Petri net analysis tool

Petri net, Coverability tree, Algorithms

During an internship at the Institute of Applied Informatics and Formal Description Methods (AIFB) at KIT we have been working in teams of two. Our team has developed a plugin for business processes modelling tool Horus Business Modeler, which had a feature allowing users to model workflow using Petri nets. Our plugin took Petri net as an input, generated and visualized Coverability Tree and computed following properties:

My responsibility was to implement the underlying algorithms while my partner took care of plugin integration and visualization of trees and graphs in the tool. Among algorithms we used to estimate properties were:

• A New Modified Reachability Tree Approach and its Applications to Unbounded Petri Nets.
• Macro Liveness Graph and Liveness of ω-independent Unbounded Nets.

• When: Summer semester 2016
• Technologies: Java, RCP, SVN
• Category: Algorithms, Software Engineering

Big Data Analysis

R, Machine Learning.

This was a team project which consisted of two parts. In the first part we have participated in the data mining cup. The topic of the competition was the prediction of the return rates for a fashion distributor. We have received raw data, which we have cleaned, prepared and built forecasting models to predict return rates. You can read more about DATA MINIG CUP here. We have used various algorithms, such as decision trees, random forest and Logistic Regression.

In the second part of the project we have received raw data from a large car manufacturer, collected by truck motor sensors. As a part of the project we cleaned, prepared the data and built a predictive maintainance model.

• When: Summer semester 2016
• Technologies: R, RStudio
• Category: Big Data Analytics, R, Machine Learning

Linked Data

Big Data, RDF, SPARQL

The aim of this seminar was to query and analyze data collected in the scope of Dynamic Linked Data Observatory (DYLDO) experiment. This is a long-term experiment to monitor the two-hop neighborhood of a core set of eighty thousand diverse Linked Data documents on a weekly basis. Data collected in the course of DYLDO experiment provides a collection to analyze and derive insights about the dynamics of Linked Data. In the course of this seminar I wrote various SPARQL queries to retrieve such information as distribution of available documents, death and change ratios of documents. Some of these queries were included in the conference paper Modelling and Analysing Dynamic Linked Data using RDF and SPARQL.

• When: Summer semester 2017
• Technologies: SPARQL, Apache Jena
• Category: Linked Data, RDF, SPARQL

Hoepfner Burghof

PHP, MySQL, JavaScript, HTML

I am currently working on a new Homepage for Hoepfner Burghof, current state can be accessed here. This project gave me a possibility to learn PHP and to boost my SQL and Database skills.

My another project is Villa Durla

• When: on it now
• Technologies: PHP, MySQL, JavaScript
• Category: Web Application, UX

System analysis tool

Web Application, SAPUI5.

During my internship at SAP I have implemented and designed a Web Application which utilized data from system error database and helped developers to navigate through it. This tool had a purpose of helping developers to find the error, which caused a particular problem and give them an overview of the overall system health.

Furthermore, I have been working on a smaller projects, which gave me a chance to get to know SAP HANA and its Predictive Analysis Library (PAL) as well as QuantLib library.

• When: Winter semester 2016/2017
• Technologies: Odata, SAPUI5, JavaScript, CSS
• Category: Web Application

Master Thesis

Algorithms, Covering arrays, Java.

The goal of my master thesis was to implement a plugin which would support product managers at keeping product models error-free. Product models consist of components and rules defining how these components can be combined in products. Product model thus describes a product family. Many product configurators then translate these rules into propositional calculus. To construct a ready product, salespeple successively pick components and each of these steps is validated by product configurator. Product configurator thus functions as SMT Solver and checks whether current configuration is compatible with the defined product model. Definition of such models is a very time-consuming and very error-prone process.

To make this process easier I have implemented a plugin which would generate regression tests for the product model. One regression test is one valid configuration. After product manager makes a change in the model, regression tests are executed and varios insides about model change can be drawn from the results. The most complicated and challenging about this task was finding a suitable algorithm for generation of test suite. Most algorithms require vast amout of calls to SMT Solver. If the model is small, it is usually not a problem, but with thousands of components and rules, the perfomance is a critical issue. Furthermore, the test suite has to be small for the exact same reason. Product managers want to get the insights fast, so the test suite must be kept small while still providing good coverage. To create a small sample with good coverage I have used Combinatorial Interaction Testing technique. I have implemented varios algorithms, such as Mixed Tabu Search, IPOG and ICPL. ICPL, the algorithm from the field of SPL Testing proved to be the most suitable for sampling of product configurations.

• When: Summer semester 2017
• Technologies: Eclipse, SVN, Java, RCP, Hibernate
• Category: Algorithms, Java