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.

The Classifier about ETM plugin


Hello everyone,

If I input a .mxml,which classifier should I choose?

I have try many kind of event logs on this plug in(they all come fromhttp://http//data.4tu.nl/repository/collection:event_logs_real ), unfortunately,none of them can be handled for 1000 generations.They usually stops in 10~20 gengrations with the error"java.lang.IllegalStateException: Fitness evaluation task execution failed."

Meanwhile,I have try all the classifiers for the log(http://http//data.4tu.nl/repository/uuid:c45dcbe9-557b-43ca-b6d0-10561e13dcb5 )But the question still exists

Now I find that only the event log generated by PLG can be handled correctly,but these logs are too simple and lacks attribute values.

Can Anyone help me?

Thank you inadvance.
Tagged:
33.jpg 27.6K

Comments

  • hverbeekhverbeek Posts: 426
    Dear,

    My guess is that the ETM runs out of memory after a number of generations. You can choose any classifier you want, but these logs seem to be just too big to be handled by the ETM miner.

    You could try to use a different import plug-in for the event log (preferably one that uses less memory), or to first clean the event log as much as possible.

    Kind regards,

    Eric.
  • TangYahuiTangYahui Posts: 59
    Dear,

    Which is the difference between the classifiers?

    How can I choose them?

    Kind regards,

    Tang.
  • hverbeekhverbeek Posts: 426
    Dear Tang,

    A classifier determines which event attributes are used to derive the activity name from the event. Assume we have an event for which the following attributes have the following values:
    • concept:name = "Select Classifier"
    • lifecycle:transition = "complete"
    • org:resource = "Tang"
    Choosing different classifiers will yield different activity names for the event:
    • MXML Legacy Classifier = "Select Classifier+complete"
    • Event Name = "Select Classifier"
    • Resource = "Tang"
    • Lifecycle transition = "complete"
    Which one to choose depends on what you deem important for the activity name. Typically, the Event Name will do, but if the event log contains both start and complete events for every activity, then the MXML Legacy Classifier would be better. Note that in such a case the Event Name classifier would result in two identical events (both "Select Classifier") for every activity, whereas the MXML Legacy Classifier would distinguish them ("Select Classifier+start" and "Select Classifier+complete"). If you want a hand-over of work network, you could choose the Resource classifier.

    Kind regards,

    Eric.

  • TangYahuiTangYahui Posts: 59
    Dear Eric,

    Thank you very much for your detailed explanation.

    "Choosing different classifiers will yield different activity names for the event",do you mean that the event is in the mined model?

    And how can I solve  ETM runs out of memory ?Using a computer with more memory?

    Thank you again .

    Kind regards,

    Tang
  • hverbeekhverbeek Posts: 426
    Dear Tang,

    No, the activity name will be in the mined model. But the activity name is derived from the event using the classifier you choose.

    Did you already change the memory setting of ProM? Using the ProM Package Manager, you can set how much memory ProM may use. If you have a computer with 8 GB of memory, then I would run ProM with at least 4 GB. If this does not help, and you really need to run this algorithm, yes, then more memory may be a solution.

    Kind regards,

    Eric.
  • TangYahuiTangYahui Posts: 59
    Dear Eric,

    I have changed the memory setting of the prom for the max value.And It still doesn't work,so I decide to choose some of the traces in the event log that can stand for the total event log.

    What is the difference between the mine a process tree with ETMd and mine configured
    process tree with ETMC? 

    Thank you again .

    Kind regards,

    Tang
  • hverbeekhverbeek Posts: 426
    Dear Tang,

    The ETMd takes a single event log as input whereas the ETMc takes a collection of event logs as input. Each event log from this collection is then considered to be an event log for some variant of the process. ETMd results in a plain process tree, whereas ETMc results in a configurable process tree (which embeds all variants).

    Kind regards,

    Eric.

  • TangYahuiTangYahui Posts: 59
    Dear Eric,

    Thank you again.

    I believe my experiment can be made as soon as possible with your help.

    Kind regards,

    Tang
  • TangYahuiTangYahui Posts: 59
    One more question,

    I generate event logs from a existing process model(which can be achieved by PLG),then use ETM mine these event log to get a process model.but  it was very different from the existing process model.

    Can I improve the mined model (to be close to the existing model)by changing the weight of one of the four quality dimensions?

    The first complex picture is gengrated by prom.
    The second picture is the existing model(bulit by PLG)

     Thank you again.

    图片1.png 15.4K
    图片2.png 12.9K
  • hverbeekhverbeek Posts: 426
    Dear Tang,

    Please be aware of the limitations of the ETM miner. Although in the long run it will return a good model, it needs time to do so. It needs time to traverse the space of process trees looking for good models. In the short run, there are no guarantees. Changing the weights will not help much here, I guess.

    Perhaps it would be better to first mine a process tree with the Inductive Miner, and then use ETMr on the event log and discovered process tree to discover a better (if possible)  process tree.

    Kind regards,

    Eric.
  • TangYahuiTangYahui Posts: 59
    Dear Eric,

      I have tried you way,and does ETMr means that the process tree mined by IM r
    eplaced randomly process tree generated by ETM? I also find that we can put more than one process tree as input.the ETM's defult population size is 20,if I put 1 process tree as input,wheather the other 19 process trees are randomly generated or not?

     Thank you again.

     kind regards,
     
     TangYahui

  • hverbeekhverbeek Posts: 426
    Dear Tang,

    Perhaps it is best if I refer you to the PhD thesis of Joos Buijs (see https://pure.tue.nl/ws/portalfiles/portal/4032592), which contains all the details for the ETM and its variants. The differnet variants are described as from page 79.

    Kind regards,

    Eric.


  • TangYahuiTangYahui Posts: 59
    Dear Eric,

     Thank you very much!I will study it carefully.

     Kind regards,
     
     TangYahui
Sign In or Register to comment.