Add MambaSL model (Published in ICLR-2026)

[yoom618]

We proposed a Mamba-based time series classification model, called MambaSL,
and evaluated on the entire UEA 30 datasets.

Paper Link: https://openreview.net/pdf?id=YDl4vqQqGP or https://iclr.cc/virtual/2026/poster/10008907
Code Link: https://github.com/yoom618/MambaSL

In this pull request, we add the model codes and test scripts for saved checkpoints:

• exp/exp_classification.py : Fixed error (set checkpoint directory using args.checkpoints)
• layers/MambaBlock.py & models/MambaSingleLayer.py : Add our model
• run.py : Add additional hyperparams, and override setting name to contain all model hyperparams .
  • It would be more awesome to override the setting name for all provided baselines (e.g. our run.py), but I only add it for MambaSL to avoid unnecessary changes to other models.
  • I hope this can be useful for better checkpoint management and reproducibility in the future.
• scripts/classification/MambaSL.sh : Add test script for MambaSL on UEA datasets. You can run the script to reproduce the results in our paper with saved checkpoints.

We are greatly thankful for all contributors who have created and maintain this library. Any kind of modification for better integration of our model into this library is welcome.