Setting c_max considering label length

kaist-dmlab / CrossMatch

Implementation of the paper "Context Consistency Regularization for Label Sparsity in Time Series" (ICML'23)

MIT License

20 stars 17 forks source link

Setting c_max considering label length #11

Closed bamiikim closed 8 months ago

bamiikim commented 8 months ago

Thank you for sharing your excellent work!

I have one question: Is it important to consider cases where labels are mixed when determining the hyperparameter c_max? I'm curious about how to set c_max when time series have significantly different label lengths.

I appreciate your response in advance.

YoojuShin commented 8 months ago

Very good point. I would chose c_max as short as possible that do not exceed the minimum length of class labels in that case to prevent information flow from different classes.