Informer2020

Why we included this project

Informer is the reference PyTorch implementation of the AAAI'21 best-paper architecture for long-sequence time-series forecasting. Its main contribution is ProbSparse attention, which keeps only the query-key pairs that carry real information instead of computing every score, so attention cost stays reasonable when inputs get long. That makes it the natural first stop for anyone forecasting far ahead on datasets like electricity, traffic, weather, or exchange rates. The repo ships the training and evaluation pipeline, experiment scripts, and Colab notebooks, so a small team can reproduce the published results and then swap in its own data. It is a research codebase rather than a maintained product, so plan to fork and adapt it to your pipeline instead of treating it as a drop-in API.