Gaussian Process on time series with a known trend break

Hi, how can I set up a Gaussian Process model on a time series data with a known trend break? Thanks.

Can you describe the nature of this trend break more concretely? Is there literally no dependence between the data on either side of the break? Or are you talking something like an elbow where the two trends must connect but otherwise have different slopes/lengthscales/etc?


It is more like the elbow situation. The goal is to extend the time series using GP. Thanks.

Wait, so you don’t have real data past the elbow? If not then this is just a case of forecasting. There was a recent post asking about forecasting this that you should search.

I have 6 years of monthly data. But there is an event that occurred in week 60, so the time series dropped from week 61 to 72. The goal is to extend to week 84.

Can you tell me which forecasting post you are referring to?

Sorry, I should have linked in my original reply. Here it is. .