Doubles only?

[culurciello]

Is there a reason why the algorithm only works on doubles?

[lvdmaaten]

Hmmm, not really I guess. You want to be able to use floats as well? I'll work on that...

[culurciello]

I tried to pass floats and the results were incorrect. Did you try or did I make a mistake?

[lvdmaaten]

I also get incorrect results when I pass floats, so I don't think it's your mistake.

[culurciello]

Hi Laurens van der Maaten, do you still have a plan to used Floats, or should we look into it?

[lvdmaaten]

Hmmm changing the Torch code is easy, but this also requires changing the C++ code, which will imply a bit more work. I'm not sure I will have time for it anytime soon.

If you want to do it yourself, then start from this version: #4

[nybbles]

Ran into this issue as well and did some investigation to find the root cause, then came here to Github to file an issue and found this.

I could take a stab at making this change, but I think an immediate and simple change would be to just check the tensor type and throw an informative error, right than having the code segfault.

[lvdmaaten]

Agreed!