DeepChem Minutes
India/Asia/Pacific Call
Date: February 4th, 2021
Attendees: Bharath, Mufei
Summary:
Bharath this week has been iterating on the AtomicConvModel. He’s made progress debugging the model but will likely require more time
Mufei made two new submissions on the HOPV dataset. This dataset doesn’t belong to pre-existing categories of MoleculeNet. We might need to add a new category to MoleculeNet v2. We probably don’t need to apply rd_filters to it. The issue applying rd_filters might need to add the filter directly to the moleculenet loaders.
Bharath mentioned as a topic for general discussion that the DeepChem 2.5.0 release is coming up for the end of the month so it would be good to plan for what we want to add in the release.
Americas/Europe/Africa/Middle East
India/Asia/Pacific Call
Date: February 5th, 2021
Attendees: Bharath, Seyone, Peter, Maria, Thiago
Summary: We had a couple new attendees so we started with intros.
Maria is from the Czech republic and is a PhD student working on QSAR modeling. We’ve developer. Thiago studies chemical engineering and does research in artificial intelligence applied to molecular property discovery
Bharath gave the same update as on the previous call.
Peter is working on getting our CI system functional again. A lot of things are failing. Part of it came from the move to Numpy 1.20 which introduced type annotations. Peter is fixing some of them, but partly downgrading to Numpy 1.19. Some of it came from TensorFlow 2.4. Part of the challenge is that github actions works differently from Travis CI since one failure triggers a shutdown. We’ve gotten in a lazy habit of merging in code with a CI failure, but that means that serious test failures are going in.
Seyone has merged in the updated ChemBERTa tutorial which merged the smiles tokenizer and BPE tutorials. The other update has been testing an updated tokenization method for ChemBERTa. Working as well on rdkit descriptor pretraining with ChemBERTa.
Vignesh has submitted a PR for the LCNN model. Once this and the LCNN featurizer model is merged in then we can merge in the model and also do a tutorial.
Joining the DeepChem Developer Calls
As a quick reminder to anyone reading along, the DeepChem developer calls are open to the public! If you’re interested in attending either or both of the calls, please send an email to X.Y@gmail.com, where X=bharath, Y=ramsundar.