Clean Up Tensor/Sptensor setitem (and getitem to match)

[ntjohnson1]

This is an extension from the long discussion here but the short summary is:
This is the example from MATLAB #92 (comment) with this being the example from Danny with additional elaboration on expected behavior #92 (comment). Will follow up on more details about the expected python interface to confirm. I think the pieces we have are fairly close already so this shouldn't impact correctness of existing algorithms but rather to confirm this as a toolbox acts as expected.

[ntjohnson1]

Closed with #116

[dmdunla]

Re-opening issue, as the behavior described in #92 (comment) seems to work for __setitem__ but not for __getitem__. Here are the cases for tensor that MATLAB supports (with examples illustrating my understanding of what the corresponding Python call should look like for a 4-way tensor). sptensor may need to be addressed as well, but this is a start for tensor.

Intended behavior (to align with MATLAB):
1a. Single subscript -> scalar: X[0,0,0,0]
1b. Slice -> tensor: X[0,0,0,:], X[:,0,0,:], or X[0:2,[1,3],0,:]
2a. Array (p x N) of subscripts -> array (p x 1): X[np.array([[0,0,0,0],[2,3,1,0]])]
2.b Array (p x 1) of linear indices -> array (p x 1): X[np.array([[0,5,6,22]]).T]

Current behavior:
1a. Working as in MATLAB.
1b. Working as in MATLAB.
2a. Does not seem to be working. Restricted to 2 x N array input and returns a slice not (p x 1) array.
2b. Does not seem to be working. Does not accept array input (only list) and return a 1-D array (which may be OK since we are using numpy, but this breaks from MATLAB).

Steps to reproduce error:

MATLAB:

>> X = tensor(1:24,[3 4 2 1])
X is a tensor of size 3 x 4 x 2 x 1
	X(:,:,1,1) = 
	     1     4     7    10
	     2     5     8    11
	     3     6     9    12
	X(:,:,2,1) = 
	    13    16    19    22
	    14    17    20    23
	    15    18    21    24

>> % Case 1a
>> X(1,1,1,1) 
ans =
     1

>> % Case 1b
>> X(1,1,1,:) 
ans is a tensor of size 1
	ans(:) = 
	     1
>> X(:,1,1,:)
ans is a tensor of size 3 x 1
	ans(:,:) = 
	     1
	     2
	     3
>> X(1:2,[2 4],1,:)
ans is a tensor of size 2 x 2 x 1
	ans(:,:,1) = 
	     4    10
	     5    11
>> X(1:2,[2 4],1,1)
ans is a tensor of size 2 x 2
	ans(:,:) = 
	     4    10
	     5    11

>> % Case 2a
>> S = [1,1,1,1;3,4,2,1]
S =
     1     1     1     1
     3     4     2     1
>> X(S) 
ans =
     1
    24
>> S = [1,1,1,1;2,2,2,1;3,4,2,1]
S =
     1     1     1     1
     2     2     2     1
     3     4     2     1
>> X(S) 
ans =
     1
    17
    24

>> % Case 2b
>> S = [1,6,7,23]'
S =
     1
     6
     7
    23
>> X(S) 
ans =
     1
     6
     7
    23

pyttb:

>>> import pyttb as ttb
>>> import numpy as np

>>> X = ttb.tensor.from_data(np.arange(1,25),(3,4,2,1))
>>> X[:,:,0,0] # printing like MATLAB
tensor of shape 3 x 4
data[:, :] =
[[ 1  4  7 10]
 [ 2  5  8 11]
 [ 3  6  9 12]]
>>> X[:,:,1,0]  # printing like MATLAB
tensor of shape 3 x 4
data[:, :] =
[[13 16 19 22]
 [14 17 20 23]
 [15 18 21 24]]

>>> # Case 1a
>>> X[0,0,0,0]
1

>>> # Case 1b
>>> X[0,0,0,:]
data[:] =
[1]
>>> X[:,0,0,:]
tensor of shape 3 x 1
[[1]
 [2]
 [3]]
>>> X[0:2,[1,3],0,:]
tensor of shape 2 x 2 x 1
data[0, :, :] =
[[ 4]
 [10]]
data[1, :, :] =
[[ 5]
 [11]]
>>> X[0:2,[1,3],0,0]
tensor of shape 2 x 2
data[:, :] =
[[ 4 10]
 [ 5 11]]

>>> # Case 2a
>>> S = [[0,0,0,0],[2,3,1,0]] # list of subscripts lists (should not be allowed)
>>> X[S] # unexpected behavior
array([[1, 1, 1, 1],
       [3, 4, 2, 1]])
>>> S = np.array([[0,0,0,0],[2,3,1,0]]) # array of 2 x N subscripts (should produce array of length 2)
>>> X[S]
array([[ 7, 19],
       [10, 22],
       [ 4, 16],
       [ 1, 13]])
>>> S = np.array([[0,0,0,0],[1,1,1,0],[2,3,1,0]]) # array of 3 x N subscripts (should produce array of length 3)
>>> X[S]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\dmdunla\dev\github.com\pyttb\pyttb\tensor.py", line 1524, in __getitem__
    a = np.squeeze(self.data[tuple(subs)])
IndexError: index 2 is out of bounds for axis 2 with size 2

>>> # Case 2b
>>> S = [0,5,6,22] # list of 4 linear indices (should not be allowed)
>>> X[S]
array([ 1,  6,  7, 23])
>>> S = np.array([[0,5,6,22]]).T # array of 4 x 1 linear indices (shoulf produce array of length 4)
>>> X[S]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\dmdunla\dev\github.com\pyttb\pyttb\tensor.py", line 1524, in __getitem__
    a = np.squeeze(self.data[tuple(subs)])
IndexError: index 5 is out of bounds for axis 1 with size 4

[ntjohnson1]

My slight preference would be to open a new ticket rather than re-opening an old one unless its the EXACT same underlying bug. Our indexing is fairly complicated, so I expect we will continue to have bugs/extensions around it. The PR I just opened ONLY attempts to address the tensor indexing pointed out above. So if that PR resolves I'd like to close the ticket and when the sptensor tutorial is prepared and we figure out bugs there we can open a new ticket. Otherwise there is some ambiguity as to when we could close this.