1D tensor not displayed correctly

[tgkolda]

When I display a 1D tensor, it has a spurious comma in the shape and is printed in row rather than column orientation:
Code:

data = np.array([5,7,8,1,3,9,5,3,5,2,8,4,5,1,7,2,9,4,7,7,8,1,3,5]);
X = ttb.tensor(data, (3, 4, 2));
print("X[1, :, 1] =\n", X[1, :, 1])

Output:

X[1, :, 1] =
 tensor of shape (4,)
data[:] =
[1 9 7 3]

Compare to MATLAB:

X = tensor([5,7,8,1,3,9,5,3,5,2,8,4,5,1,7,2,9,4,7,7,8,1,3,5],[3 4 2]);
X(2,:,2)

Output:

ans is a tensor of size 4
	ans(:) = 
	     1
	     9
	     7
	     3

[ntjohnson1]

This is consistent with our documentation and numpy. We note that we follow the numpy convention for 1D tensors to be a row.

Numpy has three flavors of 1D arrays of length N with shape (N,1), (1,N), and (N,) corresponding to explicit row, explicit column and a flexible 1D vector that can serve either.

So I think this is conceptually aligned but visually different. Have you run into functional issues because of this or just stylistic?

[tgkolda]

In MATLAB, we ensure any 1D tensor (or fiber of a larger tensor) is explicitly a column array. This is partly because MATLAB has no concept of a 1D array. But it's also consistent with the ordering of tensor modes. A row vector is technically a 1 x N tensor, whereas a column vector is an N or N x 1 tensor. This contrast to what you're saying, so I'm a bit confused.

[tgkolda]

Also, where is this in the documentation?

[ntjohnson1]

RE: Confusion
I think my ordering wasn't consistent. What you described matches for numpy as well.

>>> N = 4
>>> np.ones((N,))
array([1., 1., 1., 1.])
>>> np.ones((N,1))
array([[1.],
       [1.],
       [1.],
       [1.]])
>>> np.ones((1,4))
array([[1., 1., 1., 1.]])

I think unless there is a strong functional reason we've tried to align design choices to numpy style since that is by far the most popular array/scientific computing package in python. We can see if the true 1-D array causes any issues for use or if its mostly just a style difference (unless you already have an example to review)

[ntjohnson1]

Re: Documentation

https://pyttb.readthedocs.io/en/stable/tensor.html#pyttb.tensor Above parameters describing the default constructor or search for Numpy convention of being a row vector. We probably should make it under a Notes heading rather than just saying note to make it clearer.

[ntjohnson1]

Discussion on #350 appears to subsume this so closing here so only tracking in one location