Autoencoders, encoders and decoders, and generative adversarial networks form the basis of many modern generative ML/AI models. This project is a set of basic chemistry autoencoders and generative models that can be used as a starting point for building other ML models. Image models use a SMILES to image featurizer which embeds molecular information into a 4-channel image.
- A SMILES string autoencoder, using GRU layers.
- A 4-channel molecular graph image autoencoder using Dense layers.
- A 4-channel molecular graph image autoencoder using Convolutional layers.
- A 4-channel variational molecular graph image autoencoder using Convolutional layers.
- a 4 channel molecular graph image generative adversarial network using Convolutional layers.
- a 4 channel molecular graph image Wasserstein generative adversarial network with gradient penalty using Convolutional layers.
- a 4 channel molecular graph image Pixel CNN based on the Tensorflow distributions implementation.