v1.5.1

🐛 Bug Fixes

• Fix TwoTower retrain (#494).

📚 Documentation

• Add RsCfBase documentation (#489).

Full Changelog: v1.5.0...v1.5.1

v1.5.0

🚀 Features

• Add Swing model (#465, #468, #476, #477, #478, #482).
• Add rust implementation of ItemCF (#438, #439, #452).

🐛 Bug Fixes

• Fix similarity matrix density computation (#446).
• Fix k setting (#470).
• Fix eps for latest scikit-learn version (#471).
• Fix slow computing in multi_sparse features (#475).

📚 Documentation

• Add rust models documentation (#483).

🧻 Miscellaneous

• Update CI dependencies (#441).
• Update python action (#448).
• Refactor RsUserCF (#456).
• Refactor similarity ordering (#463).

Full Changelog: v1.4.0...v1.5.0

v1.4.0

🚀 Features

• Add SIM model (#387).
• Support dual sequence (#386).
• Add rust implementation of UserCF (#400, #411).

🐛 Bug Fixes

• Fix seq_len dtype (#383).
• Fix all consumed items in negative sampling (#427).
• Fix missing k_sim in cf prediction (#431).
• Fix none similar recommendation in CF models (#435, #436).

📚 Documentation

• Add SIM doc (#401).

🧻 Miscellaneous

• Tensor bytes serialization (#376).
• Build wheels on maturin (#394).

Full Changelog: v1.3.0...v1.4.0

v1.3.0

🚀 Features

• Support multiple attention layers (#344, #345, #347).
• Add Transformer model (#360, #361, #363).
• Online computing serving in Rust (#331).
• Add gRPC service for Rust serving (#348).

🐛 Bug Fixes

• Fix disabling output when verbose <= 0 (#333).
• Fix request timeout in CI (#355).
• Fix missing random seeds (#358).
• Fix failing to consider positive samples in eval_data. (#366).
• Fix protobuf issue (#368). Users should choose protobuf version cautiously, see known issue.

📚 Documentation

• Update recommendation content (#336).
• Add Transformer doc (#362).

🧻 Miscellaneous

• Comply with Pydantic V2 on optional fields (#326).
• Refactor tf layers to facilitate the seamless integration of new algorithms (#337, #338, #339, #341, #354).
• Follow the RUF015 rule in ruff (#352).
• Remove deprecated sampling method (#353).

Full Changelog: v1.2.2...v1.3.0

v1.2.2

🐛 Bug Fixes

• Fix ALS rebuild model (#322).

Full Changelog: v1.2.1...v1.2.2

v1.2.1

🚀 Features

• Full inference for DGL models (#315).

🐛 Bug Fixes

• Saving all past behaviors during merging (#320).

🧻 Miscellaneous

• Lazy initialization of states in rust serving (#316).

Full Changelog: v1.2.0...v1.2.1

v1.2.0

🚀 Features

• Add embedding normalization option (#278).
• Add GAUC and Coverage metrics (#280).
• Add TwoTower algorithm (#285, #287, #289).
• Support dynamic embedding generation (#293, #295).
• Support serving with dynamic feature and sequence, see doc (#303).

🐛 Bug Fixes

• Fix variable sharing in TensorFlow1 (#286).
• Fix feature saving issue (#296).
• Fix masking issue in sequence construction (#310).

📚 Documentation

• Add online computing serving documentation (#308).

Full Changelog: v1.1.1...v1.2.0

v1.1.1

🐛 Bug Fixes

• Fix YouTubeRetrieval sequence recommend (#271).

🧻 Miscellaneous

• Add Python 3.11 support (#272).

Full Changelog: v1.1.0...v1.1.1

v1.1.0

💥 Breaking Changes

• Explicit negative sampling(build_negative_samples) has been deprecated. It will be removed in the future (#263).
• Negative sampling is controlled by the neg_sampling parameter (#254). See Negative sampling.
• merge_trainset will return a new data_info (#250).

🚀 Features

• Add dynamic sequence recommendation (#257).
• Add more sampling strategies (#239).
• Multiprocess data loading (#237).

🐛 Bug Fixes

• Fix dense layer activation order (#249).

📚 Documentation

• Add Colab notebook (#230).

🧻 Miscellaneous

• Add Windows & macOs CI (#246).

Full Changelog: v1.0.1...v1.1.0

v1.0.1

🐛 Bug Fixes

• Fix PyTorch device issue (#222).

Full Changelog: v1.0.0...v1.0.1