Here is a quick summary of DeepChem updates for the 2nd week of November, including the following merged pull requests:
1. mypy fix for ferminet energy function: This PR addresses mypy-related issues in the Ferminet energy function, resolving a specific problem flagged by the mypy tool.
2. Miscellaneous methods for differentiation_utils: This PR Introduces miscellaneous methods for differentiation_utils as a new feature with associated documentation modifications.
3. CI Unit tests fix: This PR addresses and fixes dmpnn_featurizer unit tests for Python 3.10
4. Bumping torch 2.1.0 for dqc: This PR updates the torch version for DQC, ensuring compatibility for ferminet tests and considering a future move to the main branch
5. Fixes test on CI: This PR resolves failing CI tests related to featurizers, correcting issues arising from the numpy version 24, specifically addressing problems with tokenizers and updating tests accordingly.
6. BaseXC: This PR adds BaseXC, the base class for the exchange-correlation (XC) functional.
7. Added output_types in grover: This PR introduces output types in Grover, addressing a bug fix and making vocabulary parameters optional for finetuning applications
8. Ferminet: adding torch.func and changes to model input type: This PR utilizes torch.func’s hessian and jacobian calculation, modifies the model’s input type, and the forward method.
9. LightGBM - GBDT model earlystop bug fix: This PR fixes the early stopping bug in LightGBM GBDT models by adding callbacks for newer versions, with associated documentation modifications
10. GANModel Porting: This PR involves porting of GANModel to Torch, introducing a new feature.