SELFIES Example: Variational Auto Encoder (VAE) for chemistry

comparing SMILES, DeepSMILES, GrammarVAE and SELFIES representation using reconstruction quality, diversity and latent space validity as metrics of interest
- v0.1.0 -- 04. August 2019

information:

This is the original code used to generate the data in our paper.

It used a hand-written SELFIES encoding (Table2 of paper), and cannot easily be adapted to other situations. If you want to use the VAE, please see ../chemistryVAE.py

That file is connected with the selfies.encoder/selfies.decoder, and can be applied on general datasets. For more documentation, please look there.

 settings*.yml
     these four files contain the settings for values for the best model described in the paper

For comments, bug reports or feature ideas, please send an email to
mario.krenn@utoronto.ca and alan@aspuru.com