DataFreeOpen Source

DARTS

Time series forecasting made simple — from ARIMA to Transformers

Apache-2.0

ABOUT

Time series forecasting is fragmented across incompatible libraries — ARIMA in statsmodels, deep learning in PyTorch Forecasting, anomaly detection in Prophet — each with its own API, data format, and preprocessing conventions. Switching between models means rewriting data pipelines and evaluation code. Darts unifies all approaches under a single scikit-learn-style API. It supports statistical models (ARIMA, ETS, Prophet), neural networks (N-BEATS, TFT, TCN, Transformer), and ensemble methods. Data pipelines, backtesting, and hyperparameter tuning work identically whether you use Exponential Smoothing or a Transformer. It also includes built-in anomaly detection and probabilistic forecasting.

INSTALL

pip install darts

INTEGRATION GUIDE

1. Forecast product demand across thousands of SKUs using ensemble models that beat any single method 2. Detect anomalous server metrics by combining statistical baselines with deep learning detectors 3. Backtest multiple forecasting strategies (ARIMA vs N-BEATS vs TFT) with a single line-swap in the API 4. Build covariate-aware forecasts that incorporate weather, promotions, and economic indicators 5. Scale from daily retail forecasts to high-frequency financial predictions without changing the pipeline

DARTS

ABOUT

INTEGRATION GUIDE

TAGS