bpr-spark

bpr-spark copied to clipboard

If you did not use checkpoint or anything else, persist may not cut the data lineage. then the data lineage would crash your jvm. See you.

Heermosi

model parameters updating procedure in your implementation seems wrong

2

In BPR the object function is to maximize the posterior probability, so the learning algorithm should be stochastic gradient **ascent**, that is to say we need **subtract** the regularized term...

wenmin-wu

Indexes for the matrices are not present. No string indexer in place.

trickster

meaning of paramter "NUM_OF_NEGATIVE_PER_IMPLICIT"

1

Hi, I have two questions. 1) I'm a little bit confused with the paramter "NUM_OF_NEGATIVE_PER_IMPLICIT". Is it used in bootstrap sampling procedure? Would you please point out the related formula...

kiminh