mf

Most popular matrix factorization algorithms for collaborative filtering:

• SVD

svd = StochasticGradientDescent(iterations=1e5, factors=64, learning_rate=1e-4, alpha=1e-5)
svd.fit(user_to_item)

svd.similar_items(item_id=0, top_k=20)
svd.recommend(user_index=239, top_k=5)

• ALS

als = ALS(iterations=20, factors=64, confidence=40)
als.fit(user_to_item)

als.similar_items(item_id=0, top_k=20)
als.recommend(user_index=239, top_k=5)

• BPR

bpr = BPR(iterations=200, factors=64, learning_rate=1e-2, alpha=1e-5)
bpr.fit(user_to_item)

bpr.similar_items(item_id=0, top_k=20)
bpr.recommend(user_index=239, top_k=5)

• WARP

warp = WARP(iterations=50, factors=64, learning_rate=1e-3, alpha=1e-2, max_warp_sampled=100)
warp.fit(user_to_item)

warp.similar_items(item_id=0, top_k=20)
warp.recommend(user_index=239, top_k=5)