Created by: daming-lu

In PaddlePaddle's book chapter (Recommendation System), I tuned the PASS_NUM to be higher here

But the accuracy goes down... From ml-1m dataset, we can know the real rating from user 1 to movie 783 is 4

1::783::4::978824291

But the experimental result never went pass 3 ...

(Pdb) np.array(user_id)[0][0]
1
(Pdb) np.array(movie_id)[0][0]
783
NUM_PASS 
100
('inferred score: ', array([[2.1509817]], dtype=float32))

500
('inferred score: ', array([[2.3413014]], dtype=float32))

5000
('inferred score: ', array([[2.3629897]], dtype=float32))

500000
('inferred score: ', array([[2.8035383]], dtype=float32))

50000000 (because it is over 1M ???)
('inferred score: ', array([[2.038723]], dtype=float32))

1000000 (1M)
('inferred score: ', array([[2.015883]], dtype=float32))