Functime is a Python library designed for large-scale time-series machine learning, specifically targeting panel data. It offers efficient feature extraction and forecasting capabilities, aiming to provide production-ready solutions for users dealing with numerous time series.

How It Works

Functime leverages Polars for its core operations, enabling embarrassingly parallel processing of time-series data. This approach allows for rapid feature engineering and forecasting across potentially hundreds of thousands of individual time series. The library integrates various time-series preprocessing techniques, cross-validation strategies (expanding and sliding windows), and forecast metrics, all optimized as lazy Polars transformations.

Quick Start & Requirements

• Install via pip: pip install functime
• Optional features: pip install "functime[llm,lgb]" for LLM and LightGBM support.
• Prerequisites: Python. No specific hardware or GPU requirements are mentioned for basic functionality.
• Documentation: Full walkthrough on forecasting, Feature Extraction examples.

Highlighted Details

• Processes 100,000+ time series in seconds on a laptop.
• Offers over 100 time-series feature extractors accessible via a custom Polars ts namespace.
• Includes an LLM agent for analyzing, describing, and comparing forecasts.
• Supports exogenous features and automated lag/hyperparameter tuning with FLAML.

Maintenance & Community

• Active development with a Discord community available at https://discord.com/invite/JKMrZKjEwN.
• CI checks for quickstart workflows are present.

Licensing & Compatibility

• Distributed under the Apache-2.0 license.
• Permissive license suitable for commercial use and integration into closed-source projects.

Limitations & Caveats

The library is focused on panel data and may not be optimized for single, very long time series. While it claims "battle-tested" algorithms, specific benchmark results against other libraries are not detailed in the README.