Repository navigation
Applications to time series #38
Description
Activity
I remember that in the original multiview paper Etienne tested with hundreds of stock prices dataset. It is, of course, more expensive than dealing with a single dataset, but the relation is linear.
So for N datasets, it should be N times more expensive than a single dataset, given a fixed amount of data per dataset.Here we have an example on how to use the multiview approach! I would love to see how well it fairs in this problem. If you want to give it a try, I can give you support on how to do it https://github.com/folivetti/eggp/blob/main/test/mvsr_example.ipynb
Reacted by mferguson9215Thank you for your response! This sounds promising and I am very interested in trying it. Currently, I am swamped with some ongoing projects / papers due to a long medical leave, so I might not be able to get to this for some time. However, I will likely include your work for a commentary piece I am working on to raise awareness about SR in the medical informatics community. Hopefully that will promote some experimentation with applying a multi-view approach to longitudinal data. I know that you are still working on this package, so feel free to let me know if you'd rather I leave this out.
Sounds great!! Let me know if I can be of any help :-)
I'll leave this issue open so one of us might work on that eventually!
Hi, this is a very impressive package you've developed. I believe that learning a general parametric model is a very natural approach to the goal of modeling the underlying phenomenon producing data. The multiview approach you take has me wondering if perhaps this could be used to find generalized models for timeseries / longitudinal data (e.g market data, patient histories, ect.). I've seen something similar done in the following paper: https://doi.org/10.1016/j.jfds.2025.100150. How computationally expensive do you think this would be?