How unit8co/darts works
Darts is a Python library for representing, transforming, forecasting, and detecting anomalies in time series. A typical workflow creates or loads a TimeSeries, optionally transforms it or adds encoded covariates, fits a forecasting model, and predicts or backtests; alternatively, anomaly scorers produce scores that detectors classify and aggregators combine. Forecasting includes statistical, machine-learning, neural, probabilistic, and conformal models. Additional branches support filtering, metrics, explanations, and ONNX inference. The graph reflects the README and sampled sources; unsampled subsystem wiring is omitted where calls cannot be verified.
Main entry points: pipeline.py Transform pipeline · anomaly_model Anomaly wrappers · forecasting Forecasting models · inference.py ONNX inference
Components 19
Series and Preparation
- TimeSeries
darts/timeseries.py - Dataset loaders
darts/datasets/datasets.py - Transformers
darts/dataprocessing/transformers - Transform pipeline
darts/dataprocessing/pipeline.py - Covariate encoders
darts/dataprocessing/encoders/encoders.py
Forecasting
- Forecasting models
darts/models/forecasting - Forecasting base
darts/models/forecasting/forecasting_model.py - Inference datasets
darts/utils/data/torch_datasets/inference_dataset.py - Filtering models
darts/models/filtering
Anomaly Detection
- Anomaly scorers
darts/ad/scorers - Anomaly detectors
darts/ad/detectors - Score aggregators
darts/ad/aggregators - Anomaly wrappers
darts/ad/anomaly_model
Evaluation and Analysis
- Forecast metrics
darts/metrics/metrics.py - Model explanations
darts/explainability
Model Interoperability
- ONNX export
darts/utils/onnx/export.py - ONNX inference
darts/utils/onnx/inference.py
Other components
- Library user
- ONNX Runtime
Connections 23
- Library user → Dataset loaders: loads data
- Dataset loaders → TimeSeries: returns series
- Library user → TimeSeries: creates series
- Library user → Transform pipeline: transforms series
- Transform pipeline → Transformers: applies transforms
- Transform pipeline → TimeSeries: transforms data
- Library user → Covariate encoders: generates covariates
- Library user → Forecasting models: fits and predicts
- Forecasting models → Forecasting base: inherits behavior
- Forecasting base → Inference datasets: builds inference data
- Library user → Filtering models: filters series
- Library user → Anomaly scorers: scores series
- Library user → Anomaly detectors: classifies scores
- Library user → Score aggregators: combines scores
- Library user → Anomaly wrappers: runs anomaly model
- Anomaly wrappers → Forecasting models: uses forecasters
- Anomaly wrappers → Filtering models: uses filters
- Library user → Forecast metrics: evaluates forecasts
- Library user → Model explanations: explains model
- Library user → ONNX export: exports model
- Library user → ONNX inference: runs inference
- ONNX inference → ONNX Runtime: runs session
- ONNX Runtime → ONNX inference: returns outputs