Genetic Algorithm for Function Optimization

This project uses a genetic approach to minimize a specific function with N parameters. The general approach is:

Generate an initial population of randomly created members
Evaluate the fitness of all the members against the function to minimize
"Breed" the parents to create a child generation
- 10% of the time, a parent just moves into the next generation without producing children
- 90% of the time:
  - Children take a random half of their parameters from each parent
  - Children are randomly mutated 5% of the time
Iterate and continue until the function is minimized to a given tolerance (.999)

Build and Run

make
./assignment.exe <n_population> <n_parameters> <n_threads> <block_size>
- note the n_threads and block_size does not pass to all GPU functions. Many operations are preformed by the Thrust library, which controls it's threading internally.

Example Timings:

Time to 2000 Generations

1000 10: 0m9.878s
2000 10: 0m19.661s
5000 10: 1m28.506s
10000 10: 4m46.643s
1000 20: 0m10.339s
1000 30: 0m10.734s
1000 50: 0m11.556s
1000 100: 0m13.522s
1000 1000: 0m50.640s

Time to Convergence:

1000 10: 0m4.364s
2000 10: 0m4.522s
5000 10: 0m10.353s
10000 10: 0m8.888s
1000 20: 0m30.116s
1000 30: 0m50.463s
1000 50: 1m59.999s
1000 100: 8m41.778s