BPRMF Tutorial: Bayesian Personalized Ranking Matrix Factorization

BPRMF Tutorial: Bayesian Personalized Ranking Matrix Factorization#

Sandbox Model — Experimental

This model is part of the CoreRec Sandbox — an experimental collection of algorithms for research and exploration. Sandbox models are not production-tested and may have incomplete implementations. For production-ready models, see the Production Models.

Introduction#

BPRMF combines BPR ranking optimization with matrix factorization for implicit feedback collaborative filtering.

Paper: Rendle et al. 2009 - BPR-MF

How BPRMF Works#

Architecture#

Ranking-Optimized MF:

  1. Matrix Factorization: User × Item latent factors

  2. Pairwise Ranking: BPR loss on (user, pos_item, neg_item) triplets

  3. Sampling: Bootstrap sampling for efficiency

  4. Optimization: SGD with ranking gradients

Key Innovation: Standard MF with BPR loss instead of pointwise loss

Architecture:

[U×I Factors] → <p_u, q_i> - <p_u, q_j> → BPR Loss
where i = positive, j = negative

Mathematical Foundation#

Matrix Factorization:

x̂_ui = <p_u, q_i> = Σ_f p_uf · q_if

BPR-MF Loss:

L = -Σ_{(u,i,j)} log σ(x̂_ui - x̂_uj) + λ||Θ||²

Gradients:

∂L/∂p_u = σ(-x̂_uij)(q_j - q_i) + λ·p_u
∂L/∂q_i = σ(-x̂_uij)·p_u + λ·q_i
∂L/∂q_j = σ(-x̂_uij)·(-p_u) + λ·q_j

Reference Implementation#

Implementation Note

The code examples below are reference implementations for learning purposes. This is a sandbox model and has not been production-tested. Verify behavior thoroughly before using in production.

Step 1: Import and Load Data#

from corerec.sandbox.collaborative_full.bayesian_method_base.bprmf_base import BPRMF_base as BPRMF
from cr_learn import ml_1m
from sklearn.model_selection import train_test_split
import numpy as np

# Load dataset with CORRECT API
data = ml_1m.load()  # Returns dict with 'ratings', 'users', 'movies'
ratings_df = data['ratings']  # DataFrame with user_id, movie_id, rating, timestamp

print(f"Loaded {len(ratings_df)} ratings")

# Split data
train_df, test_df = train_test_split(ratings_df, test_size=0.2, random_state=42)

# Extract arrays for model
train_users = train_df['user_id'].values
train_items = train_df['movie_id'].values
train_ratings = train_df['rating'].values

test_users = test_df['user_id'].values
test_items = test_df['movie_id'].values
test_ratings = test_df['rating'].values

Step 2: Initialize Model#

model = BPRMF(
    name="BPRMF_Model",
    n_factors=64,
    learning_rate=0.05,
    reg=0.01,
    num_negatives=3,
    epochs=20,
    batch_size=256,
    verbose=True
)

print(f"Initialized {model.name}")

Step 3: Train#

model.fit(
    user_ids=train_users,
    item_ids=train_items,
    ratings=train_ratings
)

print("Training complete!")

Step 4: Predict#

# Single prediction
score = model.predict(user_id=1, item_id=100)
print(f"Predicted score: {score:.3f}")

# Batch predictions
test_predictions = model.batch_predict(list(zip(test_users[:100], test_items[:100])))

Step 5: Recommend#

# Get top-10 recommendations for user
user_id = 1
recommendations = model.recommend(
    user_id=user_id,
    top_k=10
)

print(f"Top-10 recommendations for User {user_id}:")
for rank, item_id in enumerate(recommendations, 1):
    print(f"  {rank}. Item {item_id}")

Step 6: Evaluate#

from sklearn.metrics import mean_squared_error
import numpy as np

# Predict all test ratings
test_pred = [model.predict(u, i) for u, i in zip(test_users, test_items)]
rmse = np.sqrt(mean_squared_error(test_ratings, test_pred))
print(f"Test RMSE: {rmse:.4f}")

Step 7: Save & Load#

# Save model
model.save('artifacts/bprmf_model')

# Load model
loaded = BPRMF.load('artifacts/bprmf_model')
test_score = loaded.predict(1, 100)
print(f"Loaded model prediction: {test_score:.3f}")

Key Takeaways#

When to Use BPRMF#

✅ Perfect For:

  • Implicit feedback ranking

  • Top-N recommendation

  • When you want simple MF with ranking

  • Cold-start items (better than pointwise)

  • Medium-scale CF (100K-10M interactions)

❌ Not For:

  • Explicit ratings (use SVD)

  • Need features (use neural models)

  • Very sparse data

  • Sequential patterns (use RNN)

Best Practices#

  1. Factors: 20-100 dimensions

  2. Learning Rate: 0.05-0.1 (higher than SVD)

  3. Regularization: 0.01-0.001

  4. Negative Samples: 1-5 per positive

  5. Sampling: Uniform or popularity-based

  6. Update Frequency: Shuffle every epoch

  7. Initialization: Small random ~N(0, 0.01)

  8. Convergence: Monitor AUC, not loss

Scaling & Production Considerations#

If you plan to move this sandbox model toward production use, consider:

  • Data Volume: Test with realistic dataset sizes; sandbox implementations may not be optimized for large-scale data.

  • Distributed Training: For datasets exceeding single-machine memory, consider wrapping the model in distributed training frameworks (e.g., PyTorch DDP, Horovod).

  • Hyperparameter Tuning: Use systematic search (Optuna, Ray Tune) rather than manual tuning. The defaults in sandbox models are not tuned for any specific domain.

  • Serving Latency: Profile inference time. Some architectures (e.g., attention-based) may need quantization or ONNX export for low-latency serving.

  • Monitoring: Implement A/B testing and online metrics (CTR, engagement, diversity) before replacing an existing system.

  • Validation: Write comprehensive unit tests and integration tests before deploying. Sandbox models do not have CI coverage.

Further Reading#

Contents