I want to estimate percentile statistics for time-series pairs $(t, y)$ over varying granularities of time (hour, day, month, etc)
For example, if I have the following pairs (10:59, 3), (11:01, 2), (11:32, 5), the first point is part of a different hour window, but all three are part of the same day window.
When a new point arrives, I am recomputing the hour window statistics directly, by iterating over all data points in the window. Unfortunately, this does not scale to the longer time windows.
Is there a way to incrementally update the statistics for the longer time windows?
