aadharna / untouchablethunder Goto Github PK

View Code? Open in Web Editor NEW

16.0 16.0 2.0 215.75 MB

Co-evolution of agents and environments in GVG-AI

Python 79.26% Jupyter Notebook 19.42% Shell 1.32%

coevolution neuroevolution open-ended-evolution poet reinforcement-learning transfer-learning

untouchablethunder's People

Contributors

Stargazers

Watchers

Forkers

amykhoover cryptowealth-community

untouchablethunder's Issues

Neural Net Input format

I read related paper:
https://arxiv.org/pdf/2007.08497.pdf

If I understand correctly, it is mentioned that you use a group of 13 maps (9x13 each) to encode the entire game state, each map encoding one object type.
So I was curious if you tried with one single 9x13 map instead and setting different pixel color values for the different object types. If so what was the performance difference or issues which such an approach as opposed to have one map per object type?

Discussion

This is a place for Mike and Aaron to talk about code.

Important files are:

UntouchableThunder/agent/base.py

Lines 5 to 9 in 71a2cf9

    
           class Agent: 
        
               """ 
        
               Wrap each env with a game-playing agent 
        
               """ 
        
               def __init__(self, GG, max_steps=250):

UntouchableThunder/generator/env_gen_wrapper.py

Lines 10 to 16 in 71a2cf9

    
           class GridGame(gym.Wrapper): 
        
               def __init__(self, game, play_length, path, lvl_name, mechanics=[], id=0): 
        
                   """Returns Grid instead of pixels 
        
                   Sets the reward 
        
                   Generates new level on reset 
        
                   -------- 
        
                   """

UntouchableThunder/generator/levels/base.py

Lines 19 to 26 in 71a2cf9

    
           class Generator: 
        
               def __init__(self, tile_world, path=dir+'envs/games/zelda_v0/', mechanics=[], generation=0): 
        
                   """ 
        
                   :param tile_world: 2d numpy array of map 
        
                   :param path: gym_gvgai.dir 
        
                   :param mechanics: list of sprites you would like to be able to mutate into 
        
                   """

You can see how all three of these interact inside
UntouchableThunder/combinedTesting.ipynb

aadharna / untouchablethunder Goto Github PK

untouchablethunder's People

Contributors

Stargazers

Watchers

Forkers

untouchablethunder's Issues

Neural Net Input format

Discussion

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	class Agent:
	"""
	Wrap each env with a game-playing agent
	"""
	def __init__(self, GG, max_steps=250):

	class GridGame(gym.Wrapper):
	def __init__(self, game, play_length, path, lvl_name, mechanics=[], id=0):
	"""Returns Grid instead of pixels
	Sets the reward
	Generates new level on reset
	--------
	"""

	class Generator:
	def __init__(self, tile_world, path=dir+'envs/games/zelda_v0/', mechanics=[], generation=0):
	"""

	:param tile_world: 2d numpy array of map
	:param path: gym_gvgai.dir
	:param mechanics: list of sprites you would like to be able to mutate into
	"""