def followActionBasis(self, observedState, action):
allowedDirections = observedState.getLegalPacmanActions()
pacmanPosition = observedState.getPacmanPosition()
if action == 'good':
target = classifyGhosts.getNearestGoodGhost(observedState, self.distancer)
elif action == 'bad':
target = classifyGhosts.getNearestBadGhost(observedState, self.distancer)
elif action == 'capsule':
target = classifyCapsule.closest(observedState, self.distancer)
else:
raise Exception("Unknown action '%s' passed to action basis" % action)
# find the direction that moves closes to target
best_action = Directions.STOP
best_dist = np.inf
for la in allowedDirections:
if la == Directions.STOP:
continue
successor_pos = Actions.getSuccessor(pacmanPosition,la)
new_dist = self.distancer.getDistance(successor_pos,target)
if new_dist < best_dist:
best_action = la
best_dist = new_dist
return best_action
def chooseAction(self, observedState):
"""
Pacman will eat the nearest good capsule if the ghost is not scared,
and chase the scared ghost otherwise.
"""
pacmanPos = observedState.getPacmanPosition()
ghost_states = observedState.getGhostStates() # states have getPosition() and getFeatures() methods
legalActs = [a for a in observedState.getLegalPacmanActions()]
# find the closest ghost by sorting the distances
closeGhost = ghosts.closestGhost(observedState, self.distancer, good=False)
# find the closest bad ghost by sorting the distances
badGhost = ghosts.getNearestBadGhost(observedState, self.distancer)
# position of closest good capsule to Pacman
closeCapsule = capsule.closest(observedState, self.distancer)
best_action = random.choice(legalActs)
if observedState.scaredGhostPresent():
# targets capsules and good ghosts on the way to the bad ghost
target = mapH.subTarget(pacmanPos, badGhost, closeCapsule)
target = mapH.subTarget(pacmanPos, target, closeGhost)
for dir in mapH.getDirs(pacmanPos, target):
if dir in legalActs:
return dir
else:
# sorts out directions that lead to any close ghost
legalActs = [a for a in legalActs if a not in mapH.ghostDirs(pacmanPos, closeGhost)]
for dir in mapH.getDirs(pacmanPos, closeCapsule):
if dir in legalActs:
return dir
return best_action
def chooseAction(self, observedState):
"""
Pacman will eat the nearest good capsule if the ghost is not scared,
and eat good ghosts otherwise. Along the way, he will adjust his path to
run into good ghosts.
"""
pacmanPos = observedState.getPacmanPosition()
legalActs = [a for a in observedState.getLegalPacmanActions()]
# find the closest ghost by sorting the distances
goodGhost = ghosts.closestGhost(observedState, self.distancer, good=True)
# position of closest good capsule to Pacman
closeCapsule = capsule.closest(observedState, self.distancer)
best_action = random.choice(legalActs)
if observedState.scaredGhostPresent():
# targets capsules along the way
target = mapH.subTarget(pacmanPos, goodGhost, closeCapsule)
for dir in mapH.getDirs(pacmanPos, target):
if dir in legalActs:
return dir
else:
# find the closest bad ghost by sorting the distances
badGhost = ghosts.getNearestBadGhost(observedState, self.distancer)
# sorts out directions that lead to the bad ghost
legalActs = [a for a in legalActs if a not in mapH.ghostDirs(pacmanPos, badGhost)]
# targets good ghosts along the way
target = mapH.subTarget(pacmanPos, closeCapsule, goodGhost)
for dir in mapH.getDirs(pacmanPos, closeCapsule):
if dir in legalActs:
return dir
return best_action
def goodBadGhostCapsuleDistances(self,observedState): pacmanPosition = observedState.getPacmanPosition() goodGhost = classifyGhosts.getNearestGoodGhost(observedState, self.distancer) badGhost = classifyGhosts.getNearestBadGhost(observedState, self.distancer) capsule = classifyCapsule.closest(observedState, self.distancer) distBinner = xbinner( \ (0,math.sqrt(observedState.layout.width**2 + observedState.layout.width**2)), \ goodBadGhostCapsuleDistances.buckets ) goodGhostDistance = distBinner(self.distancer.getDistance(pacmanPosition, goodGhost)) badGhostDistance = distBinner(self.distancer.getDistance(pacmanPosition, badGhost)) capsuleDistance = distBinner(self.distancer.getDistance(pacmanPosition, capsule)) hasCapsule = observedState.scaredGhostPresent() return (goodGhostDistance,badGhostDistance,capsuleDistance,int(hasCapsule))
def chooseAction(self, observedState):
"Pacman will eat the nearest bad ghost. For testing purposes."
pacmanPos = observedState.getPacmanPosition()
ghost_states = observedState.getGhostStates() # states have getPosition() and getFeatures() methods
legalActs = [a for a in observedState.getLegalPacmanActions()]
# position of closest good capsule to Pacman
badGhost = ghosts.getNearestBadGhost(observedState, self.distancer)
best_action = random.choice(legalActs)
for dir in mapH.getDirs(pacmanPos, badGhost):
if dir in legalActs:
return dir
return best_action
def localNeighborhood(self, observedState): radius = 4 pacmanPosition = observedState.getPacmanPosition() xBinner = xbinner( (pacmanPosition[0]-radius, pacmanPosition[0]+radius), localNeighborhood.buckets ) yBinner = xbinner( (pacmanPosition[1]-radius, pacmanPosition[1]+radius), localNeighborhood.buckets ) goodGhost = classifyGhosts.getNearestGoodGhost(observedState, self.distancer) badGhost = classifyGhosts.getNearestBadGhost(observedState, self.distancer) capsule = classifyCapsule.closest(observedState, self.distancer) hasCapsule = observedState.scaredGhostPresent() print pacmanPosition, goodGhost, badGhost, capsule, hasCapsule # bin the x/y position for each object of interest b = [] for p in [goodGhost, badGhost, capsule]: b += [xBinner(p[0]), yBinner(p[1])] return tuple(b + [int(hasCapsule)])
def ghostPosition(self,observedState): pacmanPosition = observedState.getPacmanPosition() ghost_positions = [ classifyGhosts.getNearestGoodGhost(observedState, self.distancer), classifyGhosts.getNearestBadGhost(observedState, self.distancer)] bins = 5 xBinner = xbinner( (0,observedState.layout.width), bins ) yBinner = xbinner( (0,observedState.layout.height), bins ) state = (xBinner(pacmanPosition[0]), yBinner(pacmanPosition[1])) for pos in ghost_positions: state += (xBinner(pos[0]),yBinner(pos[1])) return state
