AnjaliBankapur/mf-movielens20m-baseline

Matrix Factorization (BPR) - MovieLens 20M

This repository contains the trained Matrix Factorization baseline model used in a recommender systems case study on the MovieLens 20M dataset.

Model

• Model type: Matrix Factorization with BPR objective
• Final saved model: mf_bpr_model.pt

Dataset

• MovieLens 20M

Configuration

• Number of users: 138238
• Number of items: 20639
• Embedding dimension: 64
• Learning rate: 0.001
• Weight decay: 0.0
• Seed: 42
• Device: CPU

Training Summary

Training loss across 5 epochs:

• Epoch 1: 0.2113
• Epoch 2: 0.0996
• Epoch 3: 0.0849
• Epoch 4: 0.0754
• Epoch 5: 0.0689

Files

• mf_bpr_model.pt - final trained model
• config_mf_data.json - configuration/data settings
• mf_val_tuning_results.csv - validation tuning results