kevin-bretz/recsys-checkpoints

Recommender Systems checkpoints

The best trained checkpoints from my Recommender Systems coursework at Leiden University. Code and full details are in the GitHub repository kevin-bretz/recommender-systems.

Files

Path	Model	Notes
tiger/rqvae_K256_L3.pt	TIGER residual quantiser	Turns each item into a short semantic id (K=256, L=3, the default configuration).
tiger/transformer_dropout0.2.pt	TIGER generator	Generates the next item's semantic id token by token. Dropout 0.2 was the best setting in the ablation, reaching Recall@10 of 0.0352.
sasrec/sasrec_best.pt	SASRec	Self attentive sequential recommender, default configuration.

These are inference weights only, no dataset is included. See the GitHub repository for how to load them and which dataset each was trained on.