Mario level generation as an optimisation problem within the GBEA benchmark (detailed information about the optimization problems can be found there).
The data here is in compiled format (jar), while the source code is released in another Github repository: https://github.com/TheHedgeify/DagstuhlGAN/tree/v1.0
future>=0.17.1
scipy>=1.3.1
torch>=1.2.0
torchvision>=0.4.0
(Earlier versions of these packages might also work.)
For PyTorch, follow the installation instructions here.
Call
python mario_gan_evaluator.py
to run a test evaluation for some selected Mario GAN functions.
Note that this does test the correctness of the evaluation, but rather checks that the code is being executed without problems.
Usage in COCO
Running the algorithm of your choice on one of the two Mario GAN test suites in COCO
(the single-objective rw-mario-gan an the bi-objective rw-mario-gan-biobj) can range
from trivial to not-so-trivial, depending on the language used. See the explanation
here.