search

NicMcPhee / XO-bias-study

Results and write-up of our genetic programming crossover bias study
MIT License
0 stars 0 forks source link

Do larger population sizes #14

Closed NicMcPhee closed 9 years ago

NicMcPhee commented 9 years ago

If we up the population size to 10K or higher, will that cause tournament size of 7 to show an effect?

NicMcPhee commented 9 years ago

This led to a lot of out of memory errors on the Ordertree problem. On the K-landscapes problem, bigger populations (10,240 instead of 1,024) lead to better fitness, but actually washed out the impact of XO bias because so many runs were up near the best fitness.

I think it would be worth looking at the impact of population size on some of the harder regression problems, but if that doesn't do anything we may want to skip this.