Working with algorithms whose performance greatly depends on their parameters requires conducting numerous experiments and using expert knowledge in order to achieve best results. In this thesis we evaluate strategies for experimentation and parameter tuning, especially in the context of machine learning. We introduce Relaks – an embedded programming language – as a framework for conducting experiments and automatic parameter tuning. An implementation of the language is presented and later applied in the machine learning domain. We discuss strengths and weaknesses of the proposed solution and suggest directions for further development.
Sources reside in relaks/