Test Results applicable to real-time CPD? #2
Description
In your paper, you describe the application of your model to retrospective CPD which is typically synonymous with offline CPD, where the entire the dataset has been observed and the goal is to segment the time-series dataset accordingly. I assume this is the case given you did not compare some form of Time to Detection between the different methods.
My question then, if you have the entire dataset and are looking for CPs in some time series. Why would you limit yourself to only windows of the data at time? This is usually done in online (real-time) CPD because of necessity. Have you tried testing your method in an on-line fashion? (and compare it with CUSUM methods or other online CPD methods)