Date: 3/20/2020
Attendees: Bharath, Peter, Seyone
Summary: Bharath spent most of the last week’s development on getting the MoleculeNet package up and running. He’s got an open PR which is working on transferring in the needed code from DeepChem. There’s a corresponding PR in the DeepChem repo which pulls code out. This PR won’t be merged into DeepChem head until after DeepChem 2.4. Bharath also got the documentation basics for readthedocs set up on https://deepchem.readthedocs.io/en/latest/ (see PR). However, the documentation is very rudimentary at present and will need to be fleshed out.
Peter worked on a PR for a general multitask convolutional network. This is now merged into head. This could be useful for things like working with the SmilesToImage transforms that turn SMILES into images.
Seyone’s been working on trying to apply HuggingFace models to smiles generation. There was a nice discussion on Twitter he kicked off discussing some of the methods that might be relevant for this project.
Bharath and Peter had a quick discussion about the DeepChem 2.4 release. It feels like we’re getting close, but the major blocker is figuring out the issues with the graph convolutions in TF2.X. Peter mentioned that he tried the obvious fixes for the issues, so it might be useful to see if we can get a simple failing code snippet that we can use to talk to TF2.X developers. Bharath is going to put some time into working to figure this out over the next week.