aksub99 / molecular-vae Goto Github PK

Pytorch implementation of the paper "Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules"

License: MIT License

Jupyter Notebook 100.00%

cheminformatics pytorch materials-science materials-informatics python deep-learning variational-autoencoder molecular-structures machine-learning drug-discovery chemistry natural-language-processing jupyter-notebook

molecular-vae's Issues

Why reparameterize in your code multipy 1e-2?

I remebered there is no 1e-2 in vae

How to prepare data

Hi, thanks for putting this together. In your notebook with an example, I see that you are using the dataset you already processed. I wondered if you could list what the procedure to follow is. I would like to use your implementation.

Best,

Mention code contribution and a potential bug fix.

want to point out that this repo should cite the original code contribution from topazape/molecular-VAE and also cxhernandez/molencoder so that their contributions are clearly stated because what you did is essentially just reusing their code. I think this project is better started as a fork as supposed to a fresh repo.

I have also fixed a bug on conv1d layer that existed in topazape/molecular-VAE that is also carried over to this repo, you can check it out here: topazape/molecular-VAE#2. That should also help you with further minimizing the loss.

Recommend Projects