ivar_def_reflex_agent.py

from captureAgents import CaptureAgent
from util import nearestPoint
import random, time, util
from game import Directions

class ReflexCaptureAgent(CaptureAgent):
  """
  A base class for reflex agents that chooses score-maximizing actions
  """
 
  def registerInitialState(self, gameState):
    self.start = gameState.getAgentPosition(self.index)
    CaptureAgent.registerInitialState(self, gameState)

  def chooseAction(self, gameState):
    """
    Picks among the actions with the highest Q(s,a).
    """
    actions = gameState.getLegalActions(self.index)

    # You can profile your evaluation time by uncommenting these lines
    # start = time.time()
    values = [self.evaluate(gameState, a) for a in actions]
    # print('eval time for agent %d: %.4f' % (self.index, time.time() - start))

    maxValue = max(values)
    bestActions = [a for a, v in zip(actions, values) if v == maxValue]

    foodLeft = len(self.getFood(gameState).asList())

    if foodLeft <= 2:
      bestDist = 9999
      for action in actions:
        successor = self.getSuccessor(gameState, action)
        pos2 = successor.getAgentPosition(self.index)
        dist = self.getMazeDistance(self.start,pos2)
        if dist < bestDist:
          bestAction = action
          bestDist = dist
      return bestAction

    return random.choice(bestActions)

  def getSuccessor(self, gameState, action):
    """
    Finds the next successor which is a grid position (location tuple).
    """
    successor = gameState.generateSuccessor(self.index, action)
    pos = successor.getAgentState(self.index).getPosition()
    if pos != nearestPoint(pos):
      # Only half a grid position was covered
      return successor.generateSuccessor(self.index, action)
    else:
      return successor

  def evaluate(self, gameState, action):
    """
    Computes a linear combination of features and feature weights
    """
    features = self.getFeatures(gameState, action)
    weights = self.getWeights(gameState, action)
    return features * weights

  def getFeatures(self, gameState, action):
    """
    Returns a counter of features for the state
    """
    features = util.Counter()
    successor = self.getSuccessor(gameState, action)
    features['successorScore'] = self.getScore(successor)
    return features

  def getWeights(self, gameState, action):
    """
    Normally, weights do not depend on the gamestate.  They can be either
    a counter or a dictionary.
    """
    return {'successorScore': 1.0}

class DefensiveReflexAgent(ReflexCaptureAgent):
  """
  A reflex agent that keeps its side Pacman-free. Again,
  this is to give you an idea of what a defensive agent
  could be like.  It is not the best or only way to make
  such an agent.
  """

  def getFeatures(self, gameState, action):
    features = util.Counter()
    successor = self.getSuccessor(gameState, action)

    myState = successor.getAgentState(self.index)
    myPos = myState.getPosition()

    # Computes whether we're on defense (1) or offense (0)
    features['onDefense'] = 1
    if myState.isPacman: features['onDefense'] = 0

    # Computes distance to invaders we can see
    enemies = [successor.getAgentState(i) for i in self.getOpponents(successor)]
    invaders = [a for a in enemies if a.isPacman and a.getPosition() != None]
    features['numInvaders'] = len(invaders)
    if len(invaders) > 0:
      dists = [self.getMazeDistance(myPos, a.getPosition()) for a in invaders]
      features['invaderDistance'] = min(dists)

    if action == Directions.STOP: features['stop'] = 1
    rev = Directions.REVERSE[gameState.getAgentState(self.index).configuration.direction]
    if action == rev: features['reverse'] = 1

    return features

  def getWeights(self, gameState, action):
    return {'numInvaders': -1000, 'onDefense': 100, 'invaderDistance': -10, 'stop': -100, 'reverse': -2}