search

aspuru-guzik-group / Tartarus

A Benchmarking Platform for Realistic And Practical Inverse Molecular Design
https://arxiv.org/abs/2209.12487
66 stars 6 forks source link

Requesting reproducible code for GB-GA and REINVENT #10

Closed hyeok9855 closed 2 months ago

hyeok9855 commented 2 months ago

Hello,

I'd like to replicate the experiments in your paper, but some codes are missing. Could you upload a reproducible code for GB-GA and REINVENT that was presented in the original paper? For REINVENT, I've checked the provided notebook file, but it does not work properly, maybe due to #9.

Thanks!

akshat998 commented 2 months ago

Hi @hyeok9855,

The re-invent model code is available in the models directory and is set up as we used it in our experiments. You'll need to adjust the fitness functions to suit your specific benchmarks.

For the GB-GA model, I recommend checking out the original repository at https://github.com/jensengroup/GB_GA. We've adapted it by updating the initial dataset to one that's tailored for our needs and modifying the corresponding fitness function. You should find the code readily adaptable for your use. We plan to integrate our custom modifications into the main codebase in the future :)

cc @gkwt Best regards, Akshat

hyeok9855 commented 2 months ago

Thank you for your response.

I'll dig more and let you know if any other issues arise!

Thanks!

gkwt commented 2 months ago

For future reference, the notebook for REINVENT in models_jupyter folder also contains a link to a repo reinvent-benchmarking for further details.