search

cure-lab / SCINet

The GitHub repository for the paper: “Time Series is a Special Sequence: Forecasting with Sample Convolution and Interaction“. (NeurIPS 2022)
Apache License 2.0
617 stars 127 forks source link

Long Term Prediction implementation #44

Closed bendoesai closed 2 years ago

bendoesai commented 2 years ago

I've been using my own custom dataset with SCINet to great success (great work done here), but I noticed that the financial dataloader had a specific option just for long term prediction. This option is not listed in the ETT dataloader, so I am wondering what its purpose is, and if I should implement it in the use case of long term prediction on a custom dataset.

mixiancmx commented 2 years ago

Hi, that option is just for a fair comparasion with other long term forecasting baselines in our future paper. Those methods, i.e, Autoformer and Informer, use a different normalization and data division(i.e, train:valid:test=7:1:2). You can ignore it.