[Python] Add a FixedSizeTensorScalar class

[wjones127]

Describe the enhancement requested

When you get a scalar off of a FixedSizeTensorArray, you get back a generic ExtensionTensor. There's not an easy way to get the tensor contained back as a numpy ndarray. It would be nice to have:

def FixedSizeTensorScalar(pa.ExtensionScalar):
 def to_numpy_ndarray(): ...

Then one could do:

tensor_array[i].to_numpy_ndarray()

As a workaround, I currently do:

tensor_array.to_numpy_ndarray()[i]

Which isn't too bad.

Component(s)

Python

[jorisvandenbossche]

Sounds good!

[rok]

I can grab this one.

Also, another workaround would be:

tensor_array.slice(i, i+1).to_numpy_ndarray()[0]

[wjones127]

Also, another workaround would be:
tensor_array.slice(i, i+1).to_numpy_ndarray()[0]

Yes, but that doesn't work in 12.0.0. There is a bug that was fixed in 13.0.0 where to_numpy_ndarray() didn't work properly on sliced tensors.

[rok]

Oh yeah, that's right, #35573.