Test Results From Using GEMS to Find String Outputting ML Programs

Background

Parametric analysis of the performance of the GEMS system/process required performing an extensive number of GP runs. The string outputting program, while not always short in CPU or clock time, was sufficiently limited that large numbers of GP runs could be made in a practical amount of time given the computer resources available. It was also deemed a non-trivial problem for GP and was thus chosen for use in testing, characterizing and developing GEMS.

Investigations Conducted

The first step was to prove that GEMS could repeatedly generate ML code. Subsequent investigations were to quantify or qualify the factors that influenced how much computing power, e.g., time, was required to generate ML programs of different complexity.

The parameters investigated to date are the effects of the:

Pool size
Problem size (i.e., the string length to be generated)
Mutation rate
Parent selection criteria
Evolutionary Leveraging

Go to GEMS Home Page

Go to Arktist Home Page

Updated August 28, 2008