Pycaret, an open-source, low-code machine learning library, recently released a new update – PyCaret 3.0. The new update includes several new features and improvements. The library was created by Moez Ali, who is currently serving as product director for artificial intelligence at antuit.ai.
The library, which automates machine learning workflows and makes experiments exponentially fast and efficient, now includes stable Time Series Forecasting, a new Object-Oriented API, more options for Experiment Logging, Refactored Preprocessing Module, Compatibility with the latest scikit-learn version, Distributed Parallel Model Training, and Accelerated Model Training on CPU.
Key Features
The Time Series module, which is now stable, currently supports forecasting tasks, with plans to add time-series anomaly detection and clustering algorithms in the future. The Object-Oriented API allows the effortless conducting of multiple experiments in the same notebook, with parameters linked to an object and associated with various modelling and preprocessing options. The Experiment Logging feature now includes new options, such as wandb, cometml, and dagshub, alongside the default MLflow.
PyCaret 3.0 includes several new preprocessing functionalities, such as innovative categorical encoding techniques, support for text features in machine learning modelling, novel outlier detection methods, and advanced feature selection techniques. The library now guarantees to avoid target leakage as the entire pipeline is fitted at a fold level.
Moreover, PyCaret 3.0 is compatible with the latest scikit-learn version (1.X), making it easier to use both libraries simultaneously in the same environment. Distributed Parallel Model Training and Accelerated Model Training on the CPU also improve the library’s performance, making it a much more productive tool for citizen data scientists, and power users who can perform simple and moderately sophisticated analytical tasks with ease.