gt4sd / gt4sd-core Goto Github PK

Is your feature request related to a problem? Please describe.
Unify interface for prediction algorithm making use of the property predictor.

Describe alternatives you've considered

• use gt4sd.properties in gt4sd.algorithms.prediction (preferable even in case we go for namespace packaging)
• wrap gt4sd.algorithms.prediction in gt4sd.properties

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

I accept and sign the CLA to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

CLA signing should be automatically committed via github actions, but the workflow fails to clone the repo it seems:

https://github.com/GT4SD/gt4sd-core/actions/runs/2488206057

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Instead of using https://github.com/GT4SD/reinvent_models (tracked contribution here MolecularAI/Reinvent#41).
We should directly open the fork on the public repo: https://github.com/MolecularAI/reinvent-models (most likely not available at the time we went live)

Involving @Ashish13898 mostly for feedback on what is needed to implement this.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Currently, the RT supports only plain inference with user provided masked inputs like:

<QED>0.123|[C][MASK][C][O]

which is prohibitive, especially since we use SELFIES. Til now, scaffold decoration was impossible with selfies since smiles-selfies conversion mappings could not be retrieveid. This is fixed in SELFIES >=2.1.0, for details see aspuru-guzik-group/selfies#75.

Bumping selfies to >=2.1.0 requires an update of PaccMann models, so this issue is sequential to #38.

Currently on pypi we only publish a source distribution, and it's not containing the package_data. It's there when building a wheel, e.g. with python setup.py bdist_wheel.

While building a wheel is something we should anyway do, we also need to add a MANIFEST.in file.
Edit: it should be possible without MANIFEST.in

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

I accept and sign the CLA to contribute to GT4SD.

Is your feature request related to a problem? Please describe.

No problem, just an update of MoLeR dependency: https://github.com/microsoft/molecule-generation/releases/tag/v0.2.0.

I accept and sign the CLA to contribute to GT4SD.

If something is wrong in the GT4SDConfiguration (host name, secure setup) the Client enters in a loop and tries to re-connect for a long time without any error/log message. Maybe we should have a smaller max number of trials (~10/100) or a log message every time the client has to re-connect.

I accept the CLA and I would like to sign it to contribute to GT4SD.

All PaccMann based models, e.g. proteomic and omics based generator should be retrained using the pypi-distribution of pytoda. This will enable all paccmann-related repos in vcs_requirements.txt to be pip installable and could free us almost entirely from the vcs requirements. This will affect inference of PaccMannRLProteinBasedGenerator and PaccMannRLOmicBasedGenerator

The new multiprocess library version (0.70.13) gives problems when installing gt4sd-core using the development mode. I had to set multiprocess==0.70.12.2 to install the library.

I accept the CLA and I would like to sign it to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

I accept the CLA and I would like to sign it to contribute to GT4SD.

Currently predictive models such as PaccMann, adhere to the GenerativeAlgorithm interface. This is not an ideal fit, we should consider adding a dedicated interface.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

https://github.com/microsoft/molecule-generation/releases/tag/v0.2.0

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

Hey @C-nit I tried merging the PR for the bdist_wheel building and it fails with the error below.

Could you take a look? Maybe wheel is missing from the image running the action.

Currently, our TorchDrug interface does support property optimization conceptually.

However, most runs will fail due to an underlying bug in TorchDrug that raises whenever there is only invalid SMILES in a batch, see DeepGraphLearning/torchdrug#83

Once this is fixed in torchdrug and a new version is released on conda, we can enable our unittests (already written): https://github.com/gt4sd/gt4sd-core/blob/master/src/gt4sd/training_pipelines/tests/test_training_torchdrug_gcpn.py#L76

I accept and sign the CLA to contribute to GT4SD.

I accept and sign the CLA to contribute to GT4SD.

Contributor License Agreement

• I accept and sign the contributor license agreement (CLA) to contribute to GT4SD.

I accept and sign the CLA to contribute to GT4SD.