5

My question is similar to this but with a lot more noise. Basically, I want to extract repeating 'saw'-patterns from a time-series. They look like this: Example 1 Example 2 Example 3

I don't have any labelled data, but the patterns all have one thing in common. A steep rise followed by a slow decline that repeats itself for a little while. (Also recognisible in the step function in purple below the pattern).

How can I go about extracting these patterns from a big batch of different time-series? Preferably in Python. I'd rather not spend days building a model, so the simpler the better :)

Community
  • 1
  • 1
k88
  • 1,858
  • 2
  • 12
  • 33
  • 1
    Possible duplicate of [Pattern recognition in time series](http://stackoverflow.com/questions/11752727/pattern-recognition-in-time-series) – VMRuiz Apr 24 '17 at 13:29
  • @k88 Hi! Did you find the answer to your question? I'm doing the same thing now and I can't find a solution. Can you tell me anything? – Alexander Ananev Sep 15 '19 at 10:46
  • @AlexanderAnanev at the time I worked my way by grouping the time series by the on/off data I had (seen as the purple chart at the bottom). Then I extracted features about these parts of the time series (take a look at ts-fresh). At the time I manually labeled the data and did some supervised ML on it, but in hindsight clustering would have been better and easier to do. So maybe try that ? – k88 Sep 16 '19 at 04:22

0 Answers0