|
I have a sequence of events, some of which are produced by an automated system and some by users. I'm trying to find the automated events, under the assumption that they are regularly spaced (i.e. every 20 seconds, +- 10%). Specifically, I'm looking for a subsequence of events, such that each is spaced with some distribution (assume normal or even* between two intervals). There may be gaps (if an error happened and the event didn't fire) and it's somewhat of a needle in the haystack problem (guessing a hundred matches from tens of thousands of events). Are there any algorithms or tests that look at this type of problem? * i.e. if the system just does: interval = 20 + randint(-5, 5) |
Are there any general/related/somewhat comparable techniques that may be extended to this area? Any ideas?
I'm thinking maybe a Band-Pass filter tuned for your interval, so you only filter those events that happen on those intervals.
Like filtering the modulation on a transmission
Thanks Leon, I'll look into this later today.
Estimate the autocorrelation function. It may give you a good idea how to look for them.