Add FPMC model and model selection for NextItemRecommender

[hieuddo]

Description

• Add FPMC model
• Model selection by val_score
• Model selection for GRU4Rec
• Fix Diginetica session-based data loader

On the Diginetica dataset, the original data format and purpose were for session-aware recommendation. So for each user, beside the leave-one-session-per-user eval/test session, we also have the training session for that user (e.g., HGRU4Rec may feed those training sessions into a user-RNN, feed the test session into session-RNN, and combine those two). So for session-based, to minimize the changes in data loader and iteration, we add new URLs to load val/test set specifically for session-based.

Related Issues

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).

[Copilot]

Pull request overview

Adds the FPMC (Factorizing Personalized Markov Chains) next-item model, introduces an opt-in best-on-validation model selection mechanism shared by sequential models (also wired into GRU4Rec), and switches the Diginetica load_val/load_test defaults to a clean session-based held-out split (with a mode="session-aware" opt-in to recover the previous cumulative files).

Changes:

• New FPMC model (cornac/models/fpmc/) with a Torch implementation, registration in cornac.models, README entry, and a Diginetica example.
• New shared val_score helper in cornac/models/seq_utils/selection.py and best/last model_selection plumbing in both FPMC and GRU4Rec.
• cornac/datasets/diginetica.py load_val/load_test gain a mode parameter (default 'session-based' pointing to new *_sbr.zip URLs); test expectations updated accordingly.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
README.md	Adds FPMC row to the models table.
cornac/datasets/diginetica.py	Adds mode arg to load_val/load_test; new _sbr URLs.
cornac/models/init.py	Exposes FPMC from the package.
cornac/models/fpmc/init.py	Re-exports FPMC.
cornac/models/fpmc/recom_fpmc.py	New FPMC recommender with training loop and best-on-val selection.
cornac/models/fpmc/fpmc.py	Torch module computing <UI,IU> + <IL,LI> + bias.
cornac/models/fpmc/requirements.txt	torch>=1.12.0.
cornac/models/gru4rec/recom_gru4rec.py	Adds model_selection, val_eval_every, val_k, val_metric; restores best state.
cornac/models/seq_utils/init.py	Re-exports val_score.
cornac/models/seq_utils/selection.py	Helper that delegates to next_item_evaluation.ranking_eval.
examples/fpmc_diginetica.py	Quick-start example for FPMC on Diginetica.
tests/cornac/datasets/test_diginetica.py	Updates expected sizes for the new session-based defaults and asserts session-aware sizes.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[hieuddo]

Some quick results for GRU4Rec and FPMC:

Validation

Model	MRR	NDCG@10	NDCG@20	Recall@10	Recall@20	Time (s)
SPop	0.2376	0.2490	0.2537	0.2926	0.3119	0.5376
GRU4Rec	0.3310	0.3714	0.3780	0.5080	0.5338	0.4983
FPMC	0.1834	0.2117	0.2236	0.3183	0.3666	0.3962

Test

Model	MRR	NDCG@10	NDCG@20	Recall@10	Recall@20	Train (s)	Test (s)
SPop	0.2344	0.2432	0.2474	0.2776	0.2943	0.0014	0.5280
GRU4Rec	0.3024	0.3421	0.3454	0.4749	0.4883	27.2797	0.5028
FPMC	0.1752	0.2005	0.2119	0.2977	0.3411	22.0645	0.3781