Decision Miner in 6.5?

LJames

I've been using the Decision Miner to generate decision trees in Prom 5.2.  Are there any plugins with similar fuctionality in Prom 6.5?  How hard should it be to port? 

 

Massimiliano

Dear LJames,

yes, there is a similar plug-in in ProM 6.5.

This is plug-in Discovery of the Process Data-flow (Decision-Tree Miner). It requires an event log and a Petri-net model as input. It generates a so-called Data Petri Net, which is Petri nets where transitions are annotated with guards and with possible read and write operations. In particular, the plug-in discovers guards for transitions at decision points as well as the written operations.

I also enclose a zip file containing a tutorial about this plug-in with a synthetic event log for testing. You need to discover a process model before using decision miner. You can use the log in the zip file as input and discover a model, e.g., through Inductive Miner.

For further information about the algorithm:

M. de Leoni, W. M. P. van der Aalst
"Data-Aware Process Mining: Discovering Decisions in Processes Using Alignments"

In Proc. of the 28th ACM Symposium on Applied Computing (SAC 2013) 18-22 March, 2013, Coimbra, Portugal. Track "Enterprise Engineering"

Please note that, while discovering guards, the plug-in allows one to evaluate them before completing. When one is confronted with guards, theoretically one is allowed to press "Cancel", go back and change the discovery parameters and, finally, apply the technique again. However, a software bug prevents the "Cancel" button from working. This will be fixed in ProM 6.5.1, which will be soon released. For now, please simply ignore it.

speeters

Dear Massimiliano,

I've run into some issues with the Discovery of the Process Data-flow plugin (as well as the Multiperspective explorer) and was wondering if I could get your input.
I'm attempting to perform a similar experiment as the one you wrote about in the paper mentioned in your previous post. I constructed my own synthetic logs for this, both in XES and MXML. Then I discovered the process model using the alpha algorithm and attempted to apply the decision mining plugin.
The problem is that it consistently considers the continuous event attributes (which are part of the guards) as being nominal. I can't tell if this is because the way I loaded them into ProM (6.5.1a) or if it is the decision miner plugin that is causing it. Any help would be greatly appreciated!

Kind regards,

Samuel Peeters

Massimiliano

Dear Peeters,

the decision miner reads from the event log the type of attributes and considers them accordingly.

The fact that the attribute is treated as nominal is because it is recorded as literal in the event log. For example:

<string key="article" value="180"/>

If you want it to be consider as numeric, you should make sure that it is recorded as:

<int key="article" value="180"/>

or

<float key="article" value="180"/>

It is a quite easy fix. You can open the event log with a log and to a search-and-replace. 
In the nightly-build version, there is also a plug-in Enhance Log: Repair Type of Event Attributes in the Log that can do it for you.

I hope that this solves your problem.

Regarding MXML, MXML does not allow you to specify the attribute type. So, when MXML files are loaded in ProM, all attributes are considered as literal/nominal because that is the safest choice.

Cheers,
Massimiliano de Leoni