Daniel Falbel
Daniel Falbel
@cregouby Feel free to take maintainership role if you want. You would only need to change those lines: https://github.com/mlverse/tabnet/blob/f4f815f43e017ab3b6169e730b74037397041033/DESCRIPTION#L5-L6 so you get the `'cre'` role.
We still didn't do any profiling of tabnet code to detect bottlenecks so this is expected in general. It's likely that future improvements in the torch package will make it...
All the Rprof results is showing torch code. Why do you state that improvements in torch will decrease the 100 sec and not the rest?
I don't know exactly what you mean by the same C code. It's indeed different from xgboost where there's a C++ implementation and both python and R bind to that...
Sorry @egillax ! I ended up re-implementing it here: https://github.com/mlverse/torch/pull/903 to avoid conflicts when merging.
Indeed! This is a bug, we should keep the same class as the provided dataset!
@y-vectorfield Do you have any ideas how to fix this?
I am not working on LBFGS. Pinging @krzjoa as he worked on many other implementations. The main reason for not using the C++ implementations was that I wanted to allow...
Sounds good to me! +1 for a new package for fancier optimizers! that would be really cool
Yes, In theory that's possible. I'll draft something in that direction and post here. Ideally I'd really want to figure what's slowing down optimizers in R compared to PyTorch, because...