Question about data_loader.py

KimMeen / Time-LLM

[ICLR 2024] Official implementation of " 🦙 Time-LLM: Time Series Forecasting by Reprogramming Large Language Models"

https://arxiv.org/abs/2310.01728

Apache License 2.0

1.02k stars 179 forks source link

Question about data_loader.py #72

Closed Shi-Yuhong closed 1 month ago

Shi-Yuhong commented 2 months ago

Dear author, thank you for your excellent work！ I have some questions about getitem(self, index) in data_loader.py It seems that seq_x only contains data for one feature. And I don't quite understand this kind of handling. Is this related to the handling of Channel Independence？ But I checked the code of PatchTST and his DataLoader is not like this. Could you please provide a specific explanation of the data processing in getitem(self, index)，thanks a lot！！

ztb-35 commented 2 months ago

I'm also interested in this question.

Shi-Yuhong commented 2 months ago

I think maybe I have found the answer. The Dataloader of this code is the same as GPT4TS. Someone has raised a similar question and has already received an answer, please refer to NeurIPS2023-One-Fits-All/issues/15 for details.This operation does seem to be related to channel independence

kwuking commented 1 month ago

I think maybe I have found the answer. The Dataloader of this code is the same as GPT4TS. Someone has raised a similar question and has already received an answer, please refer to NeurIPS2023-One-Fits-All/issues/15 for details.This operation does seem to be related to channel independence

Yes, your understanding is completely correct. This is the data processing method that implements the channel-independent operation, and it reduces the amount of data entering the GPU at the same time, avoiding the OOM problem caused by too much data.