ML Matrix Factorization recommender

Implementation of the winning recommender system from the Netflix competition. Uses matrix decomposition to derive a P and Q matrix which can be used to make predictions.
Uses gradient descent to arrive at the solution.

How to Use.

Predict using the ProductRecommender Class.
Can be used in two ways:

1. First: To recommend using only a users x product matrix along with a k value for # of features to discover.

    #import predictor
    from recommender.matrix_factor_model import ProductRecommender
    
    # fake data. (2 users, three products)
    user_1 = [1, 2, 3]
    user_2 = [0, 2, 3]
    data = [user_1, user_2]
    
    # train model
    modelA = ProductRecommender()
    modelA.fit(data)
    
    # predict for user 2 
    modelA.predict_instance(1)
    # prints array([ 0.9053102 ,  2.02257811,  2.97001565])

2. Second: To recommend when you want to provide a list of features for movies and only learn P (user -> feature strength).

    #import predictor
    from recommender.matrix_factor_model import ProductRecommender
    
    # fake data. (2 users, three products)
    user_1 = [1, 2, 3]
    user_2 = [0, 2, 3]
    data = [user_1, user_2]
    
    # product features (year made, height)
    product_1 = [2014, 74]
    product_2 = [2016, 89]
    Q = [product_1, product_2]
    
    # train model passing in Q
    modelB = ProductRecommender()
    modelB.fit(data, Q)
    
    # predict for user 2 
    modelB.predict_instance(1)
    # prints array([ 0.9053102 ,  2.02257811,  2.97001565])

Defaults

License

MIT

Author

William Falcon