Automated Time Series Forecasting (AutoTS) is a Python package designed for the rapid deployment of high-accuracy forecasts at scale. It addresses the challenge of selecting optimal forecasting models, preprocessing techniques, and ensembling strategies for diverse time series datasets. AutoTS benefits users by automating these complex decisions through advanced AutoML, enabling efficient and robust forecasting solutions, as evidenced by its top performance in the 2023 M6 forecasting competition.

How It Works

AutoTS employs an AutoML approach driven by genetic algorithms to discover the best combination of forecasting models, transformers, and ensembling methods for a given dataset. It integrates dozens of sklearn-style forecasting models (including naive, statistical, machine learning, and deep learning) and over 30 time series transformers, all operating directly on Pandas DataFrames. Novel ensemble types, such as horizontal and mosaic, are utilized to maximize accuracy per series while maintaining scalability.

Quick Start & Requirements

• Installation: pip install autots
• Prerequisites: Basic models are included; additional packages may be required for some advanced features. Data is expected in Pandas DataFrame 'wide' (DatetimeIndex, columns as series) or 'long' (Date, Series ID, Value columns) formats.
• Docs: Extended tutorial (extended_tutorial.md) and production example (production_example.py) are available.

Highlighted Details

• Achieved top performance in the 2023 M6 forecasting competition for stock market forecasting.
• Supports multivariate and probabilistic (upper/lower bound) forecasts.
• Scales effectively to tens or hundreds of thousands of input series.
• Automated search for optimal models, preprocessing, and ensembling via genetic algorithms.
• Flagship ensembling types include horizontal and mosaic styles.

Maintenance & Community

The project encourages contributions via GitHub Issues for bug reports, feature requests, and documentation feedback. It also invites users to share model templates and search parameters. Specific community channels (e.g., Discord, Slack) or maintainer details are not explicitly provided in the README.

Licensing & Compatibility

The provided README does not explicitly state the software license. Users should verify licensing terms before adoption, especially for commercial use or integration into closed-source projects.

Limitations & Caveats

Memory shortages are noted as the primary cause of process crashes. Performance can be tuned via model/transformer lists and ensemble strategies, potentially involving accuracy trade-offs for speed. NaN handling is more efficient if data is pre-processed before input. Environment-specific setup, such as underlying BLAS libraries, can influence stability and performance.