Best way to add an alias for a dimension name

[matthew-brett]

I hope the answer is not obvious, but I could not see it on a fairly quick search.

We have need of aliases for dimensions in our brain imaging datasets.

For example, for 4D MRI images, we might reasonably chose to label the dimensions as i, j, k, time. But, any one of the i, j, k dimensions can also be the slice dimension (the scanner usually collects the images on 2D slice at a time). So slice could be an alias for any of i or j or k. Perhaps we could set this information with:

img['k'].attrs['alias'] = 'slice'

We might want to index over the slice dimension, but this would involve us first - checking whether any dimension has a slice alias, and then indexing over the found dimension. We would really like to be able to use syntax like img.isel(slice=0) - to mean the same - in this case - as img.isel(k=0).

Is there a canconical way of achieving that (using an alias for the standard dimension name)? Or would we have to make a custom accessor to do this - e.g. img.our_accessor.isel(slice=0)`?

[dcherian]

You have three options AFAICT:

1. Create a duplicate slice variable
2. https://docs.xarray.dev/en/stable/generated/xarray.Dataset.filter_by_attrs.html
3. https://cf-xarray.readthedocs.io/en/latest/custom-criteria.html for an implementation of the custom accessor idea.

Xarray tries to not interpret attrs at all (since its hard to consistently propagate them, and its easy for them to be out of date). Exceptions are reading/writing data, or plotting.

[matthew-brett]

Thanks for the quick reply. Could you say more about creating a duplicate slice variable? Won't img.isel(<name>=0) need <name> to be a value in the dims tuple?

[dcherian]

This isn't well documented unfortunately but you can do this

img = xr.DataArray(dims=("i", "j", "k"), data=np.ones((120, 120, 120)), coords={"k": ("k", np.arange(120))})
img["slice"] = img["k"]
img = img.set_xindex("slice")
img.sel(slice=40)

[matthew-brett]

Thanks again! But - I noticed that:

img.isel(slice=40)

after your code, still fails with:

ValueError: Dimensions {'slice'} do not exist. Expected one or more of ('i', 'j', 'k')

Is this expected?

[dcherian]

Yes. set_xindex sets an "index" which translates from coordinate-space to index-space. We haven't enabled "index-space" selection with .isel on such additional coordinates. (cc @benbovy )

For completeness here's the cf-xarray version if you wanted to experiment with the accessor style interface:

import xarray as xr
import numpy as np

img = xr.DataArray(
    dims=("i", "j", "k"),
    data=np.ones((120, 120, 120)),
    coords={"k": ("k", np.arange(120))},
)
img.k.attrs["alias"] = "slice"

import cf_xarray as cfxr  # registers .cf

cfxr.set_options(custom_criteria={"slice": {"alias": "slice"}})
img.cf.isel(slice=2)
img.cf.sel(slice=2)