tensor.from_function doesn't work with np.random.randn

[tgkolda]

I will be submitting a fix as part of #374

import numpy as np
import pyttb as ttb
np.random.seed(0)
T = ttb.tensor.from_function(np.random.randn, (4, 3, 2))
print(T)

produces

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[2], line 3
      1 #%%
      2 np.random.seed(0)
----> 3 T = ttb.tensor.from_function(np.random.randn, (4, 3, 2))
      4 print(T)

File ~\Documents\GIT Repos\pyttb\pyttb\tensor.py:217, in tensor.from_function(cls, function_handle, shape)
    214 shape = parse_shape(shape)
    216 # Generate data
--> 217 data = function_handle(shape)
    219 # Create the tensor
    220 return cls(data, shape, copy=False)

File numpy\\random\\mtrand.pyx:1306, in numpy.random.mtrand.RandomState.randn()

File numpy\\random\\mtrand.pyx:1467, in numpy.random.mtrand.RandomState.standard_normal()

File _common.pyx:657, in numpy.random._common.cont()

TypeError: 'tuple' object cannot be interpreted as an integer

[ntjohnson1]

This isn't a bug. Our type hints specify the function should take a tuple of ints Callable[[Tuple[int, ...]], np.ndarray], we describe it in language, and we show an example.

np.random.randn is also weird since it is a non-standard support method for MATLAB users that doesn't match the general ergonomics of the rest of numpy.

Is the request to broaden the signature of functions we support or highlighting insufficient documentation?

[tgkolda]

I understand that now and will close this. In the updated documentation, I am including more extensive hints on how to use the numpy randn function. Also, with respect to np.ones and the like, one should ideally be careful because they return in C order by default, necessitating a copy (which generated a warning). I also discuss this in the updated documentation.

[ntjohnson1]

Great! Thanks for all the documentation work.