BPR-pytorch/result.out at master · bob-lytton/BPR-pytorch · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
experiment1: batch size=4096, epochs=50, lambda=0.001, lr=0.01, num_factors=16
    End. Best epoch 049: HR = 0.850,        NDCG = 0.539

experiment2: batchsize=1024, epochs=50, lambda=0.001, lr=0.005, num_factors=32
    End. Best epoch 049: HR = 0.812,        NDCG = 0.494

experiment3: batchsize=4096, epochs=100, lambda=0.001, lr=0.01, num_factors=32
    HR: 0.861       NDCG: 0.562

experiment4: batchsize=4096, epochs=60, lambda=0.01, lr=0.01, num_factors=10
    HR: 0.838       NDCG: 0.522

experiment5: batchsize=4096, epochs=50, lambda=0.01, lr=0.01, num_factors=32
    End. Best epoch 048: HR = 0.849,        NDCG = 0.532

experiment6: batchsize=4096, epochs=33, lambda=0.1, lr=0.01, num_factors=32, optimizer=Adam
    HR: 0.736       NDCG: 0.422

experiment7: batchsize=4096, epochs=12, lambda=0.1, lr=0.001, num_factors=32, optimizer=Adam
    HR: 0.741       NDCG: 0.422

experiment8: batchsize=4096, epochs=100, lambda=0.01, lr=0.001, num_factors=32, optimizer=Adam
    End. Best epoch 092: HR = 0.861,        NDCG = 0.551

experiment9: batchsize=4096, epochs=100, lambda=0.01, lr=0.001, num_factors=64, optimizer=Adam
    End. Best epoch 080: HR = 0.866,        NDCG = 0.553

the experiments above all used only 1 positive sample for each user in test
below will use multiple positive samples and negative samples.

experiment10: batchsize=4096, epochs=83, lambda=0.01, lr=0.001, num_factors=64, optimizer=Adam
    PREC: 0.411     HR: 0.713       NDCG: 0.770

experiment10: batchsize=4096, epochs=83, lambda=0.01, lr=0.001, num_factors=64, optimizer=Adam