LPTM Implementation and Adaptive Segmentation Code

[guangyuyang001]

I'm currently exploring the implementation provided for your paper, "Large Pre-trained Time Series Models for Cross-domain Time Series Analysis Tasks," and I have encountered some confusion.
Specifically:

1. In the paper, it mentions using a GRU to obtain hidden embeddings; however, I cannot find this implementation within the provided code.
2. The paper describes an "Adaptive Segmentation" method, but the segmentation logic in Samay/src/samay/models/lptm/model/backbone.py seems to employ fixed-size patches.
3. Additionally, the function select_segments in Samay/src/samay/models/lptm/segment/selection.py does not appear to be used anywhere in the provided codebase.
4. The method LPTMPipeline.from_pretrained is referenced, but I can't locate its definition in the repository.
   Could you please clarify whether there is an error in the uploaded code or direct me to the correct implementation and usage of these components?