Диссертация (1137084), страница 27
Текст из файла (страница 27)
Theframework already has plug-ins which take care of visualisation for Petri nets, import and exportof an event log, compatibility of logs with miner plug-ins, and provide further opportunities towork with resulting data. Thus, it was not necessary to develop additional supporting software.The tool takes into account the advantages and drawbacks of other existing approaches.Nevertheless, it also has its areas to improve.
In particular, other model formalisms should besupported by the tool. Another useful feature is the simulation of process models with specifiedroles of activity performers.114Section 3.3 described an approach for the generation of event logs for high-level process models.Formal semantics for the core part of the BPMN standard, including flat control flow, hierarchy,data perspective, and participants, are defined and supported by our approach. On the basis ofthe introduced semantics, a tool has been implemented for automated event log generation asa plug-in for ProM 6 Framework. Furthermore, generator Gena is enriched with capabilities ofmodifying the behaviour of exclusive gateways and simulating models with time.The approach has been tested using hundreds of models from the Signavio BPMN ReferenceModels Collection.
The results of this evaluation showed that the approach allows the generation ofevent logs for a significant portion of models from this collection. Moreover, the tool has shown itseffectiveness, efficiency and robustness. The entire testing collection can be processed in less thanhalf an our on basic laptop.
The tool is resistant to unsound models that get stuck in deadlocksor livelocks. Experimentations also show the need to expand the number of BPMN elementssupported by the prototype tool. Several types of events are not yet supported by our tool butare present in the model collection. This is especially true for message events. Nevertheless, thecore part of BPMN model is supported by our approach.With the help of generator Gena, researchers and BPM specialists may evaluate processmining algorithms using sets of event logs with desired characteristics generated for fine-tunedmodels.
A possible extension of the approach can be to incorporate more elements from theBPMN 2.0 standard. The clear identification of dependencies between specific model parametersand statistical characteristics of corresponding event logs may also be an interesting area for futureresearch.In the future, we plan to extend this tool for other modelling formalisms. In particular, currentlyhigh-level Petri nets, EPCs, and other model notations are not supported.
We also plan to improvetool’s features when generating event logs for BPMN models.Our generator Gena allows us to generate event logs with suitable characteristics. Using thistool we evaluated modular process repair technique that has been presented in Chapter 2. Thisevaluation is presented in the next chapter.115Chapter 4Implementation and ExperimentsThis chapter describes an experimental evaluation of the modular approach for model repairthat has been described in Chapter 2. It consists of three sections. Section 4.1 describes a prototypetool that implements the modular repair technique. Using this tool and the event log generatorfrom Chapter 3, the approach has been evaluated. The results of experimental evaluation areprovided and discussed in Section 4.3.
Finally, Section 4.4 concludes the chapter.4.1Prototype Tool DescriptionA modular repair approach has been implemented as a plug-in for ProM 6 Framework [26], awell-known tool in the process mining community. This section considers how this implementationis made. Firstly, the main features of the tool are described shortly. Secondly, we describe howProM 6 Framework is organized. Thirdly, a prototype tool is considered on this basis.Consider main features of a tool that implements the model repair approach described inChapter 2.I.
The tool allows users to select a decomposition method which, in their opinion, is the mostsuitable for a given model.II. The tool makes it possible to choose a repair algorithm. The choice of the algorithm istypically based on the properties of the algorithm and a desired repaired model characteristics.
Inpractice, the task of choosing the best repair algorithm is an attempt to find appropriate alternativebetween time needed for the algorithm to do its work, presence or absence of additional transitions(including silent transitions), and conformance between a given model and an event log.III. One may specify importance of each metric for a particular repair task. This step is essentialfor automatic evaluation of how well the tool helps researchers achieve the desired repair result.116IV. The tool shows performed operations, and outputs a final repaired process model.
In thefuture, the tool will also be fitted with informative visualisations.V. Numeric results of the final model evaluation can be stored in a CSV file, either manually orautomatically. The evaluation process assesses main model quality criteria: fitness (two approachesfor fitness measurement are employed), precision, generalization, and simplicity.A tool’s GUI used for specifying repair settings is shown in Figure 4.1.Figure 4.1: Model repair plug-in — basic settingsOne may note that Figure 4.1 shows decomposition/repair options additional to the consideredin this thesis.
Actually, these algorithms are supported by the tool, but their usage to repair processmodel need to be verified theoretically, and evaluated practically. Besides, not all the algorithmsare fully tested and tuned as building blocks. Extending the presented modular repair approachusing these additional decomposers and repairers is one of future research directions.ProM 6 FrameworkThe tool is developed using Java 6 Standard Edition 1 , and is based on ProM 6.4 Framework2 .ProM 6.4 is an open-source framework specially designated for implementing process miningalgorithms in a standardized way.
It consists of the core part and disjoints parts called plug-ins.The core part of the framework is responsible for uploading available plug-ins, handling plug-inslife cycle, interaction between plug-ins and basic functions for dealing with graphical user interface.This part of ProM 6 has been developed over the last years by the Architecture of InformationSystems group3 at the Eindhoven University of Technology. Hence, programmers focus exclusively1Java 6 SE is available at the page: http://www.oracle.com/technetwork/java/javase/downloads/java-archive-downloads-javase6-419409.html2ProM 6 official page: http://www.promtools.org3The official page of AIS group is https://www.win.tue.nl/ais/doku.php. Currently, the group changed itsname to Analytics for Information Systems with the same abbreviation and key research topics.117on algorithm implementations, work with file system and visualisation of results. The frameworkis distributed under GNU Public License, although some plug-ins are distributed under differentlicenses.Once a plug-in is properly configured, ProM 6 automatically discovers, finds and uploadsit, then this plug-in is added to the list of all available plug-ins.
In addition, the list of plugins demonstrates parameters required by each plug-in. By doing this, the framework simplifiesproviding parameters needed for plug-ins. The framework takes care about parameters needed forplug-ins. Special plug-ins were developed which load data into the environment and export resultsto a hard disc as well as being stored in ProM 6 resource pool for using them in other plug-ins.Nowadays, almost all data types for working with Petri nets have been implemented and suppliedwith visualizers, so researchers and developers are eliminated of necessity to implement them fromscratch.Each plug-in has so-called context. Context acts as a bridge between a plug-in and theframework because it is the only way plug-ins can interact and collaborate with ProM 6. Forevery context child contexts may be created, each of which is designated for a specific task.
Thus,it is possible to construct a hierarchy of plug-in calls from a parent plug-in.Plug-ins may run either with or without graphical user interface. The former provides a richpossibility to interact with user or visualize data, whereas the later enables to call other plug-insin the background simultaneously with user interaction in the main plug-in. ProM 6 encouragesdevelopers to write an scalable and loose coupled software, providing a rich set of tools.One of such tools, extensively used in ProM 6 tool, is a mechanism for finding all classesannotated in a special way.
Arguably the most common way to use annotations is to mark Javaclasses that contain certain algorithms. One creates an interface for a set of related algorithms,then annotate each of them. After that, they can be easily found and used via annotations.Interaction between plug-ins is accomplished by using a plug-in manager. The plug-in managerprovides API for invoking plug-ins, makes sure that correct context is configured for a called plugin. The plug-in manager enables not only to invoke known plug-ins but also to look for plug-ins withspecific signature, to invoke them and to obtain results of their executions.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
