We construct similar problems from our benchmark TSP [9] problems
by modifying the benchmarks in five different ways. Table 1 lists the
five kinds of modifications that were used. Since we know the current best
solution to the TSP benchmarks, we train the genetic algorithm on the modified
problem 
saving and using the solutions to to help solve
the original problem. We compare the performance of the Non-Randomly
initialized GA (NRGA) with a Randomly initialized GA (RGA) and the known
optimal solution.
Table 1: Different ways of modifying TSPs
For all the problems, we use the same crossover and mutation probabilities of 0.85 and 0.05 respectively but use different population sizes and various number of generations for different problems. Table 2 shows the population sizes and the number of generations we use. For each modified problem, we run the RGA 10 times with 10 different random seeds and periodically save the best individual. We inject these chosen individuals (eight total) into the initial population of the NRGA to help solve the original problem.
Table 2: Population and generation size for different problems