The Rubix ML library for PHP has released versions 3.0, 3.1, and 3.2, introducing significant updates to performance, neural network training, data processing, and parallel computing. The project has also launched an official website to provide documentation and examples for the full machine learning lifecycle.
A key feature in version 3.0 is the introduction of parallel inference for specific algorithms. K-Nearest Neighbors and KNN Regressor now support parallel processing, while Isolation Forest supports parallel execution for both training and inference. These features require configuring a backend such as Amp or Swoole.
The updates include new tools for neural network training, such as gradient accumulation, gradient clipping, and fine-tuning with layer freezing. Additionally, the library has added the K-medoids clustering algorithm, new data extractors like ARFF and Shuffler, and new image transformations.
The library now automatically treats integer values as categorical types, which is a change from previous versions. Developers using the library for production tasks must use PHP 8.3 or newer. The official website serves as a central hub for all resources related to the toolkit.
