tensor/find need not return a column array for vals

[tgkolda]

I don't think there is any reason to have tensor/find return a column array rather than a row array. It's inconsistent with what is returned if you do T[subs] on the result, so I think it would be better to return it as a row array.

[dmdunla]

This issue has been a challenge in balancing between TTB for MATALB alignment and pythonic code in pyttb development.

NumPy provides 1D array (single dimension/index), which are different than row and column vectors. So, in the case you describe above, tensor.find returns a column vector (as you note), but T[subs] returns a 1D array, not a row vector.

Here is an example:

>>> import pyttb as ttb
>>> T = ttb.tenrand(shape=(2,2,2));
>>> [subs,vals] = T.find()
>>> vals.shape
(8, 1)
>>> T[subs].shape
(8,)

There are some cases throughout pyttb where we have not altered the specific use of row ([1,N]) or column ([N,1]) vectors as we have attempted to align with TTB for MATLAB as much as possible.

In my example above, T.find() is implemented in pyttb, whereas T[subs] passes through to NumPy and relies on the slicing operations of numpy.ndarray.

Can you let me know how you are using tensor.find that led to this issue?