This is my experiment to test how Genetic Algorithms can be used to speed up search. This repo uses a very simplistic approach to solve a simple problem, to search for magic squares.
Magic squares make the perfect candidate for testing the efficacy of
genetic algorithms. The state space grows very huge by the relation
(s * s)!, where s is side of the square, and the number of solutions
gets more and more sparse
The search was relatively fast for small boards, but on larger boards the chance of getting stuck on a local minima was greater, upon hitting a minima for the larger boards you would have to increase the mutation rate to essentially start again.
The crossover function, even though it is not very domain specific, helped a lot is speeding up search, its addition was very noticable as it quickly converged to any minimas.
I usually like to crossover about 70% to 80% of the population, and the rest generated by mutation, this is because sometimes we are extremely close to solving the problem and crossover may yield outlandish answers. Further lengthening the search
I am not sure if k way selection actually has any significant benefits, a more detailed statistical study will be needed to conclude anything regarding k-way selection.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent