Train collaborative filtering model with PyTorch

Task

Develop and train ML model for personalized recommendations

Requirements

• Implement matrix factorization with PyTorch
• Use implicit feedback signals (views, clicks, time spent)
• Add cold-start handling for new users
• Implement A/B testing framework
• Model versioning with MLflow

Data Pipeline

• Extract user interaction data from PostgreSQL
• Feature engineering (user embeddings, item embeddings)
• Train/val/test split (70/15/15)
• Hyperparameter tuning with Optuna

Metrics

• Precision@K, Recall@K
• NDCG (Normalized Discounted Cumulative Gain)
• AUC-ROC

Deliverables

• Trained model artifacts
• Training notebooks
• Model card documentation

Estimated effort

2-3 weeks