intro2ai/p2_multiagent/game.py

730 lines
25 KiB
Python
Raw Normal View History

# game.py
# -------
# Licensing Information: You are free to use or extend these projects for
# educational purposes provided that (1) you do not distribute or publish
# solutions, (2) you retain this notice, and (3) you provide clear
# attribution to UC Berkeley, including a link to http://ai.berkeley.edu.
#
# Attribution Information: The Pacman AI projects were developed at UC Berkeley.
# The core projects and autograders were primarily created by John DeNero
# (denero@cs.berkeley.edu) and Dan Klein (klein@cs.berkeley.edu).
# Student side autograding was added by Brad Miller, Nick Hay, and
# Pieter Abbeel (pabbeel@cs.berkeley.edu).
# game.py
# -------
# Licensing Information: Please do not distribute or publish solutions to this
# project. You are free to use and extend these projects for educational
# purposes. The Pacman AI projects were developed at UC Berkeley, primarily by
# John DeNero (denero@cs.berkeley.edu) and Dan Klein (klein@cs.berkeley.edu).
# For more info, see http://inst.eecs.berkeley.edu/~cs188/sp09/pacman.html
from util import *
import time, os
import traceback
import sys
#######################
# Parts worth reading #
#######################
class Agent:
"""
An agent must define a getAction method, but may also define the
following methods which will be called if they exist:
def registerInitialState(self, state): # inspects the starting state
"""
def __init__(self, index=0):
self.index = index
def getAction(self, state):
"""
The Agent will receive a GameState (from either {pacman, capture, sonar}.py) and
must return an action from Directions.{North, South, East, West, Stop}
"""
raiseNotDefined()
class Directions:
NORTH = 'North'
SOUTH = 'South'
EAST = 'East'
WEST = 'West'
STOP = 'Stop'
LEFT = {NORTH: WEST,
SOUTH: EAST,
EAST: NORTH,
WEST: SOUTH,
STOP: STOP}
RIGHT = dict([(y,x) for x, y in LEFT.items()])
REVERSE = {NORTH: SOUTH,
SOUTH: NORTH,
EAST: WEST,
WEST: EAST,
STOP: STOP}
class Configuration:
"""
A Configuration holds the (x,y) coordinate of a character, along with its
traveling direction.
The convention for positions, like a graph, is that (0,0) is the lower left corner, x increases
horizontally and y increases vertically. Therefore, north is the direction of increasing y, or (0,1).
"""
def __init__(self, pos, direction):
self.pos = pos
self.direction = direction
def getPosition(self):
return (self.pos)
def getDirection(self):
return self.direction
def isInteger(self):
x,y = self.pos
return x == int(x) and y == int(y)
def __eq__(self, other):
if other == None: return False
return (self.pos == other.pos and self.direction == other.direction)
def __hash__(self):
x = hash(self.pos)
y = hash(self.direction)
return hash(x + 13 * y)
def __str__(self):
return "(x,y)="+str(self.pos)+", "+str(self.direction)
def generateSuccessor(self, vector):
"""
Generates a new configuration reached by translating the current
configuration by the action vector. This is a low-level call and does
not attempt to respect the legality of the movement.
Actions are movement vectors.
"""
x, y= self.pos
dx, dy = vector
direction = Actions.vectorToDirection(vector)
if direction == Directions.STOP:
direction = self.direction # There is no stop direction
return Configuration((x + dx, y+dy), direction)
class AgentState:
"""
AgentStates hold the state of an agent (configuration, speed, scared, etc).
"""
def __init__( self, startConfiguration, isPacman ):
self.start = startConfiguration
self.configuration = startConfiguration
self.isPacman = isPacman
self.scaredTimer = 0
self.numCarrying = 0
self.numReturned = 0
def __str__( self ):
if self.isPacman:
return "Pacman: " + str( self.configuration )
else:
return "Ghost: " + str( self.configuration )
def __eq__( self, other ):
if other == None:
return False
return self.configuration == other.configuration and self.scaredTimer == other.scaredTimer
def __hash__(self):
return hash(hash(self.configuration) + 13 * hash(self.scaredTimer))
def copy( self ):
state = AgentState( self.start, self.isPacman )
state.configuration = self.configuration
state.scaredTimer = self.scaredTimer
state.numCarrying = self.numCarrying
state.numReturned = self.numReturned
return state
def getPosition(self):
if self.configuration == None: return None
return self.configuration.getPosition()
def getDirection(self):
return self.configuration.getDirection()
class Grid:
"""
A 2-dimensional array of objects backed by a list of lists. Data is accessed
via grid[x][y] where (x,y) are positions on a Pacman map with x horizontal,
y vertical and the origin (0,0) in the bottom left corner.
The __str__ method constructs an output that is oriented like a pacman board.
"""
def __init__(self, width, height, initialValue=False, bitRepresentation=None):
if initialValue not in [False, True]: raise Exception('Grids can only contain booleans')
self.CELLS_PER_INT = 30
self.width = width
self.height = height
self.data = [[initialValue for y in range(height)] for x in range(width)]
if bitRepresentation:
self._unpackBits(bitRepresentation)
def __getitem__(self, i):
return self.data[i]
def __setitem__(self, key, item):
self.data[key] = item
def __str__(self):
out = [[str(self.data[x][y])[0] for x in range(self.width)] for y in range(self.height)]
out.reverse()
return '\n'.join([''.join(x) for x in out])
def __eq__(self, other):
if other == None: return False
return self.data == other.data
def __hash__(self):
# return hash(str(self))
base = 1
h = 0
for l in self.data:
for i in l:
if i:
h += base
base *= 2
return hash(h)
def copy(self):
g = Grid(self.width, self.height)
g.data = [x[:] for x in self.data]
return g
def deepCopy(self):
return self.copy()
def shallowCopy(self):
g = Grid(self.width, self.height)
g.data = self.data
return g
def count(self, item =True ):
return sum([x.count(item) for x in self.data])
def asList(self, key = True):
list = []
for x in range(self.width):
for y in range(self.height):
if self[x][y] == key: list.append( (x,y) )
return list
def packBits(self):
"""
Returns an efficient int list representation
(width, height, bitPackedInts...)
"""
bits = [self.width, self.height]
currentInt = 0
for i in range(self.height * self.width):
bit = self.CELLS_PER_INT - (i % self.CELLS_PER_INT) - 1
x, y = self._cellIndexToPosition(i)
if self[x][y]:
currentInt += 2 ** bit
if (i + 1) % self.CELLS_PER_INT == 0:
bits.append(currentInt)
currentInt = 0
bits.append(currentInt)
return tuple(bits)
def _cellIndexToPosition(self, index):
x = index / self.height
y = index % self.height
return x, y
def _unpackBits(self, bits):
"""
Fills in data from a bit-level representation
"""
cell = 0
for packed in bits:
for bit in self._unpackInt(packed, self.CELLS_PER_INT):
if cell == self.width * self.height: break
x, y = self._cellIndexToPosition(cell)
self[x][y] = bit
cell += 1
def _unpackInt(self, packed, size):
bools = []
if packed < 0: raise ValueError, "must be a positive integer"
for i in range(size):
n = 2 ** (self.CELLS_PER_INT - i - 1)
if packed >= n:
bools.append(True)
packed -= n
else:
bools.append(False)
return bools
def reconstituteGrid(bitRep):
if type(bitRep) is not type((1,2)):
return bitRep
width, height = bitRep[:2]
return Grid(width, height, bitRepresentation= bitRep[2:])
####################################
# Parts you shouldn't have to read #
####################################
class Actions:
"""
A collection of static methods for manipulating move actions.
"""
# Directions
_directions = {Directions.NORTH: (0, 1),
Directions.SOUTH: (0, -1),
Directions.EAST: (1, 0),
Directions.WEST: (-1, 0),
Directions.STOP: (0, 0)}
_directionsAsList = _directions.items()
TOLERANCE = .001
def reverseDirection(action):
if action == Directions.NORTH:
return Directions.SOUTH
if action == Directions.SOUTH:
return Directions.NORTH
if action == Directions.EAST:
return Directions.WEST
if action == Directions.WEST:
return Directions.EAST
return action
reverseDirection = staticmethod(reverseDirection)
def vectorToDirection(vector):
dx, dy = vector
if dy > 0:
return Directions.NORTH
if dy < 0:
return Directions.SOUTH
if dx < 0:
return Directions.WEST
if dx > 0:
return Directions.EAST
return Directions.STOP
vectorToDirection = staticmethod(vectorToDirection)
def directionToVector(direction, speed = 1.0):
dx, dy = Actions._directions[direction]
return (dx * speed, dy * speed)
directionToVector = staticmethod(directionToVector)
def getPossibleActions(config, walls):
possible = []
x, y = config.pos
x_int, y_int = int(x + 0.5), int(y + 0.5)
# In between grid points, all agents must continue straight
if (abs(x - x_int) + abs(y - y_int) > Actions.TOLERANCE):
return [config.getDirection()]
for dir, vec in Actions._directionsAsList:
dx, dy = vec
next_y = y_int + dy
next_x = x_int + dx
if not walls[next_x][next_y]: possible.append(dir)
return possible
getPossibleActions = staticmethod(getPossibleActions)
def getLegalNeighbors(position, walls):
x,y = position
x_int, y_int = int(x + 0.5), int(y + 0.5)
neighbors = []
for dir, vec in Actions._directionsAsList:
dx, dy = vec
next_x = x_int + dx
if next_x < 0 or next_x == walls.width: continue
next_y = y_int + dy
if next_y < 0 or next_y == walls.height: continue
if not walls[next_x][next_y]: neighbors.append((next_x, next_y))
return neighbors
getLegalNeighbors = staticmethod(getLegalNeighbors)
def getSuccessor(position, action):
dx, dy = Actions.directionToVector(action)
x, y = position
return (x + dx, y + dy)
getSuccessor = staticmethod(getSuccessor)
class GameStateData:
"""
"""
def __init__( self, prevState = None ):
"""
Generates a new data packet by copying information from its predecessor.
"""
if prevState != None:
self.food = prevState.food.shallowCopy()
self.capsules = prevState.capsules[:]
self.agentStates = self.copyAgentStates( prevState.agentStates )
self.layout = prevState.layout
self._eaten = prevState._eaten
self.score = prevState.score
self._foodEaten = None
self._foodAdded = None
self._capsuleEaten = None
self._agentMoved = None
self._lose = False
self._win = False
self.scoreChange = 0
def deepCopy( self ):
state = GameStateData( self )
state.food = self.food.deepCopy()
state.layout = self.layout.deepCopy()
state._agentMoved = self._agentMoved
state._foodEaten = self._foodEaten
state._foodAdded = self._foodAdded
state._capsuleEaten = self._capsuleEaten
return state
def copyAgentStates( self, agentStates ):
copiedStates = []
for agentState in agentStates:
copiedStates.append( agentState.copy() )
return copiedStates
def __eq__( self, other ):
"""
Allows two states to be compared.
"""
if other == None: return False
# TODO Check for type of other
if not self.agentStates == other.agentStates: return False
if not self.food == other.food: return False
if not self.capsules == other.capsules: return False
if not self.score == other.score: return False
return True
def __hash__( self ):
"""
Allows states to be keys of dictionaries.
"""
for i, state in enumerate( self.agentStates ):
try:
int(hash(state))
except TypeError, e:
print e
#hash(state)
return int((hash(tuple(self.agentStates)) + 13*hash(self.food) + 113* hash(tuple(self.capsules)) + 7 * hash(self.score)) % 1048575 )
def __str__( self ):
width, height = self.layout.width, self.layout.height
map = Grid(width, height)
if type(self.food) == type((1,2)):
self.food = reconstituteGrid(self.food)
for x in range(width):
for y in range(height):
food, walls = self.food, self.layout.walls
map[x][y] = self._foodWallStr(food[x][y], walls[x][y])
for agentState in self.agentStates:
if agentState == None: continue
if agentState.configuration == None: continue
x,y = [int( i ) for i in nearestPoint( agentState.configuration.pos )]
agent_dir = agentState.configuration.direction
if agentState.isPacman:
map[x][y] = self._pacStr( agent_dir )
else:
map[x][y] = self._ghostStr( agent_dir )
for x, y in self.capsules:
map[x][y] = 'o'
return str(map) + ("\nScore: %d\n" % self.score)
def _foodWallStr( self, hasFood, hasWall ):
if hasFood:
return '.'
elif hasWall:
return '%'
else:
return ' '
def _pacStr( self, dir ):
if dir == Directions.NORTH:
return 'v'
if dir == Directions.SOUTH:
return '^'
if dir == Directions.WEST:
return '>'
return '<'
def _ghostStr( self, dir ):
return 'G'
if dir == Directions.NORTH:
return 'M'
if dir == Directions.SOUTH:
return 'W'
if dir == Directions.WEST:
return '3'
return 'E'
def initialize( self, layout, numGhostAgents ):
"""
Creates an initial game state from a layout array (see layout.py).
"""
self.food = layout.food.copy()
#self.capsules = []
self.capsules = layout.capsules[:]
self.layout = layout
self.score = 0
self.scoreChange = 0
self.agentStates = []
numGhosts = 0
for isPacman, pos in layout.agentPositions:
if not isPacman:
if numGhosts == numGhostAgents: continue # Max ghosts reached already
else: numGhosts += 1
self.agentStates.append( AgentState( Configuration( pos, Directions.STOP), isPacman) )
self._eaten = [False for a in self.agentStates]
try:
import boinc
_BOINC_ENABLED = True
except:
_BOINC_ENABLED = False
class Game:
"""
The Game manages the control flow, soliciting actions from agents.
"""
def __init__( self, agents, display, rules, startingIndex=0, muteAgents=False, catchExceptions=False ):
self.agentCrashed = False
self.agents = agents
self.display = display
self.rules = rules
self.startingIndex = startingIndex
self.gameOver = False
self.muteAgents = muteAgents
self.catchExceptions = catchExceptions
self.moveHistory = []
self.totalAgentTimes = [0 for agent in agents]
self.totalAgentTimeWarnings = [0 for agent in agents]
self.agentTimeout = False
import cStringIO
self.agentOutput = [cStringIO.StringIO() for agent in agents]
def getProgress(self):
if self.gameOver:
return 1.0
else:
return self.rules.getProgress(self)
def _agentCrash( self, agentIndex, quiet=False):
"Helper method for handling agent crashes"
if not quiet: traceback.print_exc()
self.gameOver = True
self.agentCrashed = True
self.rules.agentCrash(self, agentIndex)
OLD_STDOUT = None
OLD_STDERR = None
def mute(self, agentIndex):
if not self.muteAgents: return
global OLD_STDOUT, OLD_STDERR
import cStringIO
OLD_STDOUT = sys.stdout
OLD_STDERR = sys.stderr
sys.stdout = self.agentOutput[agentIndex]
sys.stderr = self.agentOutput[agentIndex]
def unmute(self):
if not self.muteAgents: return
global OLD_STDOUT, OLD_STDERR
# Revert stdout/stderr to originals
sys.stdout = OLD_STDOUT
sys.stderr = OLD_STDERR
def run( self ):
"""
Main control loop for game play.
"""
self.display.initialize(self.state.data)
self.numMoves = 0
###self.display.initialize(self.state.makeObservation(1).data)
# inform learning agents of the game start
for i in range(len(self.agents)):
agent = self.agents[i]
if not agent:
self.mute(i)
# this is a null agent, meaning it failed to load
# the other team wins
print >>sys.stderr, "Agent %d failed to load" % i
self.unmute()
self._agentCrash(i, quiet=True)
return
if ("registerInitialState" in dir(agent)):
self.mute(i)
if self.catchExceptions:
try:
timed_func = TimeoutFunction(agent.registerInitialState, int(self.rules.getMaxStartupTime(i)))
try:
start_time = time.time()
timed_func(self.state.deepCopy())
time_taken = time.time() - start_time
self.totalAgentTimes[i] += time_taken
except TimeoutFunctionException:
print >>sys.stderr, "Agent %d ran out of time on startup!" % i
self.unmute()
self.agentTimeout = True
self._agentCrash(i, quiet=True)
return
except Exception,data:
self._agentCrash(i, quiet=False)
self.unmute()
return
else:
agent.registerInitialState(self.state.deepCopy())
## TODO: could this exceed the total time
self.unmute()
agentIndex = self.startingIndex
numAgents = len( self.agents )
while not self.gameOver:
# Fetch the next agent
agent = self.agents[agentIndex]
move_time = 0
skip_action = False
# Generate an observation of the state
if 'observationFunction' in dir( agent ):
self.mute(agentIndex)
if self.catchExceptions:
try:
timed_func = TimeoutFunction(agent.observationFunction, int(self.rules.getMoveTimeout(agentIndex)))
try:
start_time = time.time()
observation = timed_func(self.state.deepCopy())
except TimeoutFunctionException:
skip_action = True
move_time += time.time() - start_time
self.unmute()
except Exception,data:
self._agentCrash(agentIndex, quiet=False)
self.unmute()
return
else:
observation = agent.observationFunction(self.state.deepCopy())
self.unmute()
else:
observation = self.state.deepCopy()
# Solicit an action
action = None
self.mute(agentIndex)
if self.catchExceptions:
try:
timed_func = TimeoutFunction(agent.getAction, int(self.rules.getMoveTimeout(agentIndex)) - int(move_time))
try:
start_time = time.time()
if skip_action:
raise TimeoutFunctionException()
action = timed_func( observation )
except TimeoutFunctionException:
print >>sys.stderr, "Agent %d timed out on a single move!" % agentIndex
self.agentTimeout = True
self._agentCrash(agentIndex, quiet=True)
self.unmute()
return
move_time += time.time() - start_time
if move_time > self.rules.getMoveWarningTime(agentIndex):
self.totalAgentTimeWarnings[agentIndex] += 1
print >>sys.stderr, "Agent %d took too long to make a move! This is warning %d" % (agentIndex, self.totalAgentTimeWarnings[agentIndex])
if self.totalAgentTimeWarnings[agentIndex] > self.rules.getMaxTimeWarnings(agentIndex):
print >>sys.stderr, "Agent %d exceeded the maximum number of warnings: %d" % (agentIndex, self.totalAgentTimeWarnings[agentIndex])
self.agentTimeout = True
self._agentCrash(agentIndex, quiet=True)
self.unmute()
return
self.totalAgentTimes[agentIndex] += move_time
#print "Agent: %d, time: %f, total: %f" % (agentIndex, move_time, self.totalAgentTimes[agentIndex])
if self.totalAgentTimes[agentIndex] > self.rules.getMaxTotalTime(agentIndex):
print >>sys.stderr, "Agent %d ran out of time! (time: %1.2f)" % (agentIndex, self.totalAgentTimes[agentIndex])
self.agentTimeout = True
self._agentCrash(agentIndex, quiet=True)
self.unmute()
return
self.unmute()
except Exception,data:
self._agentCrash(agentIndex)
self.unmute()
return
else:
action = agent.getAction(observation)
self.unmute()
# Execute the action
self.moveHistory.append( (agentIndex, action) )
if self.catchExceptions:
try:
self.state = self.state.generateSuccessor( agentIndex, action )
except Exception,data:
self.mute(agentIndex)
self._agentCrash(agentIndex)
self.unmute()
return
else:
self.state = self.state.generateSuccessor( agentIndex, action )
# Change the display
self.display.update( self.state.data )
###idx = agentIndex - agentIndex % 2 + 1
###self.display.update( self.state.makeObservation(idx).data )
# Allow for game specific conditions (winning, losing, etc.)
self.rules.process(self.state, self)
# Track progress
if agentIndex == numAgents + 1: self.numMoves += 1
# Next agent
agentIndex = ( agentIndex + 1 ) % numAgents
if _BOINC_ENABLED:
boinc.set_fraction_done(self.getProgress())
# inform a learning agent of the game result
for agentIndex, agent in enumerate(self.agents):
if "final" in dir( agent ) :
try:
self.mute(agentIndex)
agent.final( self.state )
self.unmute()
except Exception,data:
if not self.catchExceptions: raise
self._agentCrash(agentIndex)
self.unmute()
return
self.display.finish()