Darts is a Python library for user-friendly forecasting and anomaly detection on time series.
The library addresses the need to apply diverse forecasting approaches to time series data through a unified interface. It provides models ranging from classical statistical methods like ARIMA to deep neural networks, all accessible through consistent fit() and predict() methods similar to scikit-learn. The tool supports both univariate and multivariate time series, allows training on large datasets containing multiple series, and enables backtesting, model combination, and incorporation of external data. For anomaly detection, Darts wraps forecasting and filtering models to create detection systems, and integrates PyOD models to generate anomaly scores on time series.
Darts suits projects requiring flexible model selection across the forecasting spectrum without rewriting code for different algorithms. It works well for applications needing probabilistic forecasting capabilities, multivariate analysis, or systematic model comparison through backtesting. Teams already familiar with scikit-learn will find the API natural to adopt.
The project maintains active continuous integration with passing test workflows. Code quality is enforced through standardized formatting. Community engagement is supported through dedicated chat channels for user discussion and support.
