Max Balandat
Max Balandat
We'd also need this to be auto-differentiable in pytorch though. So the idea was to (maybe) write this using the pytorch C++ frontend as an extension.
Oh I see different implementation, my bad.
> It would be nice to have a highly-optimized C/C++ library of fundamental algorithms (not only hypervolume, but dominance filtering, nondominated sorting, other metrics like IGD+ and epsilon, etc) for...
I haven't looked into the details, but I wouldn't be surprised if the implementation of this particular algorithm were indeed many orders of magnitude slower. This algorithm is very loopy...
I'm curious about what difference in timing it makes if you set *deduplicate=False` in `is_non_dominated`.
Another related issue here: https://github.com/pytorch/botorch/issues/2310#issue-2265529284
Hmm interesting - this looks to be very deep down in the torchx stack. The fact that it's an `OSError` makes me think that this is not really an Ax...
Our setup is pretty much the same as Ax's, so let's move the discussions of the technical aspects of this to https://github.com/facebook/Ax/issues/1227.
Hmm yeah I'm not surprised, I don't think we've tried / tested this before. We'll take a look
Yeah we may well have to change something on the gpytorch end, but that's ok.