0
$\begingroup$

In the paper "Discovery of frequent episodes in event sequences" by Mannila et al., a method for finding frequent episodes in an event sequence if a class of episodes and a sequence of events is given.

It is not clear to me how to find episodes in the first place. I would like to know if there are any tools, algorithms, or published work that can be used to this end, or which describes a method to find episodes in a given stream.

$\endgroup$
1
  • 1
    $\begingroup$ I'd link to the paper and elaborate what you are asking for here. $\endgroup$
    – Sean Owen
    Jul 16, 2015 at 20:56

1 Answer 1

1
$\begingroup$

The paper that you cited references two other papers that explain the method you are looking for here and here. The second one is behind a pay wall.

But... these look like deterministic methods to me, and I suggest you use a machine learning method instead. You basically need to separate events from each other using some sort of clustering algorithm. I suggest you use DBSCAN and adjust the parameters based on whether you want every event to be assigned to an episode or not.

Another algorithm that might work well is simple k-means clustering with the addition of an elbow method or silhouette score to determine the optimal number of clusters.

$\endgroup$
2
  • $\begingroup$ I see that they are trying to get rules out of minimal episodes but if given a event stream like 'G,E,A,A,B,D,H,G,F,D,G,F,C,F,D,H,H,H,G,F,D,G,E,B,D,H,H,H,G,F,C' where each event occurs with 1 time unit gap then what are the possible episodes in this stream? $\endgroup$ Jul 17, 2015 at 10:55
  • $\begingroup$ Then this doesn't fit within the definition of finding episodes as the events in an episode do not occur at regular frequency e.g. that is the definition of an episode. The paper you referenced specifically has inhomogeneous event frequency. It looks like a sequencing problem and you should employ sequencing methods. $\endgroup$
    – AN6U5
    Jul 17, 2015 at 17:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.