PCA (SVD)

[MaxHalford]

• Incremental PCA is already implemented in scikit-learn but seems like it is straightforward to implement. See this paper too. This paper looks quite interesting too.
• Streaming Principal Component Analysis From Incomplete Data
• Maybe this is easier once SVD is implemented
• Incremental Singular Value Decomposition of Uncertain Data with Missing Values

[MarekWadinger]

Hey @MaxHalford and @hoanganhngo610, I can have a look at this feature. 💪

In similar terms, as part of my research, I work with online Dynamic Mode Decomposition (oDMD), which combines benefits of both PCA and Fourier Transformation. Based on this paper Zhang, H., et al. (2019) and papers' codebase, I implemented the oDMD in river-like API as both mini-batch regressor and transformer. Moreover, it could be used with Rolling in windowed implementation.

You can check it out and decide, whether it aligns with the trajectory of river development. I would be happy to assist with integration :)

https://github.com/MarekWadinger/odmd-subid-cpd/blob/main/functions/odmd.py

[kulbachcedric]

Hi @MarekWadinger ,
are there any news on this issue, or are you still working on it?
We would like to clean up a little bit the issue board and close the inactive ones.
Feel free to reopen it!

Best
Cedric

[MarekWadinger]

Hey @kulbachcedric 👋,

I've merged latest freshness, resolved all the failures of ruff and mypy and proceeded with alignment with the River API, which seemed to be an issue the last time. I'd love to have your review on that and find a way how we can integrate decomposition methods to River. :)

Cheers,
Marek

[hoanganhngo610]

Dear @MarekWadinger,

Thank you so much for your latest updates! I will reopen the issue and have a look at this, hopefully within this week.

Best regards,
Hoang-Anh

[MaxHalford]

Thanks for looking into this one @hoanganhngo610!