Add next-item pipeline

[hieuddo]

Description

• add data.dataset.SequentialDataset to support new data formats:
  • "SIT": Session, Item, Timestamp
  • "SITJson": Session, Item, Timestamp, Json
  • "USIT": User, Session, Item, Timestamp
  • "USITJson": User, Session, Item, Timestamp, Json
• add data.reader parser for listed formats
• add eval_method.NextItemEvaluation for next-item evaluation
• add models.SPop baseline model: session-based popular

Related Issues

#543

Todo

• write tests for listed new data/reader/model
• add next-item datasets and corresponding examples (examples done locally without local datasets)

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).

[tqtg]

@hieuddo @lthoang please take a look at some initial comments and see if they make sense. Also, can we have an example for the SPop baseline so at least we can do sanity check while making changes?

[hieuddo]

I added Pop and SPop models example with yoochoose dataset and obtained these results:

	MRR	NDCG@10	NDCG@50	Recall@10	Recall@50	Train (s)	Test (s)
Pop	0.0011	0.0000	0.0012	0.0000	0.0061	0.1965	13.0280
SPop	0.2724	0.2860	0.2871	0.3290	0.3340	0.1968	13.1694

[tqtg]

@hieuddo @lthoang are we planning to add some unit tests for the new changes?

[hieuddo]

@hieuddo @lthoang are we planning to add some unit tests for the new changes?

Yes. Unit tests are in todo list

[saghiles]

General comment. I see some formatting changes. Let's stick to "black" formatting to remain consistent across the library.

[tqtg]

@hieuddo @lthoang this PR is getting very complicated to review. Let's not keep adding more features (e.g., iterators) to this until we need them for other models in the future PRs. If this is sufficient to run the SPop baseline and do evaluation, I would suggest to revise the current code carefully and merge this PR.

[lthoang]

@hieuddo @lthoang this PR is getting very complicated to review. Let's not keep adding more features (e.g., iterators) to this until we need them for other models in the future PRs. If this is sufficient to run the SPop baseline and do evaluation, I would suggest to revise the current code carefully and merge this PR.

@tqtg I keep some iterators that i think should be standard enough for accessing data in sequence level and user level. The non-standard iterator (i.e., input-output iterators) has been removed from the current PR.
We already have enough components for supporting any standard next-item recommendation model atm. Please spend some time to review the source code and merge it when ready.

[tqtg]

LGTM