This is code for training bots to play Mortal Kombat 3 for Genesis using NEAT-python and gym-retro liberty
To use this project you need to install following python libertys:
1.gym-retro 2.numpy 3.opencv 4.python-neat
Before runig any of this moduls you need to add states to gym-retro's site-package in "data\stable\MortalKombat3-Genesis" folder, states that i used are in "playGame/states". You also need to provide game rom that i have no rights to and can not publish.
To train new network you need to change file to witch final network will be saved in "neat_mortal.py" and then run it. (not changing file will override last saved network). Do to limitation on my device i could train only on tree stats at the time. If you wish to do more you cant change states list in "neat_mortal.py"
To see results of traning network change runing file in "play.py" and run it. You also can change state to which one you want to start game from.
Do to limitation on my device i can only geave weak examples.
I managed to train networks that can always win matches that they where trind for.
sub-zero-winner6.bin
sub-zero-winner5.bin
But when it comes to fighting oponents that i didnt train network to fight result is unpridcitible:
sub-zero-winner6.bin
Network sub-zero-winner6.bin was traind on "sub-zero-mid1", "sub-zero-mid2", "sub-zero-hard1" states and it took over 36h and over 70 generations for me.
