chaoxel / pyboyenv

Turn any gameboy game into a reinforcement learning environment.

PyBoyEnv

This package allow you to turn any gameboy memory event into a reinforcement learning environment rule.

(generated with test.py)

Installation

Install the dependencies:

pip install -r requirements.txt

Install the package with pip:

pip install ./

Quickstart

import gym
import pyboyenv

# Create the environment
env = gym.make('Pyboy-v0', game=<FILE>)

# Add rules
env.set_reward_rule(ADDRESS, TYPE, VALUE, LABEL)

# Available rules:
# - increase: add VALUE if the memory at the address ADDRESS increases
# - decrease: add VALUE if the memory at the address ADDRESS decreases
# - smaller X: add VALUE if the memory at the address ADDRESS is smaller than X
# - bigger X: add VALUE if the memory at the address ADDRESS is bigger than X
# - equals X: add VALUE if the memory at the address ADDRESS equals X
# - in X1,..,XN: add VALUE if the memory at the address ADDRESS is equal to X1 or ... or XN

# Useful values
cumul = 0 # Sum of rewards 
done = False # Is the game done ?
action = 16 # Initial action (action list available below, 16=nothing)

# Game loop
while not done:
    # State = Window screen
    state, reward, done, info = env.step(action)
    cumul += reward
    action = Agent.get_action(state, cumul, reward) # Next action
    # Print infos
    for i in info:
        print(f"{i[0]}: {i[1]}")

Actions

• 0: WindowEvent.PRESS_ARROW_UP
• 1: WindowEvent.PRESS_ARROW_DOWN
• 2: WindowEvent.PRESS_ARROW_LEFT
• 3: WindowEvent.PRESS_ARROW_RIGHT
• 4: WindowEvent.PRESS_BUTTON_A
• 5: WindowEvent.PRESS_BUTTON_B
• 6: WindowEvent.PRESS_BUTTON_SELECT
• 7: WindowEvent.PRESS_BUTTON_START
• 8: WindowEvent.RELEASE_ARROW_UP
• 9: WindowEvent.RELEASE_ARROW_DOWN
• 10: WindowEvent.RELEASE_ARROW_LEFT
• 11: WindowEvent.RELEASE_ARROW_RIGHT
• 12: WindowEvent.RELEASE_BUTTON_A
• 13: WindowEvent.RELEASE_BUTTON_B
• 14: WindowEvent.RELEASE_BUTTON_SELECT
• 15: WindowEvent.RELEASE_BUTTON_START
• 16: WindowEvent.PASS