Question About the Implementation of Channel Independence

[cccccrj]

Hello, thank you for sharing your excellent work and code! While reading your paper, I noticed the mention of "channel independence" as part of the design or mechanism. However, I couldn't locate the specific implementation of this in the provided code.

Could you please clarify where the channel independence mentioned in the paper is implemented in the code? Looking forward to your response, and thank you in advance!

[WenWeiTHU]

Hello, you can refer to our data loader for the implementation of channel independence. ---- Replied Message ---- ***@***.***>Date12/12/2024 00:03 ***@***.***> ***@***.***>Subject[thuml/AutoTimes] Question About the Implementation of Channel Independence (Issue #19) Hello, thank you for sharing your excellent work and code! While reading your paper, I noticed the mention of "channel independence" as part of the design or mechanism. However, I couldn't locate the specific implementation of this in the provided code. Could you please clarify where the channel independence mentioned in the paper is implemented in the code? Looking forward to your response, and thank you in advance! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[cccccrj]

Hello, I noticed that in the current implementation, the data is split by channels, which indeed ensures the independence of the input data. However, I have a question: all channel data is ultimately fed into the same model, and the model's weights are shared. Does this mean that the data from different channels is not entirely modeled independently? Instead, the model might learn the same features for all channels, rather than capturing the specific characteristics of each channel?

[WenWeiTHU]

Yes, you are right! It is a well-acknowledged approach (channel-independence). There are also some channel-dependent models for explicit cross-channel modeling, such as iTransformer and Timer-XL :)