return opts
if __name__ == '__main__':
opts = parseOptions()
###########################
# GET THE GRIDWORLD
###########################
import gridworld
mdpFunction = getattr(gridworld, "get"+opts.grid)
mdp = mdpFunction()
mdp.setLivingReward(opts.livingReward)
mdp.setNoise(opts.noise)
env = gridworld.GridworldEnvironment(mdp)
###########################
# GET THE DISPLAY ADAPTER
###########################
import textGridworldDisplay
display = textGridworldDisplay.TextGridworldDisplay(mdp)
if not opts.textDisplay:
import graphicsGridworldDisplay
display = graphicsGridworldDisplay.GraphicsGridworldDisplay(mdp, opts.gridSize, opts.speed)
display.start()
###########################
return opts
if __name__ == '__main__':
opts = parseOptions()
###########################
# GET THE GRIDWORLD
###########################
import gridworld
mdpFunction = getattr(gridworld, "get" + opts.grid)
mdp = mdpFunction()
mdp.setLivingReward(opts.livingReward)
mdp.setNoise(opts.noise)
env = gridworld.GridworldEnvironment(mdp)
###########################
# GET THE DISPLAY ADAPTER
###########################
import textGridworldDisplay
display = textGridworldDisplay.TextGridworldDisplay(mdp)
if not opts.textDisplay:
import graphicsGridworldDisplay
display = graphicsGridworldDisplay.GraphicsGridworldDisplay(
mdp, opts.gridSize, opts.speed)
try:
display.start()
except KeyboardInterrupt:
sys.exit(0)
# Attribution Information: The Pacman AI projects were developed at UC Berkeley. # The core projects and autograders were primarily created by John DeNero # ([email protected]) and Dan Klein ([email protected]). # Student side autograding was added by Brad Miller, Nick Hay, and # Pieter Abbeel ([email protected]). import mdp, util import gridworld from learningAgents import ValueEstimationAgent import collections # VOOR DEBUGGEN mdpFunction = gridworld.getBridgeGrid mdp = mdpFunction() mdp.setLivingReward(0) mdp.setNoise(0.2) env = gridworld.GridworldEnvironment(mdp) #a = ValueIterationAgent(mdp, 0.9, 100) class ValueIterationAgent(ValueEstimationAgent): """ * Please read learningAgents.py before reading this.* A ValueIterationAgent takes a Markov decision process (see mdp.py) on initialization and runs value iteration for a given number of iterations using the supplied discount factor. """ def __init__(self, mdp, discount=0.9, iterations=100): """
def main(myargs):
sys.argv = myargs.split()
opts = parseOptions()
###########################
# GET THE GRIDWORLD
###########################
if opts.grid == 'VerticalBridgeGrid':
opts.gridSize = 120
import gridworld
mdpFunction = getattr(gridworld, "get"+opts.grid)
mdp = mdpFunction()
mdp.setLivingReward(opts.livingReward)
mdp.setNoise(opts.noise)
env = gridworld.GridworldEnvironment(mdp)
###########################
# GET THE DISPLAY ADAPTER
###########################
import textGridworldDisplay
display = textGridworldDisplay.TextGridworldDisplay(mdp)
if not opts.textDisplay:
import graphicsGridworldDisplay
display = graphicsGridworldDisplay.GraphicsGridworldDisplay(mdp, opts.gridSize, opts.speed)
try:
display.start()
except KeyboardInterrupt:
sys.exit(0)
###########################
# GET THE AGENT
###########################
import valueIterationAgents, qlearningAgents
a = None
if opts.agent == 'value':
a = valueIterationAgents.ValueIterationAgent(mdp, opts.discount, opts.iters)
elif opts.agent == 'q':
#env.getPossibleActions, opts.discount, opts.learningRate, opts.epsilon
#simulationFn = lambda agent, state: simulation.GridworldSimulation(agent,state,mdp)
gridWorldEnv = GridworldEnvironment(mdp)
actionFn = lambda state: mdp.getPossibleActions(state)
qLearnOpts = {'gamma': opts.discount,
'alpha': opts.learningRate,
'epsilon': opts.epsilon,
'actionFn': actionFn}
a = qlearningAgents.QLearningAgent(**qLearnOpts)
elif opts.agent == 'random':
# # No reason to use the random agent without episodes
if opts.episodes == 0:
opts.episodes = 10
class RandomAgent:
def getAction(self, state):
return random.choice(mdp.getPossibleActions(state))
def getValue(self, state):
return 0.0
def getQValue(self, state, action):
return 0.0
def getPolicy(self, state):
"NOTE: 'random' is a special policy value; don't use it in your code."
return 'random'
def update(self, state, action, nextState, reward):
pass
a = RandomAgent()
else:
if not opts.manual: raise 'Unknown agent type: '+opts.agent
###########################
# RUN EPISODES
###########################
# DISPLAY Q/V VALUES BEFORE SIMULATION OF EPISODES
try:
if not opts.manual and opts.agent == 'value':
if opts.valueSteps:
for i in range(opts.iters):
tempAgent = valueIterationAgents.ValueIterationAgent(mdp, opts.discount, i)
display.displayValues(tempAgent, message = "VALUES AFTER "+str(i)+" ITERATIONS")
display.pause()
display.displayValues(a, message = "VALUES AFTER "+str(opts.iters)+" ITERATIONS")
display.pause()
display.displayQValues(a, message = "Q-VALUES AFTER "+str(opts.iters)+" ITERATIONS")
display.pause()
except KeyboardInterrupt:
sys.exit(0)
# FIGURE OUT WHAT TO DISPLAY EACH TIME STEP (IF ANYTHING)
displayCallback = lambda x: None
if not opts.quiet:
if opts.manual and opts.agent == None:
displayCallback = lambda state: display.displayNullValues(state)
else:
if opts.agent == 'random': displayCallback = lambda state: display.displayValues(a, state, "CURRENT VALUES")
if opts.agent == 'value': displayCallback = lambda state: display.displayValues(a, state, "CURRENT VALUES")
if opts.agent == 'q': displayCallback = lambda state: display.displayQValues(a, state, "CURRENT Q-VALUES")
messageCallback = lambda x: printString(x)
if opts.quiet:
messageCallback = lambda x: None
# FIGURE OUT WHETHER TO WAIT FOR A KEY PRESS AFTER EACH TIME STEP
pauseCallback = lambda : None
if opts.pause:
pauseCallback = lambda : display.pause()
# FIGURE OUT WHETHER THE USER WANTS MANUAL CONTROL (FOR DEBUGGING AND DEMOS)
if opts.manual:
decisionCallback = lambda state : getUserAction(state, mdp.getPossibleActions)
else:
decisionCallback = a.getAction
# RUN EPISODES
if opts.episodes > 0:
print()
print("RUNNING", opts.episodes, "EPISODES")
print()
returns = 0
for episode in range(1, opts.episodes+1):
returns += runEpisode(a, env, opts.discount, decisionCallback, displayCallback, messageCallback, pauseCallback, episode)
if opts.episodes > 0:
print()
print("AVERAGE RETURNS FROM START STATE: "+str((returns+0.0) / opts.episodes))
print()
print()
# DISPLAY POST-LEARNING VALUES / Q-VALUES
if opts.agent == 'q' and not opts.manual:
try:
display.displayQValues(a, message = "Q-VALUES AFTER "+str(opts.episodes)+" EPISODES")
display.pause()
display.displayValues(a, message = "VALUES AFTER "+str(opts.episodes)+" EPISODES")
display.pause()
except KeyboardInterrupt:
sys.exit(0)
return opts
if __name__ == '__main__':
opts = parseOptions()
stepsLimit=9
###########################
# GET THE GRIDWORLD
###########################
import gridworld
mdpFunction = getattr(gridworld, "get"+opts.grid)
mdp = mdpFunction()
mdp.setLivingReward(opts.livingReward)
mdp.setNoise(0.1)
env = gridworld.GridworldEnvironment(mdp)
###########################
# GET THE DISPLAY ADAPTER
###########################
import textGridworldDisplay
display = textGridworldDisplay.TextGridworldDisplay(mdp)
if not opts.textDisplay:
import graphicsGridworldDisplay
display = graphicsGridworldDisplay.GraphicsGridworldDisplay(mdp, opts.gridSize, opts.speed)
try:
display.start()
except KeyboardInterrupt:
sys.exit(0)