Parallel Wilcoxon signed-rank tests
Name
Stenver Jerkku
Abstract
Statistical tests are used to find out if some sort of experimental stimulation affects observable features. In this paper we researched Wilcoxon signed-rank test which is one of the few statistical tests that can be used when the natural variation inside the group is not normally distributed. The test is used by Bioinformatics, Algorithmics and Data mining group research for gene regulation, gene expression data analysis, biological data mining and others. BIIT is a joint research group between the Department of Computer Science (University of Tartu), Quretec, and the Estonian Biocenter.
The current implementations of the Wilcoxon signed-rank tests are slow and unoptimized. This project looked into the foundations of Wilcoxon signed-rank test, its current implementations and how to optimize it. In order to make the implementation more accurate, the relationship between Wilcoxon statistic and Gaussian approximate was observed. In order to make the implementation faster, some dynamic programming methods were used to save computation time. The purpose of optimizing was to make it more accurate and speed up the test running.
In this project an accurate and fast Wilcoxon test shared library was created. In the scope of this project, the library was integrated with two tools - command line and GNU-R. Due to the nature of shared library, it will be easy integrate the library with any other tools one might desire.
Graduation Thesis language
English
Graduation Thesis type
Bachelor - Information Technology
Supervisor(s)
Sven Laur
Defence year
2014