This post will demonstrate the PELT algorithm from the changepoint package–a competing algorithm to the twitter package’s breakout detection algorithm. While neither of these algorithms produce satisfactory results, one change point location approximation algorithm that makes no distributional assumptions shows potentially promising results.
I received some feedback regarding my first foray into change point analysis from Brian Peterson. While some of it was good, a fair bit more was how I can add more to my analysis, more boxes I could check off, and so on. One of those boxes was the PELT algorithm, devised by one Rebecca Killick of Lancaster University, which I’ll give a quick run-through of below.
In the twitter paper, PELT was the competing algorithm that the paper compared itself to, and while I didn’t think that replicating the competing algorithm would be necessary at first go, it turns out, that, well, it…
Ver la entrada original 1.815 palabras más