A general game playing research project using Tangled Program Graphs (TPG), a genetic programming algorithm.
-
individuals: Each environment is trained on individually.
-
15-shrink-novir: Environments to train on selected randomly, with 9 in the training pool at a time, and 3 being swapped out per generation. Starts with 15 in the training pool, and slowly shrinks to 9.
-
15-shrink-vir: Environments to train on selected based on the environment fitness which is based on a host-parasite model, with 9 in the training pool at a time, and 3 being swapped out per generation. Starts with 15 in the training pool, and slowly shrinks to 9.
-
8-all-at-once: All environments being trained on every generation.
-
8-all-at-once-window-4: All environment being trained on from the ground up but only up to 4 at a time selected randomly.
-
8-merge: All environments being trained on every generation, but continuing on from individually trained populations.
-
8-merge-window-4: Only up to 4 of the 8 environments being trained on every generation, selected randomly, but continuing on from individually trained populations.