search

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
MIT License
2k stars 429 forks source link

Electricity dataset #175

Closed khairulislam closed 1 year ago

khairulislam commented 1 year ago

Hi, I was looking through the datasets, but I found two issues.

  1. The electricity dataset uploaded in both the Goode drive and Tsinghua Cloud. In both locations, the electricity dataset is from 7/1/2016 and 7/2/2019, which is clearly not the UCI electricity dataset from 2011-2014.

  2. The UCI electricity dataset mentions 370 users, but the paper mentioned 321.

wuhaixu2016 commented 1 year ago

Hi, thanks for your question. （1）The time stamps in electricity is added by us and not correctly matched to the real world. We add this just for temporal embedding. Thus, the relative interval is essential but the exact value will not affact the final performance. （2）We obtain the data from this repo https://github.com/laiguokun/multivariate-time-series-data, which contains 321 variates.