search

ngruver / llmtime

https://arxiv.org/abs/2310.07820
MIT License
628 stars 139 forks source link

Autoformer experiments #12

Open rajatsen91 opened 8 months ago

rajatsen91 commented 8 months ago

Hi, Thanks for the great repository. I could not find a run script for the datasets in the informer/autoformer papers. Are there plans to add them?

ngruver commented 7 months ago

Hi Rajat,

I do not currently have a plan for adding them. Essentially the script would be identical to the Monash experiment script (https://github.com/ngruver/llmtime/blob/main/experiments/run_monash.py) but pointed at these datasets (https://drive.google.com/drive/folders/1ZOYpTUa82_jCcxIdTmyr0LXQfvaM9vIy). The series in each dataset are predicted independently (not jointly as a multivariate forecasting problem), as described in the paper.

To run the baseline models, we used the original Autoformer (https://github.com/thuml/Autoformer) and FEDformer repos (https://github.com/MAZiqing/FEDformer) but modified the test set to be a single chunk (the last chunk in the time series).

Nate

rajatsen91 commented 7 months ago

Thanks Nate.