CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.18.0] - 21.02.2026

Added

• Two-stage candidate ranking system with CandidateRankingModel and supporting classes (CandidateGenerator, CandidateFeatureCollector, Reranker, CatBoostReranker, PerUserNegativeSampler) (#296)

Changed

• Used pm-implicit instead of implicit for Python>=3.10 to support Cuda 12.x (#298)

[0.17.0] - 03.09.2025

Added

• LiGR transformer layers from "From Features to Transformers: Redefining Ranking for Scalable Impact" (#295)

[0.16.0] - 27.07.2025

Added

• HSTU Model from "Actions Speak Louder then Words..." implemented in the class HSTUModel (#290)
• leave_one_out_mask function (rectools.models.nn.transformers.utils.leave_one_out_mask) for applying leave-one-out validation during transformer models training.(#292)
• logits_t argument to TransformerLightningModuleBase. It is used to scale logits when computing the loss. (#290)
• use_scale_factor argument to LearnableInversePositionalEncoding. It scales embeddings by the square root of their dimension — following the original approach from the "Attention Is All You Need" (#290)
• Optional context argument to recommend method of models and get_context function to rectools.dataset.context.py (#290)

Fixed

• [Breaking] Corrected computation of cosine distance in DistanceSimilarityModule(#290)
• Installation issue with cupy extra on macOS (#293)
• torch.dtype object has no attribute 'kind' error in TorchRanker (#293)

Removed

• [Breaking] Dropout module from IdEmbeddingsItemNet. This changes model behaviour during training, so model results starting from this release might slightly differ from previous RecTools versions even when the random seed is fixed.(#290)

[0.15.0] - 17.07.2025

Added

• extras argument to SequenceDataset, extra_cols argument to TransformerDataPreparatorBase, session_tower_forward and item_tower_forward methods to SimilarityModuleBase (#287)
• Support for resaving transformer models multiple times and loading trainer state (#289)

Fixed

• [Breaking] Now LastNSplitter guarantees taking the last ordered interaction in dataframe in case of identical timestamps (#288)

[0.14.0] - 16.05.2025

Added

• Python 3.13 support (#227)
• fit_partial implementation for transformer-based models (#273)
• map_location and model_params_update arguments for the function load_from_checkpoint for Transformer-based models. Use map_location to explicitly specify the computing new device and model_params_update to update original model parameters (e.g. remove training-specific parameters that are not needed anymore) (#281)
• get_val_mask_func_kwargs and get_trainer_func_kwargs arguments for Transformer-based models to allow keyword arguments in custom functions used for model training. (#280)

[0.13.0] - 10.04.2025

Added

• TransformerNegativeSamplerBase and CatalogUniformSampler classes, negative_sampler_type and negative_sampler_kwargs parameters to transformer-based models (#275)
• SimilarityModuleBase, DistanceSimilarityModule, similarity module to TransformerTorchBackbone parameters to transformer-based models similarity_module_type, similarity_module_kwargs (#272)
• out_dim property to IdEmbeddingsItemNet, CatFeaturesItemNet and SumOfEmbeddingsConstructor (#276)
• TransformerBackboneBase, backbone_type and backbone_kwargs parameters to transformer-based models (#277)
• sampled_softmax loss option for transformer models (#274)

[0.12.0] - 24.02.2025

Added

• CatalogCoverage metric (#266, #267)
• divide_by_achievable argument to NDCG metric (#266)

Changed

• Interactions extra columns are not dropped in Dataset.filter_interactions method #267

[0.11.0] - 17.02.2025

Added

• SASRecModel and BERT4RecModel - models based on transformer architecture (#220)
• Transfomers extended theory & practice tutorial, advanced training guide and customization guide (#220)
• use_gpu for PureSVD (#229)
• from_params method for models and model_from_params function (#252)
• TorchRanker ranker which calculates scores using torch. Supports GPU. #251
• Ranker ranker protocol which unify rankers call. #251

Changed

• ImplicitRanker rank method compatible with Ranker protocol. use_gpu and num_threads params moved from rank method to __init__. #251

[0.10.0] - 16.01.2025

Added

• ImplicitBPRWrapperModel model with algorithm description in extended baselines tutorial (#232, #239)
• All vector models and EASEModel support for enabling ranking on GPU and selecting number of threads for CPU ranking. Added recommend_n_threads and recommend_use_gpu_ranking parameters to EASEModel, ImplicitALSWrapperModel, ImplicitBPRWrapperModel, PureSVDModel and DSSMModel. Added recommend_use_gpu_ranking to LightFMWrapperModel. GPU and CPU ranking may provide different ordering of items with identical scores in recommendation table, so this could change ordering items in recommendations since GPU ranking is now used as a default one. (#218)

[0.9.0] - 11.12.2024

Added

• from_config, get_config and get_params methods to all models except neural-net-based (#170)
• fit_partial implementation for ImplicitALSWrapperModel and LightFMWrapperModel (#203, #210, #223)
• save and load methods to all of the models (#206)
• Model configs example (#207,#219)
• use_gpu argument to ImplicitRanker.rank method (#201)
• keep_extra_cols argument to Dataset.construct and Interactions.from_raw methods. include_extra_cols argument to Dataset.get_raw_interactions and Interactions.to_external methods (#208)
• dtype adjustment to recommend, recommend_to_items methods of ModelBase (#211)
• load_model function (#213)
• model_from_config function (#214)
• get_cat_features method to SparseFeatures (#221)
• LightFM Python 3.12+ support (#224)

Fixed

• Implicit ALS matrix zero assignment size (#228)

Removed

• Python 3.8 support (#222)

[0.8.0] - 28.08.2024

Added

• Debias mechanism for classification, ranking and auc metrics. New parameter is_debiased to calc_from_confusion_df, calc_per_user_from_confusion_df methods of classification metrics, calc_from_fitted, calc_per_user_from_fitted methods of auc and rankning (MAP) metrics, calc_from_merged, calc_per_user_from_merged methods of ranking (NDCG, MRR) metrics. (#152)
• nbformat >= 4.2.0 dependency to [visuals] extra (#169)
• filter_interactions method of Dataset (#177)
• on_unsupported_targets parameter to recommend and recommend_to_items model methods (#177)
• Use nmslib-metabrainz for Python 3.11 and upper (#180)

Fixed

• display() method in MetricsApp (#169)
• IntraListDiversity metric computation in cross_validate (#177)
• Allow warp-kos loss for LightFMWrapperModel (#175)

Removed

• [Breaking] assume_external_ids parameter in recommend and recommend_to_items model methods (#177)

[0.7.0] - 29.07.2024

Added

• Extended Theory&Practice RecSys baselines tutorial (#139)
• MetricsApp to create plotly scatterplot widgets for metric-to-metric trade-off analysis (#140, #154)
• Intersection metric (#148)
• PartialAUC and PAP metrics (#149)
• New params (tol, maxiter, random_state) to the PureSVD model (#130)
• Recommendations data quality metrics: SufficientReco, UnrepeatedReco, CoveredUsers (#155)
• r_precision parameter to Precision metric (#155)

Fixed

• Used rectools-lightfm instead of pure lightfm that allowed to install it using poetry>=1.5.0 (#165)
• Added restriction to pytorch version for MacOSX + x86_64 that allows to install it on such platforms (#142)
• PopularInCategoryModel fitting for multiple times, cross_validate compatibility, behaviour with empty category interactions (#163)

[0.6.0] - 13.05.2024

Added

• Warm users/items support in Dataset (#77)
• Warm and cold users/items support in ModelBase and all possible models (#77, #120, #122)
• Warm and cold users/items support in cross_validate (#77)
• [Breaking] Default value for train dataset type and params for user and item dataset types in DSSMModel (#122)
• [Breaking] n_factors and deterministic params to DSSMModel (#122)
• Hit Rate metric (#124)
• Python 3.11 support (without nmslib) (#126)
• Python 3.12 support (without nmslib and lightfm) (#126)

Changed

• Changed the logic of choosing random sampler for RandomModel and increased the sampling speed (#120)
• [Breaking] Changed the logic of RandomModel: now the recommendations are different for repeated calls of recommend methods (#120)
• Torch datasets to support warm recommendations (#122)
• [Breaking] Replaced include_warm parameter in Dataset.get_user_item_matrix to pair include_warm_users and include_warm_items (#122)
• [Breaking] Renamed torch datasets and dataset_type to train_dataset_type param in DSSMModel (#122)
• [Breaking] Updated minimum versions of numpy, scipy, pandas, typeguard (#126)
• [Breaking] Set restriction scipy < 1.13 (#126)

Removed

• [Breaking] return_external_ids parameter in recommend and recommend_to_items model methods (#77)
• [Breaking] Python 3.7 support (#126)

[0.5.0] - 22.03.2024

Added

• VisualApp and ItemToItemVisualApp widgets for visual comparison of recommendations (#80, #82, #85, #115)
• Methods for conversion Interactions to raw form and for getting raw interactions from Dataset (#69)
• AvgRecPopularity (Average Recommendation Popularity) to metrics (#81)
• Added normalized parameter to AvgRecPopularity metric (#89)
• Added EASE model (#107)

Changed

• Loosened pandas, torch and torch-light versions for python >= 3.8 (#58)

Fixed

• Bug in Interactions.from_raw method (#58)
• Mistakes in formulas for Serendipity and MIUF in docstrings (#115)
• Examples reproducibility on Google Colab (#115)

[0.4.2] - 01.12.2023

Added

• Ability to pass internal ids to recommend and recommend_to_items methods and get internal ids back (#70)
• rectools.model_selection.cross_validate function (#71, #73)

Changed

• Loosened lightfm version, now it's possible to use 1.16 and 1.17 (#72)

Fixed

• Small bug in LastNSplitter with incorrect i_split in info (#70)

[0.4.1] - 31.10.2023

Added

• LightFM wrapper inference speed benchmark (#60)
• iALS with features quality benchmark (#60)

Changed

• Updated attrs version (#56)
• Optimized inference for vector models with EUCLIDEAN distance using implicit library topk method (#57)
• Changed features processing example (#60)

[0.4.0] - 27.10.2023

Added

• MRR (Mean Reciprocal Rank) to metrics (#29)
• F1beta, MCC (Matthew correlation coefficient) to metrics (#32)
• Base Splitter class to construct data splitters (#31)
• RandomSplitter to model_selection (#31)
• LastNSplitter to model_selection (#33)
• Support for Python 3.10 (#47)

Changed

• Bumped implicit version to 0.7.1 (#45)
• Bumped lightfm version to 1.17 (#43)
• Bumped pylint version to 2.17.6 (#43)
• Moved nmslib from main dependencies to extras (#36)
• Moved lightfm to extras (#51)
• Renamed nn extra to torch (#51)
• Optimized inference for vector models with COSINE and DOT distances using implicit library topk method (#52)
• Changed initialization of TimeRangeSplitter (instead of date_range argument, use test_size and n_splits) (#53)
• Changed split infos key names in splitters (#53)

Fixed

• Bugs with new version of pytorch_lightning (#43)
• pylint config for new version (#43)
• Cyclic imports (#45)

Removed

• Markdown dependancy (#54)