pacman-decision-tiny / example.json
guilindev's picture
Release 17,601-parameter policy, training recipe and measured pilot
1f7a5ce verified
Raw History Blame Contribute Delete
1.37 kB
{
"state": {
"lives": 2,
"pellets_left": 181,
"to_junction": 5.8,
"options": {
"up": {
"ghost": null,
"pellets": 6,
"food": 0.8,
"power": 10.8
},
"left": {
"ghost": null,
"pellets": 6,
"food": 0.8
},
"back": {
"ghost": null,
"pellets": 11,
"food": 1.2,
"power": 11.2
}
}
},
"questions": {
"move": {
"type": "choice",
"instructions": "You control Pac-Man in a maze. Choose which way to go at the next junction. Eating pellets scores points. Touching a normal ghost loses a life. After a power pellet, ghosts are frightened for a few seconds and can be eaten for bonus points. Each option lists facts about the path through that exit: ghost = steps to the nearest normal ghost (null if none within 15), ghost_coming = that ghost is moving toward you, edible = steps to the nearest frightened ghost, pellets = pellets within 6 steps, food = steps to the nearest pellet, power = steps to the nearest power pellet. Steps count from Pac-Man, including the corridor up to the junction. 'back' means turn around right now instead of going to the junction.",
"criteria": {
"up": null,
"left": null,
"back": "turn around now"
}
}
}
}