Emmanuel Bengio

It's great that z.lua has this feature, but I'm suggesting to change the default. Rather than have users look for such environment variables for every tool they install, we could...

I also went looking for the RL and found none. For posterity (happy to be corrected by the authors), what the finetuning step (and `property_filter`) seems to consist in is...

Hello, not the author, but I managed to load the sklearn models with `scikit-learn==0.21.3`, `scipy==1.5.4`, `numpy==1.17.0`. To run this in a newer environment I used `skl2onnx==1.11.0` to export the models...

FWIW applying this patch solved another problem for me where when rendering multiple images some erroneous textured triangles would appear between images (as if the batching went wrong within OpenGL).

I'd rather not have models in the git repo itself, we could follow the pattern of `bengio2021flow.py`, and link to the original (or, in case it disappears, I made a...

Not all molecules can be expressed by the default fragment set used in the code base. Normally the search stops after a while if that's the case, but I don't...

It does look like these molecules contain subgraphs which are not within the [default fragment set](https://github.com/recursionpharma/gflownet/blob/trunk/src/gflownet/models/bengio2021flow.py#L34), so indeed converting them would not work. You may be able to generate your...

> is it right if finally I consider the fragments and their respective atom_idx to be: Yes, that's more or less how we did it. We then look for fragments...

You're right, colab updated their torch version which broke the torch_geometric `pip install` command. I updated it, should now be `!pip install torch_geometric torch_sparse torch_scatter rdkit gitpython omegaconf wandb --find-links...

I can't really help with the specific files you're sharing, but maybe one thing to make sure of is that your reward is always strictly positive so that the log-reward...