Finding Winning Patterns in ICPC data
Boer, R.H. de
MetadataShow full item record
The International Collegiate Programming Contest is a worldwide, multi-tier competition where students from universities all over the globe compete to be the best programmers. These competitions result in data in the form of scoreboards, which are never analyzed and collected before. The purpose of this master thesis is to gather and structure all this data, and provide insights. It details characteristics of highly performing teams, compares different competitions and looks deeper at the way problem characteristics influence each other. This thesis answers the question how competitions compare and what patterns exist throughout the recent years. Results show that competitions can be explored and summarized using the measures of popularity and difficulty and a visualization technique to show the distribution of solutions over a single year. This information and other metrics are used to compare within and between regions, where the European competitions are found to be most similar to the World Finals. Further analysis was done on the best performing teams, where was found that they are better in all aspects of their game compared to other teams; they on average require less attempts, solve faster, can handle more difficult problems, are slightly more efficient, and hand in more solutions. Finally, several different ways of gathering information problem descriptions were used, where was found that it is difficult to build a model that is both accurate in identifying algorithmic topics and has a high recall. A model that with reasonably high precision was made and run on some ICPC problems, where was found that (of the most prevalent topics) problems labeled as being about data structures are most significantly different from others. All these results can be used by multiple stakeholders of the ICPC.