DeepChem Minutes 9/23/2021

bharath · October 1, 2021, 2:10am

DeepChem Minutes

India/Asia/Pacific Call

Date: September 23rd, 2021

Attendees: Bharath, Peter, Arun

Summary:

Bharath this week has been working on shepherding GSoC projects across the finish line.

Arun this week worked on a number of YouTube tutorials and read the MegNet paper for the model wishlist. Bharath suggested some time to review the paper might be useful.

Peter has been working primarily on OpenMM.

Arun asked where Bharath saw DeepChem going over the coming year. Bharath said he’d written up a few thread:

github.com/deepchem/deepchem

Making DeepChem Layers Multibackend

opened 05:30AM - 17 Sep 21 UTC

closed 11:52PM - 05 Apr 23 UTC

rbharath

Here's a current list of DeepChem layers and the backends for which they are ava…ilable | Layer Name | TensorFlow | Torch | Jax | |---|---|---|---| |InteratomicL2Distances|X|-|-| |GraphConv|X|-|-| |GraphPool|X|-|-| |GraphGather|X|-|-| |MolGANConvolutionalLayer|X|-|-| |MolGANAggregationLayer|X|-|-| |MolGANMultiConvolutionLayer|X|-|-| |MollGANEncoderLayer|X|-|-| |LSTMStep|X|-|-| |AttnLSTMEmbedding|X|-|-| |IterRefLSTMEmbedding|X|-|-| |SwitchedDropout|X|-|-| |WeightedLinearCombo|X|-|-| |CombineMeanStd|X|-|-| |Stack|X|-|-| |Variable|X|-|-| |VinaFreeEnergy|X|-|-| |NeighborLlist|X|-|-| |AtomicConvolution|X|-|-| |AlphaShareLayer|X|-|-| |SluiceLoss|X|-|-| |BetaShare|X|-|-| |ANIFeat|X|-|-| |GraphEmbedPoolLayer|X|-|-| |GraphCNN|X|-|-| |Highway|X|-|-| |WeaveLayer|X|-|-| |WeaveGather|X|-|-| |DTNNEmbedding|X|-|-| |DTNNStep|X|-|-| |DTNNGather|X|-|-| |DAGLayer|X|-|-| |DAGGather|X|-|-| |MessagePassing|X|-|-| |EdgeNetwork|X|-|-| |GatedRecurrentUnit|X|-|-| |SetGather|X|-|-| |Stem|X|-|-| |InceptionResnetA|X|-|-| |InceptionResnetB|X|-|-| |InceptionResnetC|X|-|-| |ReductionA|X|-|-| |ReductionB|X|-|-| |ScaleNorm|-|X|-| |MultiHeadedMATAttention|-|X|-| |MATEncoderLayer|-|X|-| |SublayerConnection|-|X|-| |PositionwiseFeedForward|-|X|-| |MATEmbedding|-|X|-| |MATGenerator|-|X|-| |Linear|-|-|X| Our layers are a powerful library of differentiable primitives that we should be able to leverage to build more complex differentiable programs, but the fact that layers are currently poorly documented and not available on multiple backends limits their usability. I'm raising this issue to help brainstorm some ideas to make layers more useful.

github.com/deepchem/deepchem

which one is the preferred API service for deepchem TorchModel?

opened 05:51AM - 23 Sep 21 UTC

closed 11:52PM - 05 Apr 23 UTC

biotech7

help wanted Contribution Welcome

## ❓ Questions & Help I want to deploy deepchem TorchModels on a server which …can be queried by users . there are many choices for deployment of torch models. e.g. - 1. apache+flask+wsgi+pytorch - 2. springboot+pytorch - 3. pytorch serve tools(https://github.com/pytorch/serve) - 4. many third-party SDKs or libraries... so what's the preferred strategy for serving a deepchem TorchModel ?

As a short answer, Bharath said he saw DeepChem becoming a general library for AI powered science.

Americas/Europe/Africa/Middle East

Date: September 24th, 2021

Attendees: Bharath, Tony, David, Seyone

Summary:

Tony is a former mechanical engineer who just started a master’s program at Florida Atlantic university.

David has been working on a bug that exists in dataset loading when there are no labels. He will try to submit a PR once he figures out how to solve it. Bharath asked about the status of the Swissprot PR [WIP] Swisprot database into molnet by davidRFB · Pull Request #2557 · deepchem/deepchem · GitHub. David said he would rebase the PR and do a second review.

Seyone has been working on getting one of the older PRs up. He’s working on removing the dual inheritance for the RobertaTokenizer.

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.