mllf.cb.deepset_autoencoder

DeepSet Autoencoder for pretraining atom-level embeddings.

This implements Step 2 of the 4-step pretraining process: Build an autoencoder with symmetric encoder/decoder networks to learn compressed representations of atomic environments (AEV + charge).

Functions

create_autoencoder([input_dim, hidden_dim, ...])	Factory function to create a DeepSet autoencoder.
load_pretrained_atombondgnn(encoder_path[, ...])	Load a pretrained AtomBondGNN encoder from a checkpoint.
load_pretrained_deepset(encoder_path[, ...])	Factory function to load a pretrained DeepSet for inference.

Classes

AtomBondGNNAutoencoder([aev_length, ...])	Autoencoder for pretraining AtomBondGNN atom-level embeddings.
DeepSetAutoencoder([input_dim, hidden_dim, ...])	Complete autoencoder for pretraining DeepSet embeddings.
DeepSetDecoder([embedding_dim, hidden_dim, ...])	Decoder network that reconstructs atom features from embeddings.
DeepSetEncoder([input_dim, hidden_dim, ...])	Encoder network that compresses atom features to embedding space.
PretrainedDeepSet(encoder_path[, freeze_weights])	Pretrained DeepSet with sum-pooling for RGCN integration.