unit8co/darts

What it solves

Darts is designed to simplify time series forecasting and anomaly detection in Python. It provides a unified interface for a wide range of models, allowing users to easily switch between classic statistical methods and modern deep learning architectures without changing their core workflow.

How it works

The library introduces a TimeSeries object that standardizes how data is handled across different models. It uses a scikit-learn-like API (fit() and predict()) to ensure consistency. Darts supports both univariate and multivariate time series, as well as the integration of external data (past and future covariates) and static metadata to improve predictions.

Who it’s for

It is intended for data scientists and ML engineers who need to perform time series analysis, from simple baseline forecasts to complex deep neural networks, and those who need to build robust anomaly detection systems.

Highlights

• Diverse Model Suite: Includes everything from ARIMA and Exponential Smoothing to deep learning models like N-BEATS.
• Anomaly Detection: Dedicated module for anomaly scorers, detectors, and aggregators, including integration with PyOD.
• Global Model Training: Ability to train ML-based models on multiple time series simultaneously to scale to large datasets.
• Probabilistic Forecasting: Support for stochastic time series to generate confidence intervals and quantile predictions.
• Comprehensive Tooling: Built-in utilities for backtesting, data processing (scaling, filling missing values), and hierarchical forecast reconciliation.
• Infrastructure Integration: Deep learning models are powered by PyTorch Lightning and integrate with MLflow for experiment tracking.