torch.chunk fails to split correctly

[vivanov879]

th> torch.rand(11):chunk(5)
{
1 : DoubleTensor - size: 3
2 : DoubleTensor - size: 3
3 : DoubleTensor - size: 3
4 : DoubleTensor - size: 2
}

[vivanov879]

splits in only 4 chunks, while 5 is specified

[andreaskoepf]

@vivanov879 I created an alternative impl for chunk which always tries to return the a table with the number of entries requested by the user. My implementation has two 'effects' which need to be discussed:

1. The size of more than the last tensor in the result table can vary.
2. There are cases with empty tensors if the desired chunk count exceeds the size of the source dimension.

[vivanov879]

your PR works ok

[andreaskoepf]

After discussion with @soumith & @nicholas-leonard the original torch.chunk() behavior is considered 'by-design'. For anybody interested in a solution that always returns exactly the requested number of chunks I created this gist.