next up previous
Next: Fitness Landscape Up: Solution of the Previous: Ramped-Half-and-Half Random Search

Comparison with Other Methods


Table 3 gives E values for various methods of solving the Ant problem. Rows 2--6 are from calculations in the previous sections, row 7 is from [Koza1992], while rows 8 onwards have been calculated from runs previously reported which used three types of varible length subtree mutation (except the last row using PDGP has not previously been reported).

Table 3:  Effort to Solve Santa Fe Trail

From Table 3 it is clear that there are many techniques capable of finding solutions to the Ant problem and although these have different performance the best typically only do marginally better than the best performance that could be obtained with random search.

In the following sections we investigate the Ant problem fitness landscape to explain the comparitively poor performance of these search techniques.

William B Langdon
Tue Feb 3 19:53:29 GMT 1998