Add next-basket recommendation evaluation method

[lthoang]

Description

Next-basket recommendation takes history baskets (e.g., a sequence of baskets that have been purchased) as input to predict what the next-basket (a set of items) is.
To support modeling basket data, we must allow modeling repeating items.
To support next-basket recommendation, it should be a NextBasketRecommender with following function implemented:

• fit()
• score()

TODO:

The conventional metrics for evaluating next-basket recommendation are Recall, NDCG, and PHR. This paper also evaluate performance on repeating/exploring items separately.

Related Issues

#543

Checklist:

• I have added tests.
• I have updated the documentation accordingly.
• I have updated README.md (if you are adding a new model).
• I have updated examples/README.md (if you are adding a new example).
• I have updated datasets/README.md (if you are adding a new dataset).

[tqtg]

@lthoang Can we have a short description on the philosophy behind this? And some best practices if any.

[tqtg]

Overall comments:

• It seems that we can avoid a lot of redundancy by inheriting existing base classes. This will help future development and maintenance.
• Please also think about how this fits in with next-item recommendation. Can we have one evaluation method that supports both scenarios or we need to separate them?
• For the GPTop baseline, it's good to have it in the model list and give a link to paper where we can find description for better reference.

@lthoang @hieuddo

[tqtg]

@lthoang one idea, how about maintaining UIR tuple as it is and keeping basket ids as another array like timestamps? It will be backward compatible with the current Dataset, and we could use UBIT datasets for other not-next-basket-rec models.

[tqtg]

@lthoang please consider some comments above. In addition, I refactored NextBasketRecommender and NextBasketEvaluation, mainly to simplify things but not the logic. Below is the result running gp_top_tafeng.py example, please double check if that's something we should expect.

VALIDATION:
...
      | NDCG@10 | NDCG@20 | NDCG@50 | Recall@10 | Recall@20 | Recall@50 | Time (s)
----- + ------- + ------- + ------- + --------- + --------- + --------- + --------
PTop  |  0.0062 |  0.0064 |  0.0067 |    0.0090 |    0.0097 |    0.0109 |  18.2922
GTop  |  0.0228 |  0.0278 |  0.0359 |    0.0367 |    0.0553 |    0.0932 |  88.7690
GPTop |  0.0239 |  0.0290 |  0.0373 |    0.0410 |    0.0602 |    0.0991 |  94.4805

TEST:
...
      | NDCG@10 | NDCG@20 | NDCG@50 | Recall@10 | Recall@20 | Recall@50 | Train (s) | Test (s)
----- + ------- + ------- + ------- + --------- + --------- + --------- + --------- + --------
PTop  |  0.0124 |  0.0133 |  0.0137 |    0.0192 |    0.0224 |    0.0240 |    0.0003 |  24.1674
GTop  |  0.0264 |  0.0326 |  0.0425 |    0.0385 |    0.0597 |    0.1024 |    0.0774 | 114.1030
GPTop |  0.0295 |  0.0363 |  0.0467 |    0.0483 |    0.0717 |    0.1168 |    0.0783 | 122.7305

[tqtg]

@lthoang I did some changes to reuse the code better. We're almost there. Please check if the following result looks fine.

VALIDATION:
...
      | NDCG@10 | NDCG@20 | NDCG@50 | Recall@10 | Recall@20 | Recall@50 | Time (s)
----- + ------- + ------- + ------- + --------- + --------- + --------- + --------
PTop  |  0.0494 |  0.0525 |  0.0572 |    0.0438 |    0.0578 |    0.0712 |   2.6300
GTop  |  0.0492 |  0.0549 |  0.0703 |    0.0398 |    0.0587 |    0.1011 |  10.3374
GPTop |  0.0672 |  0.0767 |  0.0951 |    0.0594 |    0.0908 |    0.1414 |  11.4246

TEST:
...
      | NDCG@10 | NDCG@20 | NDCG@50 | Recall@10 | Recall@20 | Recall@50 | Train (s) | Test (s)
----- + ------- + ------- + ------- + --------- + --------- + --------- + --------- + --------
PTop  |  0.0489 |  0.0528 |  0.0586 |    0.0436 |    0.0589 |    0.0742 |    0.0003 |   6.0762
GTop  |  0.0497 |  0.0561 |  0.0722 |    0.0394 |    0.0596 |    0.1046 |    0.0767 |  25.9023
GPTop |  0.0697 |  0.0788 |  0.0987 |    0.0628 |    0.0915 |    0.1460 |    0.0769 |  28.5510

[lthoang]

@lthoang I did some changes to reuse the code better. We're almost there. Please check if the following result looks fine.

VALIDATION:
...
      | NDCG@10 | NDCG@20 | NDCG@50 | Recall@10 | Recall@20 | Recall@50 | Time (s)
----- + ------- + ------- + ------- + --------- + --------- + --------- + --------
PTop  |  0.0494 |  0.0525 |  0.0572 |    0.0438 |    0.0578 |    0.0712 |   2.6300
GTop  |  0.0492 |  0.0549 |  0.0703 |    0.0398 |    0.0587 |    0.1011 |  10.3374
GPTop |  0.0672 |  0.0767 |  0.0951 |    0.0594 |    0.0908 |    0.1414 |  11.4246

TEST:
...
      | NDCG@10 | NDCG@20 | NDCG@50 | Recall@10 | Recall@20 | Recall@50 | Train (s) | Test (s)
----- + ------- + ------- + ------- + --------- + --------- + --------- + --------- + --------
PTop  |  0.0489 |  0.0528 |  0.0586 |    0.0436 |    0.0589 |    0.0742 |    0.0003 |   6.0762
GTop  |  0.0497 |  0.0561 |  0.0722 |    0.0394 |    0.0596 |    0.1046 |    0.0767 |  25.9023
GPTop |  0.0697 |  0.0788 |  0.0987 |    0.0628 |    0.0915 |    0.1460 |    0.0769 |  28.5510

The result looks fine. I have update the example. The new result as follows:

VALIDATION:
...
      | HitRatio@10 | HitRatio@50 | NDCG@10 | NDCG@50 | Recall@10 | Recall@50 | Time (s)
----- + ----------- + ----------- + ------- + ------- + --------- + --------- + --------
PTop  |      0.3701 |      0.5000 |  0.0836 |  0.0997 |    0.0779 |    0.1290 |   1.7190
GTop  |      0.2606 |      0.5121 |  0.0562 |  0.0791 |    0.0438 |    0.1105 |   6.8573
GPTop |      0.4003 |      0.6722 |  0.0917 |  0.1269 |    0.0861 |    0.1869 |   7.2663

TEST:
...
      | HitRatio@10 | HitRatio@50 | NDCG@10 | NDCG@50 | Recall@10 | Recall@50 | Train (s) | Test (s)
----- + ----------- + ----------- + ------- + ------- + --------- + --------- + --------- + --------
PTop  |      0.3533 |      0.4962 |  0.0794 |  0.0953 |    0.0689 |    0.1211 |    0.0004 |   4.4665
GTop  |      0.2418 |      0.4820 |  0.0540 |  0.0746 |    0.0402 |    0.1013 |    0.0812 |  17.1021
GPTop |      0.3802 |      0.6500 |  0.0857 |  0.1181 |    0.0784 |    0.1740 |    0.0807 |  18.1752

[tqtg]

@lthoang LGTM