人工智能Pacman(三)(2018-05-24)

# 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__):

                if 'drawExpandedCells' in dir(__main__._display): #@UndefinedVariable

                    __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

        costFn = lambda pos: .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

        costFn = lambda pos: 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 ***"

        util.raiseNotDefined()

    def isGoalState(self, state):

        """

        Returns whether this search state is a goal state of the problem.

        """

        "*** YOUR CODE HERE ***"

        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:

            #  x,y = currentPosition

            #  dx, dy = Actions.directionToVector(action)

            #  nextx, nexty = int(x + dx), int(y + dy)

            #  hitsWall = self.walls[nextx][nexty]

            "*** YOUR CODE HERE ***"

        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

    walls = problem.walls # These are the walls of the maze, as a Grid (game.py)

    "*** YOUR CODE HERE ***"

    return 0 # 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 ***"

    return 0

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))

你可能感兴趣的:(人工智能Pacman(三)(2018-05-24))