Date: February 25th, 2021
Attendees: Bharath, Mufei, Vignesh
Summary: Bharath this week worked primarily on reviewing PRs and on sourcing this list of DeepChem papers and discoveries. Putting together a formal list can help us when we apply for grants or try to get a wikipedia page set up.
Mufei has been mostly busy with a new release of DGL, but in the meanwhile Yuanqi has been adding some new datasets and submissions. Mufei mentioned that Bharath opened an issue about a potential discrepancy in BBBP benchmark. Mufei also mentioned some issues in loading and using MoleculeNetV2 datasets, including:
- band gap
- Datasets for which we might want to apply RD-Filters
Mufei mentioned that we would need to work out these issues before we could release the MoleculeNet paper. Vignesh this week worked on his PR for the LCNN model. There are some failures there so Bharath suggested that it might be useful to rebase.
Date: February 26th, 2021
Attendees: Bharath, Peter, Amrita, Seyone, Nathan
Summary: Bharath gave the same update this week.
Peter got a PR in fixing the tests that were failing and making some of the flaky tests more robust. Peter is also working on converting the test suite to use conda-forge to reduce failures.
Amrita is getting up to speed with DeepChem.
Seyone is working on pretraining for ChemBERTa and converting it to a generalized script. Seyone hopes this moves towards getting ChemBERTa into DeepChem. Seyone was also working on a review paper on uncertainty estimation for part of his ime.
Nathan has mostly been working on the atomic convolution models and featurizers this week. The slow tests are fixed (PR) and the atomic conv models are fixed at least for unit tests (PR). Also fixed a bug with the new version of pytorch spline conv (PR).
We moved to the general discussion. We made a call to push back the DeepChem 2.5.0 release by 1 week to try to get in the RDKit grid featurizer fixes and take a look at the TensorFlow 2.4.0 upgrade.
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.