The "evaluate process tree" plug in.

TangYahui

Hello everyone:
 I want to get the four quality dimension values of the model I mined,so I use the "evaluate process tree"plug in.
 But I get a non-visualized result.and he also can not be operated.
 I want to know where the problem is.


 Thank you in advance.
TangYahui

JBuijs

Hi TangYahui,

Please us the 'calculatePTQualityString' variant which returns the tree fitness as a string representation.

TangYahui

Hi Buijs:

  I don't understand what you said. Where can I find the 'calculatePTQualityString' variant ?

yours,

TangYahui

TangYahui

Hi Buijs:
 
  In fact, I have found plug-ins to measure the other three quality dimensions, only don't know how to calculate the simplicity .

Yours,

Tangyahui

[Deleted User]

Hi Yahui,

For the simplicity metric, you can find the definition from "On the Role of Fitness, Precision, Generalization and Simplicity in Process Discovery". http://wwwis.win.tue.nl/~wvdaalst/publications/p688.pdf. I guess you need to implement it by yourself. 

Hope this helps.
Cheers,
Cong

TangYahui

Hi Cong,

 I have read it.And now I almost know how to calculate manually.

 But it is better if there is  a plug-in I can use.

 Yours
 TangYahui

JBuijs

Hi TangYahui,

The plug-in I mentioned should produce a string with four quality dimensions encoded within it. I guess it's not available in ProM lite, but should be in ProM 6.7.

The resulting string should contain all four quality dimensions.

TangYahui

Hi Bujis：

I think that the plug-in you said is the plug-in for the image I uploaded earlier.

But it is unreadable and inoperable.
I don’t know where the problem is.


Thank you for your reply 

yours

TangYahui

                                                                                                                                    

JBuijs

Hi TangYahui,

The plug-in you have executed now returns a 'TreeFitness' object. For this there is indeed no visualizer or plug-in that takes this as input.
There should be another plug-in in the list with a very similar name that produces a 'String' object, which can be visualized. Can you check?

DStekel

Hello, 

I use ETM CLI version: the plugin variant is called "evaluate_process_tree_returns_string(Xlog, PTree)". It results in a string like this:
0,3529 (Gv: 0,0000 Pe: 1,0000 Fr: 0,0000 Of: 0,3529 Su: 1,0000 )

The 0,3529 is the overall quality value, but I am not sure which of the Gv, Pe, Fr, Su stands for Replay Fitness, Precision, Simplicity and Generalization. I have looked into the source code but did not get any wiser.

Does anyone know what these values mean?

Many thanks!

hverbeek

Dear Daniel,

• Gv = Generalization (using the average number of visits to each node)
• Pe = Precision (using escaping edges)
• Fr = Fitness (using replay)
• Su = Simplicity (using useless nodes)
• Of = Overall Fitness

Cheers,

Eric.

DStekel

Thank you!!!

JBuijs

Thanks Eric! You're right. See the ETM folder and then the fitness.metrics package. Each metric has an info object with a key ('Gv'), dimension ('Generalization') and description (as Eric showed). The default weights to calculate the Of is (off the top of my head) 10 x Fr + 5 x Pe + Gv + Su normalized by dividing by 17 (sum of the weights).

Hope this helps!

DStekel

Thanks guys. Unfortunately, the "evaluate_process_tree_returns_string(log, tree)" variant does not work correctly in my case.

Applying ETMd to one of my workflow logs results in the following output, where it states that the resulting process tree has an overall quality of >0.97 (great!).




However, when I export the resulting process tree to a file (ETMd.ptml) and use the same workflow log file and process tree as input for "import_process_tree_from_ptml", then the algorithm returns a quality of only 0.1845... Any idea if I miss something?

The image beneath shows a ProM script, which I use as input for ProM CLI. The console contains the output from ProM.




Any help is greatly appreciated!

Thanks in advance,
Daniël 

JBuijs

Hi Daniel,

When exporting a process tree from ETM as ptml there is a conversion from the ETM internal model (NAryTree) to a ProcessTree instance.

Given the fact that your Fr is 0 in your recalculation I guess that the alignment settings are not correct (my guess: the XEventClassifier).

My recommendation would be to hack a bit into the code and work directly with NAryTrees. I have save and load functions, even functions to translate NAryTrees to nice looking LaTeX pictures, even with quality values etc. etc. See my thesis for the end result and the ETM and JoosBuijs packages for the code.

Hope this helps!!!