Irregularly sampled data

[TalGrbr]

Hello! I want to use the library to detect change points in a multivariate irregularly sampled signal.
Meaning, time gaps between samples isn't constant, and on each timestamp some of the dimensions might be null.
I want to use bottom up / kernelcpd with rbf cost function, what modifications should I do to my data and to the algorithm?
I can do interpolation and resample to a constant grid, but my data might be too noisy and interpolation might be too bad for me.
Thank you!

[deepcharles]

Hi, sorry for the late reply.

Great question!

ruptures doesn't natively handle irregularly sampled signals.

However, if you consider a piecewise constant model (e.g., kernelcpd), irregular sampling may not change anything. This cost function computes the mean distribution (using a kernel) over segments. The formula for the kernel mean remains unchanged under irregular sampling.

Side note: ruptures returns the indices of changes. To get the times of changes, you need to find the timestamps at the indices given by ruptures.

I hope this helps,

Charles