has anyone seen MLP-Mixer for Time Series yet?

[BradKML]

I found the following repo, and it mentioned at least 4 other references of forecasting https://github.com/EmorZz1G/PatchAD

• https://github.com/giuliomattolin/tsmixer-pytorch
• https://github.com/yuqinie98/PatchTST
• https://github.com/Zeying-Gong/PatchMixer
• https://github.com/aikunyi/FreTS

Would they be good to add to PyPOTS if it really is that good?

[WenjieDu]

Hi Brad, PatchTST and FreTS have been integrated into PyPOTS, just not extended to the forecasting task yet. New models are welcome of course. Will you be interested in helping with the integration?

[BradKML]

If there is a clean way of dragging code from research papers into standard papers, why not? (still surprised that nobody has done anything with AI agents yet to do this)

[WenjieDu]

You raise a good question. Actually, I did try AI assistants like Cursor to add new algorithms but the output still needs a bunch of modifications to make everything ready. I would say the modularized framework of PyPOTS is convenient enough for quick integrations of new time series models. To people who are familiar with PyPOTS code base, it may take only half an hour or less to add a new algorithm if the original code is in good format

Anyway, talk is cheap, you can give it a try to see how much workload agents can reduce for you, please let me know if you have anything to share ;-)