# thuml/Autoformer

About Code release for "Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting" (NeurIPS 2021), https://arxiv.org/abs/2106.13008

Repository: https://github.com/thuml/Autoformer
Canonical: https://ross.abutalabs.com/products/autoformer
Language: Jupyter Notebook
License: MIT
License Family: permissive
Topics: deep-learning, time-series
Last push: 2025-02-28T01:14:54+00:00

## Health v2 (maintenance only)
Score: 36/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 8, release rhythm 35, longevity 100
- inputs: {"age_days": 1773, "days_push": 552, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2481, forks 497 (observed 2026-08-28T04:06:55.292616+00:00)

## What it is
Autoformer is the official PyTorch implementation of the NeurIPS 2021 paper 'Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting'. It provides a deep decomposition architecture with a series-wise Auto-Correlation mechanism for long-term time series forecasting across benchmarks in energy, traffic, economics, weather, and disease domains.

## Use cases
- forecast long-term time series with transformers
- train a decomposition-based forecasting model on benchmark datasets
- predict energy load, traffic, or weather trends
- reproduce NeurIPS 2021 time series forecasting results
- compare Auto-Correlation attention against standard self-attention for forecasting
- run long-horizon forecasting experiments in PyTorch

## When to choose
- you need state-of-the-art long-term time series forecasting with a proven Transformer variant
- you want a research-grade PyTorch codebase with pre-processed benchmark datasets
- your data has strong trend and seasonality components suited to decomposition
- you want a model with log-linear complexity instead of standard self-attention

## When to avoid
- you need short-term forecasting, imputation, anomaly detection, or classification - use the authors' Time-Series-Library instead
- you need a production-ready forecasting service rather than research code
- you work outside Python/PyTorch ecosystems
- you need lightweight classical forecasting methods like ARIMA or Prophet

## Facets
- artifact type: library
- maturity: stable
- function: deep-learning, machine-learning
- domain: time-series, deep-learning, machine-learning
- platform: python
- tags: time-series-forecasting, transformer, auto-correlation, research-code, pytorch, neurips-2021

## Member repositories
- thuml/Autoformer (main) score 36

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:55.292616+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 2026-08-30T02:28:14.780283+00:00, confidence not recorded.
  - readme: https://github.com/thuml/Autoformer (fetched 2026-08-28T04:06:55.292616+00:00, sha 66898e182593)
- Data as of 2026-08-30T08:39:29.467469+00:00.