maxhodak / keras-molecules

Autoencoder network for learning a continuous representation of molecular structures.
MIT License
519 stars 146 forks source link

Add code for generator based training, testing, sampling #43

Closed pechersky closed 7 years ago

pechersky commented 7 years ago

The new files allow a user to skip preprocessing SMILES string output. Instead, the one-hot-matrices are generated on the fly.

This also allows less memory-heavy training (train_gen.py) and sampling (sample_gen.py), which addresses #35.