Date: 3/6/2020
Attendees: Bharath, Peter
Summary: Bharath’s spent the last week working on overhauling the DeepChem tutorial and notebook series. The tutorials are now ordered in a cogent sequence and have all been modified to run on Google colab. The hope is that working through the tutorial sequence will now be accessible to more developers, which could help grow out the community.
Peter’s nearing completion of the TensorFlow 2.X issue. One hurdle was the in_silico_mutagenesis() function which was used for some models. It looks like it can be reworked so it works for general models rather than the original very limited subset.
The other issues are the DAG and GraphConv models. At this point, it looks like the source is just bugs in TensorFlow 2.X that have broken the original implementation. We’ll have to re-implement the GraphConv model and probably sunset the DAG model (since it doesn’t appear anyone uses it at present). The plan is to get the TensorFlow 2.X PR merged in then work on fixing the GraphConv code and the tutorial sequence in subsequent PRs.
As a summary, a number of pull requests were merged in over the last week. One community PR [1] improved the scaffold splitting functionality with a follow-up [2] to fix a minor bug. A number of tutorial PRs were merged in as well [3], [4], [5], [6], [7]
[1] https://github.com/deepchem/deepchem/pull/1751
[2] https://github.com/deepchem/deepchem/pull/1762
[3] https://github.com/deepchem/deepchem/pull/1753
[4] https://github.com/deepchem/deepchem/pull/1755
[5] https://github.com/deepchem/deepchem/pull/1759
[6] https://github.com/deepchem/deepchem/pull/1763
[7] https://github.com/deepchem/deepchem/pull/1765