We plan to relax this constraint soon, as part of the ongoing "Explicit index" refactor being led by @benbovy :

• Explicit indexes in xarray's data-model (Future of MultiIndex) #1603
• https://github.com/pydata/xarray/projects/1

The desired behavior is the possibility to set x_str as an index without changing its name.

[benbovy]

The desired behavior is the possibility to set x_str as an index without changing its name.

This will be enabled by #6971, hopefully merged before the next release:

arr = arr.set_xindex("x_str").set_xindex("x_num")

arr
# <xarray.DataArray (t: 4, x: 4)>
# array([[ 0,  1,  2,  3],
#        [ 4,  5,  6,  7],
#        [ 8,  9, 10, 11],
#        [12, 13, 14, 15]])
# Coordinates:
#   * x_str    (x) <U1 'a' 'b' 'c' 'd'
#   * x_num    (x) int64 1 2 3 4
# Dimensions without coordinates: t, x

arr.sel(x_str="a")
# <xarray.DataArray (t: 4)>
# array([ 0,  4,  8, 12])
# Coordinates:
#     x_str    <U1 'a'
#     x_num    int64 1
# Dimensions without coordinates: t

arr.sel(x_num=1)
# <xarray.DataArray (t: 4)>
# array([ 0,  4,  8, 12])
# Coordinates:
#     x_str    <U1 'a'
#     x_num    int64 1
# Dimensions without coordinates: t

[benbovy]

Closed in #6971.