-
Install rdkit (chemical informatics, http://www.rdkit.org/docs/Install.html) using Anaconda.
-
Install pytorch. The project uses pytorch 0.3.1, later versions are not backward compatible and will not run.
-
First prepare data by running
python -m data data/zinc.smi
(Generate samples from pretrained VAE for testing the grammar model) -
To train your own VAE.
python -m vae --restore vae-pretrained --generate-samples
-
Train grammar_rnn model, the main paper contribution. This takes approximately ten minutes.
python -m grammar_model