How to test the evolutionary framework quickly

The ImageNet experiment takes ~4 days x 128 cores to run on the cluster. To test evolving images quickly, you can choose to either:

1. Run the MNSIT experiment dl_map_elites_images_mnist here.

2. Run the test ImageNet experiment dl_map_elites_images_test.cpp here. If you choose this option #2, please follow the steps below:

• Then configure waf to compile your experiment. In wscript file, update the obj.source property to: obj.source = 'dl_map_elites_images_test.cpp'

This experiment should evolve 28x28 color images in 10 ImageNet classes from 1-10. And it should take it a few minutes to run on a single machine of 4 cores.

Please note the settings configured differently from the real experiment:

     // Evolve towards 10 ImageNet classes instead of 1000
     SFERES_CONST size_t res_y = 10;  // e.g. 1000
     SFERES_CONST int num_categories = 10;  //e.g. 1000

     //number of initial random points
     static const size_t init_size = 10; //e.g. 400

     // size of the population
     SFERES_CONST unsigned size = 10; //e.g. 400;

     // number of generations
     SFERES_CONST unsigned nb_gen = 10; //e.g. 5000;

     // Evolve images of size 28 x 28
     SFERES_CONST int size = 28;  // e.g. 256

     // Use only 1 center crop as configured in the prototxt
     SFERES_CONST bool use_crops = false;