random choice should support multidimensional arrays

[alan-isaac]

random.choice enhancement request:

random.choice should support multidimensional arrays by taking an axis number. With this enhancement, axis=None would choose from a flattened array, while an integer argument would chose from the subarrays along that axis.

Apologies if this is a duplicate. My previous submission seems to have vanished.

[seberg]

This should be simple if you are thinking about doing it. All that seems necessary is to add a special check for axis=None and ravel in that case and replace make axis=1. Then set pop_size = a.shape[axis] and pass axis into take. (Plus possibly an error if a is just an int and axis is given).

Actually thinking about it, it could also be reasonable to have size=None as default and have it return a scalar/the given axes removed in that case. That would be a real change in functionality unfortunately, but it would make sense for similarity to import random; random.choice mostly.

One could think about also allowing an out argument which is just passed into take mostly. While none of the other random functions have that it could be a nice performance option in some cases.

[samuela]

+1

[jhollowed]

+1. Is this still not implemented?

[jbachh]

+1

[slidenerd]

Is this implemented?

[mikesmith1611]

Is this ever going to be implemented?

[JulianKlug]

+1

[stephenpardy]

+1
I also tried implementing this, before realizing that @masasin had already done it.

[mattip]

Implemented in the new randomgen API: np.random.Generator().choice(..., axis=int). Closing since we will not be updating the legacy code to support this. If you feel differently, please reopen for discussion.