CC BY 4.0 （除特别声明或转载文章外）
如果这篇博客帮助到你，可以请我喝一杯咖啡~
The Pac-Man projects were developed for UC Berkeley’s introductory artificial intelligence course, CS 188. They apply an array of AI techniques to playing Pac-Man. However, these projects don’t focus on building AI for video games. Instead, they teach foundational AI concepts, such as informed state-space search, probabilistic inference, and reinforcement learning. These concepts underly real-world application areas such as natural language processing, computer vision, and robotics.
We designed these projects with three goals in mind. The projects allow students to visualize the results of the techniques they implement. They also contain code examples and clear directions, but do not force students to wade through undue amounts of scaffolding. Finally, Pac-Man provides a challenging problem environment that demands creative solutions; real-world AI problems are challenging, and Pac-Man is too.
In our course, these projects have boosted enrollment, teaching reviews, and student engagement. The projects have been field-tested, refined, and debugged over multiple semesters at Berkeley. We are now happy to release them to other universities for educational use. )
Search
search.py
由于要实现很多不同的搜索，这里封装了一个frontierSearch类，根据传入容器的不同（队列、栈、优先队列），就可以直接实现不同的搜索方式（广度优先、深度优先、花费优先等等）。
# search.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
# ([email protected]) and Dan Klein ([email protected]).
# Student side autograding was added by Brad Miller, Nick Hay, and
# Pieter Abbeel ([email protected]).
"""
In search.py, you will implement generic search algorithms which are called by
Pacman agents (in searchAgents.py).
"""
import util
class SearchProblem:
"""
This class outlines the structure of a search problem, but doesn't implement
any of the methods (in object-oriented terminology: an abstract class).
You do not need to change anything in this class, ever.
"""
def getStartState(self):
"""
Returns the start state for the search problem.
"""
util.raiseNotDefined()
def isGoalState(self, state):
"""
state: Search state
Returns True if and only if the state is a valid goal state.
"""
util.raiseNotDefined()
def getSuccessors(self, state):
"""
state: Search state
For a given state, this should return a list of triples, (successor,
action, stepCost), where 'successor' is a successor to the current
state, 'action' is the action required to get there, and 'stepCost' is
the incremental cost of expanding to that successor.
"""
util.raiseNotDefined()
def getCostOfActions(self, actions):
"""
actions: A list of actions to take
This method returns the total cost of a particular sequence of actions.
The sequence must be composed of legal moves.
"""
util.raiseNotDefined()
def tinyMazeSearch(problem):
"""
Returns a sequence of moves that solves tinyMaze. For any other maze, the
sequence of moves will be incorrect, so only use this for tinyMaze.
"""
from game import Directions
s = Directions.SOUTH
w = Directions.WEST
return [s, s, w, s, w, w, s, w]
def frontierSearch(problem, frontier):
frontier.push(([], problem.getStartState()))
visited = []
while not frontier.isEmpty():
moves, curr = frontier.pop()
if problem.isGoalState(curr):
return moves
if curr not in visited:
visited.append(curr)
for successor, action, _stepCost in problem.getSuccessors(curr):
frontier.push((moves + [action], successor))
def depthFirstSearch(problem):
"""
Search the deepest nodes in the search tree first.
Your search algorithm needs to return a list of actions that reaches the
goal. Make sure to implement a graph search algorithm.
To get started, you might want to try some of these simple commands to
understand the search problem that is being passed in:
print "Start:", problem.getStartState()
print "Is the start a goal?", problem.isGoalState(problem.getStartState())
print "Start's successors:", problem.getSuccessors(problem.getStartState())
"""
"*** YOUR CODE HERE ***"
return frontierSearch(problem, util.Stack())
# util.raiseNotDefined()
def breadthFirstSearch(problem):
"""Search the shallowest nodes in the search tree first."""
"*** YOUR CODE HERE ***"
return frontierSearch(problem, util.Queue())
# util.raiseNotDefined()
def uniformCostSearch(problem):
"""Search the node of least total cost first."""
"*** YOUR CODE HERE ***"
return aStarSearch(problem, nullHeuristic)
# util.raiseNotDefined()
def nullHeuristic(state, problem=None):
"""
A heuristic function estimates the cost from the current state to the nearest
goal in the provided SearchProblem. This heuristic is trivial.
"""
return 0
def aStarSearch(problem, heuristic=nullHeuristic):
"""Search the node that has the lowest combined cost and heuristic first."""
"*** YOUR CODE HERE ***"
def cost((moves, pos)):
return problem.getCostOfActions(moves) + heuristic(pos, problem)
return frontierSearch(problem, util.PriorityQueueWithFunction(cost))
# util.raiseNotDefined()
# Abbreviations
bfs = breadthFirstSearch
dfs = depthFirstSearch
astar = aStarSearch
ucs = uniformCostSearch
searchAgents.py
# searchAgents.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
# ([email protected]) and Dan Klein ([email protected]).
# Student side autograding was added by Brad Miller, Nick Hay, and
# Pieter Abbeel ([email protected]).
"""
This file contains all of the agents that can be selected to control Pacman. To
select an agent, use the '-p' option when running pacman.py. Arguments can be
passed to your agent using '-a'. For example, to load a SearchAgent that uses
depth first search (dfs), run the following command:
> python pacman.py -p SearchAgent -a fn=depthFirstSearch
Commands to invoke other search strategies can be found in the project
description.
Please only change the parts of the file you are asked to. Look for the lines
that say
"*** YOUR CODE HERE ***"
The parts you fill in start about 3/4 of the way down. Follow the project
description for details.
Good luck and happy searching!
"""
from game import Directions
from game import Agent
from game import Actions
import util
import time
import search
class GoWestAgent(Agent):
"An agent that goes West until it can't."
def getAction(self, state):
"The agent receives a GameState (defined in pacman.py)."
if Directions.WEST in state.getLegalPacmanActions():
return Directions.WEST
else:
return Directions.STOP
#######################################################
# This portion is written for you, but will only work #
# after you fill in parts of search.py #
#######################################################
class SearchAgent(Agent):
"""
This very general search agent finds a path using a supplied search
algorithm for a supplied search problem, then returns actions to follow that
path.
As a default, this agent runs DFS on a PositionSearchProblem to find
location (1,1)
Options for fn include:
depthFirstSearch or dfs
breadthFirstSearch or bfs
Note: You should NOT change any code in SearchAgent
"""
def __init__(self, fn='depthFirstSearch', prob='PositionSearchProblem', heuristic='nullHeuristic'):
# Warning: some advanced Python magic is employed below to find the right functions and problems
# Get the search function from the name and heuristic
if fn not in dir(search):
raise AttributeError, fn + ' is not a search function in search.py.'
func = getattr(search, fn)
if 'heuristic' not in func.func_code.co_varnames:
print('[SearchAgent] using function ' + fn)
self.searchFunction = func
else:
if heuristic in globals().keys():
heur = globals()[heuristic]
elif heuristic in dir(search):
heur = getattr(search, heuristic)
else:
raise AttributeError, heuristic + ' is not a function in searchAgents.py or search.py.'
print('[SearchAgent] using function %s and heuristic %s' %
(fn, heuristic))
# Note: this bit of Python trickery combines the search algorithm and the heuristic
self.searchFunction = lambda x: func(x, heuristic=heur)
# Get the search problem type from the name
if prob not in globals().keys() or not prob.endswith('Problem'):
raise AttributeError, prob + ' is not a search problem type in SearchAgents.py.'
self.searchType = globals()[prob]
print('[SearchAgent] using problem type ' + prob)
def registerInitialState(self, state):
"""
This is the first time that the agent sees the layout of the game
board. Here, we choose a path to the goal. In this phase, the agent
should compute the path to the goal and store it in a local variable.
All of the work is done in this method!
state: a GameState object (pacman.py)
"""
if self.searchFunction == None:
raise Exception, "No search function provided for SearchAgent"
starttime = time.time()
problem = self.searchType(state) # Makes a new search problem
self.actions = self.searchFunction(problem) # Find a path
totalCost = problem.getCostOfActions(self.actions)
print('Path found with total cost of %d in %.1f seconds' %
(totalCost, time.time() - starttime))
if '_expanded' in dir(problem):
print('Search nodes expanded: %d' % problem._expanded)
def getAction(self, state):
"""
Returns the next action in the path chosen earlier (in
registerInitialState). Return Directions.STOP if there is no further
action to take.
state: a GameState object (pacman.py)
"""
if 'actionIndex' not in dir(self):
self.actionIndex = 0
i = self.actionIndex
self.actionIndex += 1
if i < len(self.actions):
return self.actions[i]
else:
return Directions.STOP
class PositionSearchProblem(search.SearchProblem):
"""
A search problem defines the state space, start state, goal test, successor
function and cost function. This search problem can be used to find paths
to a particular point on the pacman board.
The state space consists of (x,y) positions in a pacman game.
Note: this search problem is fully specified; you should NOT change it.
"""
def __init__(self, gameState, costFn=lambda x: 1, goal=(1, 1), start=None, warn=True, visualize=True):
"""
Stores the start and goal.
gameState: A GameState object (pacman.py)
costFn: A function from a search state (tuple) to a non-negative number
goal: A position in the gameState
"""
self.walls = gameState.getWalls()
self.startState = gameState.getPacmanPosition()
if start != None:
self.startState = start
self.goal = goal
self.costFn = costFn
self.visualize = visualize
if warn and (gameState.getNumFood() != 1 or not gameState.hasFood(*goal)):
print 'Warning: this does not look like a regular search maze'
# For display purposes
self._visited, self._visitedlist, self._expanded = {}, [], 0 # DO NOT CHANGE
def getStartState(self):
return self.startState
def isGoalState(self, state):
isGoal = state == self.goal
# For display purposes only
if isGoal and self.visualize:
self._visitedlist.append(state)
import __main__
if '_display' in dir(__main__):
# @UndefinedVariable
if 'drawExpandedCells' in dir(__main__._display):
__main__._display.drawExpandedCells(
self._visitedlist) # @UndefinedVariable
return isGoal
def getSuccessors(self, state):
"""
Returns successor states, the actions they require, and a cost of 1.
As noted in search.py:
For a given state, this should return a list of triples,
(successor, action, stepCost), where 'successor' is a
successor to the current state, 'action' is the action
required to get there, and 'stepCost' is the incremental
cost of expanding to that successor
"""
successors = []
for action in [Directions.NORTH, Directions.SOUTH, Directions.EAST, Directions.WEST]:
x, y = state
dx, dy = Actions.directionToVector(action)
nextx, nexty = int(x + dx), int(y + dy)
if not self.walls[nextx][nexty]:
nextState = (nextx, nexty)
cost = self.costFn(nextState)
successors.append((nextState, action, cost))
# Bookkeeping for display purposes
self._expanded += 1 # DO NOT CHANGE
if state not in self._visited:
self._visited[state] = True
self._visitedlist.append(state)
return successors
def getCostOfActions(self, actions):
"""
Returns the cost of a particular sequence of actions. If those actions
include an illegal move, return 999999.
"""
if actions == None:
return 999999
x, y = self.getStartState()
cost = 0
for action in actions:
# Check figure out the next state and see whether its' legal
dx, dy = Actions.directionToVector(action)
x, y = int(x + dx), int(y + dy)
if self.walls[x][y]:
return 999999
cost += self.costFn((x, y))
return cost
class StayEastSearchAgent(SearchAgent):
"""
An agent for position search with a cost function that penalizes being in
positions on the West side of the board.
The cost function for stepping into a position (x,y) is 1/2^x.
"""
def __init__(self):
self.searchFunction = search.uniformCostSearch
def costFn(pos): return .5 ** pos[0]
self.searchType = lambda state: PositionSearchProblem(
state, costFn, (1, 1), None, False)
class StayWestSearchAgent(SearchAgent):
"""
An agent for position search with a cost function that penalizes being in
positions on the East side of the board.
The cost function for stepping into a position (x,y) is 2^x.
"""
def __init__(self):
self.searchFunction = search.uniformCostSearch
def costFn(pos): return 2 ** pos[0]
self.searchType = lambda state: PositionSearchProblem(state, costFn)
def manhattanHeuristic(position, problem, info={}):
"The Manhattan distance heuristic for a PositionSearchProblem"
xy1 = position
xy2 = problem.goal
return abs(xy1[0] - xy2[0]) + abs(xy1[1] - xy2[1])
def euclideanHeuristic(position, problem, info={}):
"The Euclidean distance heuristic for a PositionSearchProblem"
xy1 = position
xy2 = problem.goal
return ((xy1[0] - xy2[0]) ** 2 + (xy1[1] - xy2[1]) ** 2) ** 0.5
#####################################################
# This portion is incomplete. Time to write code! #
#####################################################
class CornersProblem(search.SearchProblem):
"""
This search problem finds paths through all four corners of a layout.
You must select a suitable state space and successor function
"""
def __init__(self, startingGameState):
"""
Stores the walls, pacman's starting position and corners.
"""
self.walls = startingGameState.getWalls()
self.startingPosition = startingGameState.getPacmanPosition()
top, right = self.walls.height-2, self.walls.width-2
self.corners = ((1, 1), (1, top), (right, 1), (right, top))
for corner in self.corners:
if not startingGameState.hasFood(*corner):
print 'Warning: no food in corner ' + str(corner)
self._expanded = 0 # DO NOT CHANGE; Number of search nodes expanded
# Please add any code here which you would like to use
# in initializing the problem
"*** YOUR CODE HERE ***"
def getStartState(self):
"""
Returns the start state (in your state space, not the full Pacman state
space)
"""
"*** YOUR CODE HERE ***"
return (self.startingPosition, [])
# util.raiseNotDefined()
def isGoalState(self, state):
"""
Returns whether this search state is a goal state of the problem.
"""
"*** YOUR CODE HERE ***"
pos = state[0]
cnt = len(state[1])
if pos in self.corners:
if pos not in state[1]:
cnt = cnt+1
return cnt == 4
return False
# util.raiseNotDefined()
def getSuccessors(self, state):
"""
Returns successor states, the actions they require, and a cost of 1.
As noted in search.py:
For a given state, this should return a list of triples, (successor,
action, stepCost), where 'successor' is a successor to the current
state, 'action' is the action required to get there, and 'stepCost'
is the incremental cost of expanding to that successor
"""
successors = []
for action in [Directions.NORTH, Directions.SOUTH, Directions.EAST, Directions.WEST]:
# Add a successor state to the successor list if the action is legal
# Here's a code snippet for figuring out whether a new position hits a wall:
"*** YOUR CODE HERE ***"
dx, dy = Actions.directionToVector(action)
nex = (int(state[0][0] + dx), int(state[0][1] + dy))
hitsWall = self.walls[nex[0]][nex[1]]
if not hitsWall:
path = list(state[1])
if nex in self.corners:
if nex not in path:
path.append(nex)
successors.append(((nex, path), action, 1))
self._expanded += 1 # DO NOT CHANGE
return successors
def getCostOfActions(self, actions):
"""
Returns the cost of a particular sequence of actions. If those actions
include an illegal move, return 999999. This is implemented for you.
"""
if actions == None:
return 999999
x, y = self.startingPosition
for action in actions:
dx, dy = Actions.directionToVector(action)
x, y = int(x + dx), int(y + dy)
if self.walls[x][y]:
return 999999
return len(actions)
def cornersHeuristic(state, problem):
"""
A heuristic for the CornersProblem that you defined.
state: The current search state
(a data structure you chose in your search problem)
problem: The CornersProblem instance for this layout.
This function should always return a number that is a lower bound on the
shortest path from the state to a goal of the problem; i.e. it should be
admissible (as well as consistent).
"""
corners = problem.corners # These are the corner coordinates
# These are the walls of the maze, as a Grid (game.py)
walls = problem.walls
"*** YOUR CODE HERE ***"
h_sum = 0
unVis = []
cur = state[0]
for i in range(4):
if corners[i] not in state[1]:
unVis.append(corners[i])
while(len(unVis) != 0):
dis, corner = min([(util.manhattanDistance(
cur, corner), corner) for corner in unVis])
h_sum = h_sum + dis
cur = corner
unVis.remove(corner)
return h_sum # Default to trivial solution
class AStarCornersAgent(SearchAgent):
"A SearchAgent for FoodSearchProblem using A* and your foodHeuristic"
def __init__(self):
self.searchFunction = lambda prob: search.aStarSearch(
prob, cornersHeuristic)
self.searchType = CornersProblem
class FoodSearchProblem:
"""
A search problem associated with finding the a path that collects all of the
food (dots) in a Pacman game.
A search state in this problem is a tuple ( pacmanPosition, foodGrid ) where
pacmanPosition: a tuple (x,y) of integers specifying Pacman's position
foodGrid: a Grid (see game.py) of either True or False, specifying remaining food
"""
def __init__(self, startingGameState):
self.start = (startingGameState.getPacmanPosition(),
startingGameState.getFood())
self.walls = startingGameState.getWalls()
self.startingGameState = startingGameState
self._expanded = 0 # DO NOT CHANGE
self.heuristicInfo = {} # A dictionary for the heuristic to store information
def getStartState(self):
return self.start
def isGoalState(self, state):
return state[1].count() == 0
def getSuccessors(self, state):
"Returns successor states, the actions they require, and a cost of 1."
successors = []
self._expanded += 1 # DO NOT CHANGE
for direction in [Directions.NORTH, Directions.SOUTH, Directions.EAST, Directions.WEST]:
x, y = state[0]
dx, dy = Actions.directionToVector(direction)
nextx, nexty = int(x + dx), int(y + dy)
if not self.walls[nextx][nexty]:
nextFood = state[1].copy()
nextFood[nextx][nexty] = False
successors.append((((nextx, nexty), nextFood), direction, 1))
return successors
def getCostOfActions(self, actions):
"""Returns the cost of a particular sequence of actions. If those actions
include an illegal move, return 999999"""
x, y = self.getStartState()[0]
cost = 0
for action in actions:
# figure out the next state and see whether it's legal
dx, dy = Actions.directionToVector(action)
x, y = int(x + dx), int(y + dy)
if self.walls[x][y]:
return 999999
cost += 1
return cost
class AStarFoodSearchAgent(SearchAgent):
"A SearchAgent for FoodSearchProblem using A* and your foodHeuristic"
def __init__(self):
self.searchFunction = lambda prob: search.aStarSearch(
prob, foodHeuristic)
self.searchType = FoodSearchProblem
def foodHeuristic(state, problem):
"""
Your heuristic for the FoodSearchProblem goes here.
This heuristic must be consistent to ensure correctness. First, try to come
up with an admissible heuristic; almost all admissible heuristics will be
consistent as well.
If using A* ever finds a solution that is worse uniform cost search finds,
your heuristic is *not* consistent, and probably not admissible! On the
other hand, inadmissible or inconsistent heuristics may find optimal
solutions, so be careful.
The state is a tuple ( pacmanPosition, foodGrid ) where foodGrid is a Grid
(see game.py) of either True or False. You can call foodGrid.asList() to get
a list of food coordinates instead.
If you want access to info like walls, capsules, etc., you can query the
problem. For example, problem.walls gives you a Grid of where the walls
are.
If you want to *store* information to be reused in other calls to the
heuristic, there is a dictionary called problem.heuristicInfo that you can
use. For example, if you only want to count the walls once and store that
value, try: problem.heuristicInfo['wallCount'] = problem.walls.count()
Subsequent calls to this heuristic can access
problem.heuristicInfo['wallCount']
"""
position, foodGrid = state
"*** YOUR CODE HERE ***"
gs = problem.startingGameState
foodList = foodGrid.asList()
foodCount = len(foodList)
if foodCount == 1:
return mazeDistance(position, foodList[0], gs)
"Find the 'diam' of the points and the min distance to one of the end points of the diam"
diam = (0, 0)
dis = []
for i in range(foodCount):
dis.append(mazeDistance(position, foodList[i], gs))
for ii in range(i):
diam = max(
diam, (mazeDistance(foodList[i], foodList[ii], gs), -min(dis[i], dis[ii])))
return diam[0]-diam[1]
class ClosestDotSearchAgent(SearchAgent):
"Search for all food using a sequence of searches"
def registerInitialState(self, state):
self.actions = []
currentState = state
while(currentState.getFood().count() > 0):
nextPathSegment = self.findPathToClosestDot(
currentState) # The missing piece
self.actions += nextPathSegment
for action in nextPathSegment:
legal = currentState.getLegalActions()
if action not in legal:
t = (str(action), str(currentState))
raise Exception, 'findPathToClosestDot returned an illegal move: %s!\n%s' % t
currentState = currentState.generateSuccessor(0, action)
self.actionIndex = 0
print 'Path found with cost %d.' % len(self.actions)
def findPathToClosestDot(self, gameState):
"""
Returns a path (a list of actions) to the closest dot, starting from
gameState.
"""
# Here are some useful elements of the startState
startPosition = gameState.getPacmanPosition()
food = gameState.getFood()
walls = gameState.getWalls()
problem = AnyFoodSearchProblem(gameState)
"*** YOUR CODE HERE ***"
util.raiseNotDefined()
class AnyFoodSearchProblem(PositionSearchProblem):
"""
A search problem for finding a path to any food.
This search problem is just like the PositionSearchProblem, but has a
different goal test, which you need to fill in below. The state space and
successor function do not need to be changed.
The class definition above, AnyFoodSearchProblem(PositionSearchProblem),
inherits the methods of the PositionSearchProblem.
You can use this search problem to help you fill in the findPathToClosestDot
method.
"""
def __init__(self, gameState):
"Stores information from the gameState. You don't need to change this."
# Store the food for later reference
self.food = gameState.getFood()
# Store info for the PositionSearchProblem (no need to change this)
self.walls = gameState.getWalls()
self.startState = gameState.getPacmanPosition()
self.costFn = lambda x: 1
self._visited, self._visitedlist, self._expanded = {}, [], 0 # DO NOT CHANGE
def isGoalState(self, state):
"""
The state is Pacman's position. Fill this in with a goal test that will
complete the problem definition.
"""
x, y = state
"*** YOUR CODE HERE ***"
util.raiseNotDefined()
def mazeDistance(point1, point2, gameState):
"""
Returns the maze distance between any two points, using the search functions
you have already built. The gameState can be any game state -- Pacman's
position in that state is ignored.
Example usage: mazeDistance( (2,4), (5,6), gameState)
This might be a useful helper function for your ApproximateSearchAgent.
"""
x1, y1 = point1
x2, y2 = point2
walls = gameState.getWalls()
assert not walls[x1][y1], 'point1 is a wall: ' + str(point1)
assert not walls[x2][y2], 'point2 is a wall: ' + str(point2)
prob = PositionSearchProblem(
gameState, start=point1, goal=point2, warn=False, visualize=False)
return len(search.bfs(prob))
Multi-Agent Search
multiAgents.py
# multiAgents.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
# ([email protected]) and Dan Klein ([email protected]).
# Student side autograding was added by Brad Miller, Nick Hay, and
# Pieter Abbeel ([email protected]).
from util import manhattanDistance
from game import Directions
import random
import util
import sys
from game import Agent
class ReflexAgent(Agent):
"""
A reflex agent chooses an action at each choice point by examining
its alternatives via a state evaluation function.
The code below is provided as a guide. You are welcome to change
it in any way you see fit, so long as you don't touch our method
headers.
"""
def getAction(self, gameState):
"""
You do not need to change this method, but you're welcome to.
getAction chooses among the best options according to the evaluation function.
Just like in the previous project, getAction takes a GameState and returns
some Directions.X for some X in the set {North, South, West, East, Stop}
"""
# Collect legal moves and successor states
legalMoves = gameState.getLegalActions()
# Choose one of the best actions
scores = [self.evaluationFunction(
gameState, action) for action in legalMoves]
bestScore = max(scores)
bestIndices = [index for index in range(
len(scores)) if scores[index] == bestScore]
# Pick randomly among the best
chosenIndex = random.choice(bestIndices)
"Add more of your code here if you want to"
return legalMoves[chosenIndex]
def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current and proposed successor
GameStates (pacman.py) and returns a number, where higher numbers are better.
The code below extracts some useful information from the state, like the
remaining food (newFood) and Pacman position after moving (newPos).
newScaredTimes holds the number of moves that each ghost will remain
scared because of Pacman having eaten a power pellet.
Print out these variables to see what you're getting, then combine them
to create a masterful evaluation function.
"""
# Useful information you can extract from a GameState (pacman.py)
successorGameState = currentGameState.generatePacmanSuccessor(action)
newPos = successorGameState.getPacmanPosition()
newFood = successorGameState.getFood()
newGhostStates = successorGameState.getGhostStates()
newScaredTimes = [
ghostState.scaredTimer for ghostState in newGhostStates]
"*** YOUR CODE HERE ***"
if successorGameState.isWin():
return sys.maxint
if successorGameState.isLose():
return -sys.maxint
ghostPositions = [ghostState.getPosition(
) for ghostState in newGhostStates if ghostState.scaredTimer == 0]
if ghostPositions:
closestGhost = min([util.manhattanDistance(
newPos, ghostPos) for ghostPos in ghostPositions])
if closestGhost == 0:
return -sys.maxint
else:
return sys.maxint
closestFood = min([util.manhattanDistance(newPos, foodPos)
for foodPos in newFood.asList()])
return successorGameState.getScore() + sum(newScaredTimes) + 1.0 / (closestFood * closestGhost)
def scoreEvaluationFunction(currentGameState):
"""
This default evaluation function just returns the score of the state.
The score is the same one displayed in the Pacman GUI.
This evaluation function is meant for use with adversarial search agents
(not reflex agents).
"""
return currentGameState.getScore()
class MultiAgentSearchAgent(Agent):
"""
This class provides some common elements to all of your
multi-agent searchers. Any methods defined here will be available
to the MinimaxPacmanAgent, AlphaBetaPacmanAgent & ExpectimaxPacmanAgent.
You *do not* need to make any changes here, but you can if you want to
add functionality to all your adversarial search agents. Please do not
remove anything, however.
Note: this is an abstract class: one that should not be instantiated. It's
only partially specified, and designed to be extended. Agent (game.py)
is another abstract class.
"""
def __init__(self, evalFn='scoreEvaluationFunction', depth='2'):
self.index = 0 # Pacman is always agent index 0
self.evaluationFunction = util.lookup(evalFn, globals())
self.depth = int(depth)
class MinimaxAgent(MultiAgentSearchAgent):
"""
Your minimax agent (question 2)
"""
def getAction(self, gameState):
"""
Returns the minimax action from the current gameState using self.depth
and self.evaluationFunction.
Here are some method calls that might be useful when implementing minimax.
gameState.getLegalActions(agentIndex):
Returns a list of legal actions for an agent
agentIndex=0 means Pacman, ghosts are >= 1
gameState.generateSuccessor(agentIndex, action):
Returns the successor game state after an agent takes an action
gameState.getNumAgents():
Returns the total number of agents in the game
"""
"*** YOUR CODE HERE ***"
return self.MinimaxSearch(gameState, 1, 0) # util.raiseNotDefined()
def MinimaxSearch(self, gameState, currentDepth, agentIndex):
if agentIndex >= gameState.getNumAgents():
return self.MinimaxSearch(gameState, currentDepth+1, 0)
if currentDepth > self.depth or gameState.isWin() or gameState.isLose():
return self.evaluationFunction(gameState)
legalMoves = [action for action in gameState.getLegalActions(
agentIndex) if action != 'Stop']
scores = [self.MinimaxSearch(gameState.generateSuccessor(
agentIndex, action), currentDepth, agentIndex + 1) for action in legalMoves]
if agentIndex == 0:
bestScore = max(scores)
if currentDepth == 1: # pacman first move
bestIndices = [index for index in range(
len(scores)) if scores[index] == bestScore]
chosenIndex = random.choice(bestIndices)
return legalMoves[chosenIndex]
return bestScore
else:
return min(scores)
class AlphaBetaAgent(MultiAgentSearchAgent):
"""
Your minimax agent with alpha-beta pruning (question 3)
"""
def getAction(self, gameState):
"""
Returns the minimax action using self.depth and self.evaluationFunction
"""
"*** YOUR CODE HERE ***"
# util.raiseNotDefined()
return self.AlphaBeta(gameState, 1, 0, -sys.maxint, sys.maxint)
def AlphaBetaSearch(self, gameState, currentDepth, agentIndex, alpha, beta):
if agentIndex >= gameState.getNumAgents():
return self.AlphaBetaSearch(gameState, currentDepth+1, 0, alpha, beta)
if currentDepth > self.depth or gameState.isWin() or gameState.isLose():
return self.evaluationFunction(gameState)
legalMoves = [action for action in gameState.getLegalActions(
agentIndex) if action != 'Stop']
if agentIndex == 0:
if currentDepth == 1: # pacman first move
scores = [self.AlphaBetaSearch(gameState.generateSuccessor(
agentIndex, action), currentDepth, agentIndex + 1, alpha, beta) for action in legalMoves]
bestScore = max(scores)
bestIndices = [index for index in range(
len(scores)) if scores[index] == bestScore]
chosenIndex = random.choice(bestIndices)
return legalMoves[chosenIndex]
bestScore = -sys.maxint
for action in legalMoves:
bestScore = max(bestScore,
self.AlphaBetaSearch(gameState.generateSuccessor(agentIndex, action), currentDepth, agentIndex + 1, alpha, beta))
if bestScore >= beta:
return bestScore
alpha = max(alpha, bestScore)
return bestScore
else:
bestScore = sys.maxint
for action in legalMoves:
bestScore = min(bestScore,
self.AlphaBetaSearch(gameState.generateSuccessor(agentIndex, action), currentDepth, agentIndex + 1, alpha, beta))
if alpha >= bestScore:
return bestScore
beta = min(beta, bestScore)
return bestScore
class ExpectimaxAgent(MultiAgentSearchAgent):
"""
Your expectimax agent (question 4)
"""
def getAction(self, gameState):
"""
Returns the expectimax action using self.depth and self.evaluationFunction
All ghosts should be modeled as choosing uniformly at random from their
legal moves.
"""
"*** YOUR CODE HERE ***"
util.raiseNotDefined()
def betterEvaluationFunction(currentGameState):
"""
Your extreme ghost-hunting, pellet-nabbing, food-gobbling, unstoppable
evaluation function (question 5).
DESCRIPTION: <write something here so we know what you did>
"""
"*** YOUR CODE HERE ***"
util.raiseNotDefined()
# Abbreviation
better = betterEvaluationFunction