Gabrijel Boduljak

@NeptuneIsTheBest This is a good suggestion. I am willing to implement this after we get some feedback on MaxPooling and AvgPooling. I would like to add these in a separate...

> Hello, when I looked at the PR #357 I found that there are only a few basic pooling layer implementations such as MaxPooling and AvgPooling. Will we add an...

> > @NeptuneIsTheBest This is a good suggestion. I am willing to implement this after we get some feedback on MaxPooling and AvgPooling. I would like to add these in...

> @gboduljak You are right, but I actually use AdaptiveAvgPool2D and AdaptiveMaxPool2D in many places. And (as a consequence of not spending the money to upgrade to 36GB of unified...

> I get this: `ValueError: [tril] array must be at least 2-D` I am also getting this error. I am working on M2.

> The main question is whether we can automatically detect the above and replace it with `scatter_add` or `scatter_{op}` depending on the operation. It might also simply not be worth...

> We can but it is not that simple as they are not equivalent operations. The `scatter_add` is atomic while the other one isn't (still atomic but not the increment)....

> Honestly the more I think about it I think this shouldn't be implemented. It would also break numpy compatibility which implements sliced `__iadd__` the same way. > > The...

> > Honestly the more I think about it I think this shouldn't be implemented. It would also break numpy compatibility which implements sliced `__iadd__` the same way. > >...