To prevent spam users, you can only post on this forum after registration, which is by invitation. If you want to post on the forum, please send me a mail (h DOT m DOT w DOT verbeek AT tue DOT nl) and I'll send you an invitation in return for an account.

How can I measure generalization?

TangYahui
TangYahui
in - Usage
Hello everyone:
  I have a Petri net, execute the plug-in 'Replay a Log on Petri Net for Conformance Analysis' and then execute the plug-in 'Measure Precision/Generalization' to compute generalization.
 While I met a error when execute the plug-in 'Replay a Log on Petri Net for Conformance Analysis' ,which says that my petri net doesn't have innitial marking and final marking.I add some as it mentioned,but it still doesn't work.
 I find that the measure of generalization quality demension must based on replay log,and replay must based on a petri net with initial marking and final marking.But when I use heuristic miner,sometimes mined model doesn't have them.
 Dose the plug-in 'Replay a Log on Petri Net for Conformance Analysis' is the only plug-in I can use to measure generalization?
 In fact I have compute all the other 3 quality demension.I find that generalization may only have one plug-in I can use.
Thank you in advance.
TangYahui

Comments

  • TangYahui
    TangYahui
    edited May 2019
    I also have used "anti-alignment precision/generalization" it still doesn't work.
  • hverbeek
    hverbeek
    Hello,

    To compute metrics like the generalization metric, some alignment between the model and the log is required. The replay plug-in you mention computes such an alignment, after which the measure plug-in can compute the generalization. There are other plug-ins than the replay plug-in that also can compute an alignment.

    If no initial and/or final marking is known for a model, the replay plug-in allows you to add them 'on-the-fly', this is not an error. You mention that "it still doesn't work", but for me it is unclear what does not work (and why). Is it possible to be more specific? Which model, which log, which steps do you take?

    A possibility is that in your situation the final marking is simply not reachable in the model from the initial marking. As a result, no alignment can be computed. Please check whether the final marking is reachable.

    Kind regards,
    Eric.
  • TangYahui
    TangYahui
    Hi Eric,

    I think the problem happens in alignment.For example,I use "Replay a log on petri net for conformance analysis" plug-in,the result shows that no trace can be found.Then if I use"Measure Precision/Generalization",the result show the value of them are 0.Same problem happened on others alignment plug-ins .By the way I use alpha miner,also happens on ILP and heuristic.

    I found that in this situation,only use "discover using decomposition" and "compute project fitness and precision"plug-ins can get fitness and precision .But still cannot get generalization value.

    kind regards,
    TangYahui




    1.PNG 53.3K
    2.PNG 7.1K
  • hverbeek
    hverbeek
    Hi TangYahui,

    In the net shown above, the final marking is not reachable from the initial marking. Both DES and CONF add a token to the same place, and the second token will not be removed.

    A precondition for alignments is that the final marking can be reached. Some miners do not guarantee that this is the case. Please check!

    Kind regards,
    Eric.
  • TangYahui
    TangYahui
    Hi Eric，
     
    So if the final marking can not be reached(which I think equals to can not align log and model)，the generalization can not be measured with existing scientific and technological support?

    Kind regards,
    TangYahui



  • hverbeek
    hverbeek
    Hi TangYahui,

    Yes. An alignment between a trace and a model guarantees that the result of the alignment in the model brings you from the initial to the final state. If the final state cannot be reached from the initial state, no alignment is possible. If no alignment is possible, generalization cannot be computed.

    Kind regards,
    Eric.
  • TangYahui
    TangYahui
    Hi Eric，
     
    Thank you!
    Now I'm convinced.

    Kind regards,
    TangYahui
Sign In or Register to comment.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Sign In Register

Categories

In this Discussion