Is there a worked out example for time series prediction?
Thanks for your reply, am currently using FB Prophet, however either am not using it properly or its not providing enough information for my use case, I am working with some sensor reads that are collected every five seconds, the reads are bouncing around and sometimes spike, I would like to know the probability of the last read read being a peak or a trough?
Any thoughts / suggestions?
I started looking at PYMC3 as it has a different approach to time series analysis and forecasting I guess, but not being an expert not sure how to use it.
I have uploaded a sample data set or sensor readings
Thankssample-data-pmprophet.csv (120.4 KB)
You might be interested in Gaussian Process Regression, e.g.: https://docs.pymc.io/notebooks/GP-MaunaLoa.html
Hi twiecki, I looked at it and it was quite helpful, thanks.
I am also interested in identifying the state of the system, sometimes readings are range bound and dont change much but at other times there is a very strong move either up or down.
I looked at the possibility of using Hidden Markov Model to identify the state of the system. Is there any worked out example in PyMc3 for something similar? thanks
I’m not sure how you would combine GPs and HMM although it should be possible in principle. PyMC3 is mainly tuned towards continuous models (but doesn’t rule out discrete models either), so that might not be a smooth path. If you’ve been using pm-prophet, you might want to get in touch with @luca_giacomel who has been experimenting with change-points in Prophet: https://twitter.com/luca_giacomel/status/1049985944403812353