How amazon-science/chronos-forecasting works
Chronos is a Python package for probabilistic time-series forecasting, with original Chronos, Chronos-Bolt, and Chronos-2 model families. Users provide series (optionally in DataFrames with covariates), choose a pretrained pipeline, and receive forecasts; Chronos-2 additionally supports fine-tuning and multivariate or covariate-informed forecasting. The repository also includes model training and benchmark-evaluation scripts. Relationships below follow the excerpts, SOURCE INDEX references, or README; unsampled components are represented only where their purpose is documented or named.
Main entry points: pipeline.py Chronos-2 Pipeline · base.py Pipeline Base · chronos.py Chronos Pipeline · model.py Chronos-2 Model
Components 16
Forecasting API
- Chronos-2 Pipeline
src/chronos/chronos2/pipeline.py - Pipeline Base
src/chronos/base.py - DataFrame Utilities
src/chronos/df_utils.py - Forecast Utilities
src/chronos/utils.py - AWS Utilities
src/chronos/boto_utils.py
Model Families
- Chronos Pipeline
src/chronos/chronos.py - Chronos-Bolt
src/chronos/chronos_bolt.py - Chronos-2 Model
src/chronos/chronos2/model.py - Chronos-2 Config
src/chronos/chronos2/config.py - Chronos-2 Layers
src/chronos/chronos2/layers.py
Chronos-2 Data
- Input Preprocessing
src/chronos/chronos2/preprocess.py - Chronos-2 Dataset
src/chronos/chronos2/dataset.py
Training and Evaluation
- Chronos-2 Trainer
src/chronos/chronos2/trainer.py - Model Training
scripts/training/train.py - Benchmark Evaluation
scripts/evaluation/evaluate.py
Other components
- Forecast User
Connections 19
- Forecast User → Chronos-2 Pipeline: requests forecasts
README.md - Chronos-2 Pipeline → Input Preprocessing: prepares inputs
src/chronos/chronos2/pipeline.py - Chronos-2 Pipeline → Chronos-2 Dataset: builds datasets
src/chronos/chronos2/pipeline.py - Chronos-2 Pipeline → Chronos-2 Model: runs model
src/chronos/chronos2/pipeline.py - Chronos-2 Pipeline → Pipeline Base: uses base
src/chronos/chronos2/pipeline.py - Chronos-2 Pipeline → DataFrame Utilities: uses utilities
src/chronos/chronos2/pipeline.py - Chronos-2 Pipeline → Forecast Utilities: uses utilities
src/chronos/chronos2/pipeline.py - Chronos-2 Dataset → Input Preprocessing: converts inputs
src/chronos/chronos2/dataset.py - Chronos-2 Model → Chronos-2 Config: reads config
src/chronos/chronos2/model.py - Chronos-2 Model → Chronos-2 Layers: uses layers
src/chronos/chronos2/model.py - Chronos Pipeline → Pipeline Base: extends base
src/chronos/chronos.py - Chronos Pipeline → Forecast Utilities: uses utilities
src/chronos/chronos.py - Chronos-Bolt → Pipeline Base: extends base
src/chronos/chronos_bolt.py - Pipeline Base → DataFrame Utilities: uses utilities
src/chronos/base.py - Pipeline Base → Forecast Utilities: uses utilities
src/chronos/base.py - Pipeline Base → AWS Utilities: uses AWS helpers
src/chronos/base.py - Chronos-2 Trainer → Chronos-2 Dataset: trains on
src/chronos/chronos2/trainer.py - Model Training → Chronos Pipeline: imports model
scripts/training/train.py - Benchmark Evaluation → Chronos Pipeline: imports model
scripts/evaluation/evaluate.py