Daniil Polykovskiy
Daniil Polykovskiy
@SeemonJ, can you please comment on this?
Hi, @Dom303! If you install MOSES using pip (`pip install molsets`), the requirement is torch>=1.1.0 — it should not affect your local version of PyTorch. We specify a fixed (older)...
@lilleswing, the code from MOSES you've sent implements teacher forcing. Did you mean that we should add free running for training?
Yes, we’ll add free run soon. It will probably be denoted as a separate model at the metrics table.
Hi, @liujunhongznn Hi! Low reconstruction quality is due to the posterior collapse that frequently happens in VAEs. Since the goal of MOSES is to produce the generative distribution as good...
Hi, @bokertof! VAE in MOSES uses teacher forcing—we pass the correct token, not the sampled one.
If you feed sampled tokens, you have to propagate the gradient through sampling (e.g., with REINFORCE), which has notoriously high variance. You could use variance reduction techniques, but it lies...
Hi, @yuhuafang! We do not specify the stereochemistry of compounds in the dataset since none of the metrics take stereochemistry into account. However, you can obtain a dataset with stereochemistry...
PR #8 should resolve the issue :)
I think you should remove pip version (`pip uninstall hunga-bunga`) and install the package again from source (`python setup.py install`).