You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In PyMC3 the dims feature that allows naming of shapes leads to great UX. It would be great to push this down into Aesara and extend it. One thing that would be nice is to have named indices so that one could do: x["foo"] instead of x[0].
Same with dimensions: x.sum(axis="bar"). A lot could probably be borrowed from xarray who are doing this really well.
The text was updated successfully, but these errors were encountered:
twiecki
changed the title
Add named dimensions
Add support for named dimensions and coordinates like xarray
Apr 5, 2021
twiecki
changed the title
Add support for named dimensions and coordinates like xarray
Add support for named dimensions and coordinates like in xarray
Apr 5, 2021
Just to link this, as it could be useful, Torch does something like this with named tensors. Numpy also has its structured arrays, but those are much more flexible than what we need to get names into dimensions.
I think that the key thing that we would need to do as a bare minimum is to handle dimension name propagation through the graph.
This came up in a conversation with @aseyboldt.
In PyMC3 the
dims
feature that allows naming of shapes leads to great UX. It would be great to push this down into Aesara and extend it. One thing that would be nice is to have named indices so that one could do:x["foo"]
instead ofx[0]
.Same with dimensions:
x.sum(axis="bar")
. A lot could probably be borrowed fromxarray
who are doing this really well.The text was updated successfully, but these errors were encountered: