Heuristic Miner not deterministic?

splitcells

Hi,
why is the Miner seem not deterministic?
Example:
I created a Petri Net in ProM 6.3 via Log and exported it to file.
After that I started ProM 5.0 and imported the log.
I added atrifical Start and End Events via Filter to the log.
Then I imported the petri net and connected it to the filtered log.
I started Analysis/Conformance Checker and clicked two times on "Start Analysis".
The results differ a little. For instance the Fitness: 0.890 ; 0.889.
Also the diagnostic token counter value vary: 7, 9, 6....

There is not much "randomness" but
its strange for me that the algorithm is a bit random.

splitcells

Sorry, my post or more precisely my question is false...

The question is: why is the conformance Checker not deterministic?

JBuijs

My guess would be that your model allows for 'a lot' of different behaviour. Since considering all possible options is not feasible some random choices could be made to prevent an exhaustive exploration of all possibilities.

I'm curious what your findings are when you try the alignment based approach to calculate the replay fitness (only available in ProM 6).

splitcells

Yes, my model allows a lot of different behaviour.

"Since considering all possible options is not feasible some random
choices could be made to prevent an exhaustive exploration of all
possibilities."
You are right. The "Resctrict search depth for invisible task" option seems to do this job in a non deterministic
(I guess a HashSet or something like that is the cause on the implementation side).

"I'm curious what your findings are when you try the alignment based
approach to calculate the replay fitness (only available in ProM 6)."
I guess you mean "Replay a Log on Petri Net for All Optimal Aligment"?
Alignment statistifcs ( Avarage/case):
-Min Fitness Cost 106,20
-Max Fitness Cost 132,60
-Trace length 171,80
-#Algiments 10

Unfortunately I don't know the meaning of the result yet and I didn't find a description yet.

I am new to Processes Mining and going throw the algorithms to find the best ones for my tasks.
I have logs that contain two Groups of cases.
I am searching for similarities and differences in the logs of the two Groups.
I suspect that there are properties in the logs that appear in the logs of one group
more often or always but less often or never in the logs of the other group.
(Properties that are in the logs of both groups are not as important right now
but these would be helpful too)

I am searching for a way to find these properties and to describe
them and their appearance in the logs.


Do you know which algortihm could be suitable for this task?

Thanks for answer