# functime-org/functime

Time-series machine learning at scale. Built with Polars for embarrassingly parallel feature extraction and forecasts on panel data.

Repository: https://github.com/functime-org/functime
Canonical: https://ross.abutalabs.com/products/functime
Homepage: https://functime-org.github.io/functime/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: forecasting, machine-learning, python, time-series, polars, feature-engineering, panel-data
Last push: 2026-05-03T10:37:52+00:00

## Health v2 (maintenance only)
Score: 70/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 80, release rhythm 50, longevity 84
- inputs: {"age_days": 1185, "days_push": 122, "days_rel": 122, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1183, forks 65 (observed 2026-08-28T04:03:54.447098+00:00)

## What it is
functime is a Python library for production-ready global forecasting and time-series feature extraction on large panel datasets, built on lazy Polars transforms for parallelism. It includes preprocessing, cross-validation splitters, forecast metrics, automated hyperparameter tuning, and an LLM agent for forecast analysis.

## Use cases
- forecast 100,000 time series in seconds on a laptop
- extract tsfresh and Catch22 features from panel data in parallel
- backtest forecasts with expanding and sliding window splitters
- tune forecast hyperparameters and lags automatically with FLAML
- produce point and probabilistic forecasts with exogenous features
- preprocess time series with box-cox and differencing
- analyze and compare forecasts with an LLM agent

## When to choose
- you need fast forecasting across many related time series (panel data)
- you want Polars-based parallel feature engineering for time series
- you need an end-to-end forecasting pipeline with backtesting and metrics
- you want ML-based forecasters with exogenous feature support

## When to avoid
- you need classical statistical forecasting only (e.g. simple ARIMA on a single series)
- your stack is pandas-only and you cannot adopt Polars
- you need deep-learning forecasting models like N-BEATS or TFT

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, data-science, etl
- domain: machine-learning, data-science, time-series, analytics
- platform: python
- tags: time-series, forecasting, polars, feature-engineering, panel-data, backtesting, global-forecasting

## Member repositories
- functime-org/functime (main) score 70

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:54.447098+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:24:34.226641+00:00, confidence not recorded.
  - readme: https://github.com/functime-org/functime (fetched 2026-08-28T04:03:54.447098+00:00, sha ea69b270f3dc)
  - homepage: https://functime-org.github.io/functime/ (fetched 2026-08-29T12:31:46.543348+00:00, sha edeb3c043990)
- Data as of 2026-08-30T08:39:29.467469+00:00.