def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current `pacai.bin.pacman.PacmanGameState`
and an action, and returns a number, where higher numbers are better.
Make sure to understand the range of different values before you combine them
in your evaluation function.
"""
successorGameState = currentGameState.generatePacmanSuccessor(action)
# Useful information you can extract.
newPosition = successorGameState.getPacmanPosition()
newFood = successorGameState.getFood()
oldFood = currentGameState.getFood()
newGhostStates = successorGameState.getGhostStates()
distance = 0
if newFood.count() == oldFood.count():
distance = manhattan(newFood.asList()[0], newPosition)
for pos in newFood.asList():
if manhattan(pos, newPosition) < distance:
distance = manhattan(pos, newPosition)
for ghost in newGhostStates:
distance += 2**(2 - manhattan(ghost.getPosition(), newPosition))
return 0 - distance
def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current `pacai.bin.pacman.PacmanGameState`
and an action, and returns a number, where higher numbers are better.
Make sure to understand the range of different values before you combine them
in your evaluation function.
"""
successorGameState = currentGameState.generatePacmanSuccessor(action)
# Useful information you can extract.
# newPosition = successorGameState.getPacmanPosition()
# oldFood = currentGameState.getFood()
# newGhostStates = successorGameState.getGhostStates()
# newScaredTimes = [ghostState.getScaredTimer() for ghostState in newGhostStates]
newPosition = successorGameState.getPacmanPosition()
newGhostStates = successorGameState.getGhostStates()
# found out through trial and error, and hardcoding the constants
# this gives us the average ghost distance
ghostDistance = sum([
manhattan(newPosition, ghostState.getPosition())
for ghostState in newGhostStates
]) / (currentGameState.getNumAgents() - 1)
oldFood = successorGameState.getFood()
foodDistances = [
manhattan(newPosition, food) for food in oldFood.asList()
] + [10000]
return successorGameState.getScore() + 10 / (min(foodDistances) + 0.0001) - \
10 / (ghostDistance + 0.0001)
def offenseEvaluationFunction(self, currentGameState, action):
successorGameState = currentGameState.generateSuccessor(self.index, action)
# Useful information you can extract.
newPosition = successorGameState.getAgentPosition(self.index)
oldFood = self.getOppFood(currentGameState)
newGhostStates = self.getOppGhostState(successorGameState)
newGhostPositions = self.getOppGhost(successorGameState)
oldOppPacmanPositions = self.getOppPacman(currentGameState)
score = currentGameState.getScore() * 100 # weighted current score
closestDistToFood = float('-inf')
if newPosition in oldOppPacmanPositions:
return 10000
if self.onOppositeSide(newPosition, successorGameState):
score += 100
for food in oldFood.asList():
dist = distance.maze(newPosition, food, currentGameState)
if dist == 0: # check if the new position is food
score += 1000
else: # add reciprocol manhattan dist to food (b/c close food = more points) * weight
score += (1 / dist) * 100
# calculate ghost score
eat_ghost = 1 # coefficient to switch values if ghost are edible
for i in range(len(newGhostPositions)):
if newGhostStates[i].getScaredTimer() > 0: # if ghosts edible, change weight to inverse reciprocol
eat_ghost = -0.1
if distance.manhattan(newGhostPositions[i], newPosition) <= 1: # if near ghost run away
score -= 50 * eat_ghost
else: # subtract reciprocol ghost distance (b/c closer ghost = less points) * weight
score -= (1 / distance.manhattan(newGhostPositions[i], newPosition)) * 50 * eat_ghost
return score
def ABEvaluationFunction(self, currentGameState):
food = self.getOppFood(currentGameState).asList()
minManhattanFood = None
minFoodDist = 999999
for f in food:
curDist = distance.manhattan(f, self.getPosition(currentGameState))
if curDist < minFoodDist:
minFoodDist = curDist
minManhattanFood = f
features = {
"closestDistToFood":
distance.maze(minManhattanFood, self.getPosition(currentGameState),
currentGameState),
"closestDistToGhost":
min([
distance.manhattan(self.getPosition(currentGameState), ghost)
for ghost in self.getOppGhost(currentGameState)
])
}
weights = {"closestDistToFood": -1, "closestDistToGhost": 100}
return sum(
[features[feat] * weights[feat] for feat in features.keys()])
def betterEvaluationFunction(currentGameState):
"""
Your extreme ghost-hunting, pellet-nabbing, food-gobbling, unstoppable evaluation function.
DESCRIPTION: <write something here so we know what you did>
This is nearly the same as the reflex one, but with modified constants,
and I added food and capsule distances
"""
oldFood = currentGameState.getFood()
foodDistances = [
manhattan(currentGameState.getPacmanPosition(), food)
for food in oldFood.asList()
] + [10000]
minDist = min(foodDistances)
newGhostStates = currentGameState.getGhostStates()
ghostDistance = sum([
manhattan(currentGameState.getPacmanPosition(),
ghostState.getPosition()) for ghostState in newGhostStates
]) / (currentGameState.getNumAgents() - 1)
capsules = currentGameState.getCapsules()
capsuleDistances = [
manhattan(currentGameState.getPacmanPosition(), capsule)
for capsule in capsules
] + [10000]
minCapDist = min(capsuleDistances)
return 100 * currentGameState.getScore() + 10 / (minDist + 0.001) +\
10 / (ghostDistance + 0.001) + 100 / (minCapDist + 0.001)
def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current `pacai.bin.pacman.PacmanGameState`
and an action, and returns a number, where higher numbers are better.
Make sure to understand the range of different values before you combine them
in your evaluation function.
"""
successorGameState = currentGameState.generatePacmanSuccessor(action)
# Useful information you can extract.
newPosition = successorGameState.getPacmanPosition()
oldFood = currentGameState.getFood()
newGhostStates = successorGameState.getGhostStates()
newScaredTimes = [ghostState.getScaredTimer() for ghostState in newGhostStates]
# *** Your Code Here ***
x = oldFood.getWidth()
y = oldFood.getHeight()
maxlen = distance.manhattan((x, y), (0, 0))
halflen = maxlen
minfooddist = maxlen
baddist = maxlen
badflag = True
for i in range(x):
for j in range(y):
m = oldFood[i][j]
if m:
d = distance.manhattan((i, j), newPosition)
if d < minfooddist:
minfooddist = d
for i in range(len(newScaredTimes)):
ghost = newGhostStates[i]
d = distance.manhattan(newPosition, ghost._position)
if newScaredTimes[i] == 0:
badflag = False
if d < baddist:
baddist = d
if baddist == 0:
return -100
if badflag:
baddist = 1
elif baddist > halflen:
baddist = 1
ev = (-1 / baddist) + (successorGameState.getScore() / 2)
if minfooddist == 0:
ev += 10
else:
ev += (2 / minfooddist)
if badflag:
ev += 1
elif baddist > halflen:
ev += 1
return ev
def ABEvaluationFunction(self, currentGameState):
food = self.getOppFood(currentGameState).asList()
minManhattanFood = None
minFoodDist = 999999
for f in food:
curDist = distance.manhattan(f, self.getPosition(currentGameState))
if curDist < minFoodDist:
minFoodDist = curDist
minManhattanFood = f
currentPos = self.getPosition(currentGameState)
numWalls = self.getNumWalls(currentPos, currentGameState)
# print("Num Walls =", numWalls)
ateCapsule = self.oppIsScared(currentGameState)
ghostDistMultiplier = 1
if ateCapsule:
# print ("ate capsule")
ghostDistMultiplier = -2
closestDistToGhost = 0
try:
closestDistToGhost = min([
distance.manhattan(self.getPosition(currentGameState), ghost)
for ghost in self.getOppGhost(currentGameState)
])
except:
pass
if closestDistToGhost > 10:
return -10000
features = {
"closestDistToFood":
distance.maze(minManhattanFood, self.getPosition(currentGameState),
currentGameState),
"closestDistToGhost":
min([
distance.manhattan(self.getPosition(currentGameState), ghost)
for ghost in self.getOppGhost(currentGameState)
]),
# "closestDistToGhost": closestDistToGhost,
"surroundingWalls":
numWalls
}
weights = {
"closestDistToFood": -50,
"closestDistToGhost": 100 * ghostDistMultiplier,
"surroundingWalls": -100
}
return sum(
[features[feat] * weights[feat] for feat in features.keys()])
def applyAction(state, action, agentIndex):
"""
Edits the state to reflect the results of the action.
"""
legal = AgentRules.getLegalActions(state, agentIndex)
if (action not in legal):
raise ValueError('Illegal action: ' + str(action))
agentState = state.getAgentState(agentIndex)
# Update position.
vector = Actions.directionToVector(action, AgentRules.AGENT_SPEED)
agentState.updatePosition(vector)
# Eat.
nextPosition = agentState.getPosition()
nearest = nearestPoint(nextPosition)
if (agentState.isPacman() and manhattan(nearest, nextPosition) <= 0.9):
AgentRules.consume(nearest, state, state.isOnRedTeam(agentIndex))
# Potentially change agent type.
if (nextPosition == nearest):
# Agents are pacmen when they are not on their own side.
position = agentState.getPosition()
agentState.setIsPacman(state.isOnRedTeam(agentIndex) != state.isOnRedSide(position))
def checkDeath(state, agentIndex):
agentState = state.getAgentState(agentIndex)
if (state.isOnRedTeam(agentIndex)):
teamPointModifier = 1
otherTeam = state.getBlueTeamIndices()
else:
teamPointModifier = -1
otherTeam = state.getRedTeamIndices()
for otherAgentIndex in otherTeam:
otherAgentState = state.getAgentState(otherAgentIndex)
# Ignore agents with a matching type (e.g. two ghosts).
if (agentState.isPacman() == otherAgentState.isPacman()):
continue
otherPosition = otherAgentState.getPosition()
# Ignore other agents that are too far away.
if (otherPosition is None
or manhattan(otherPosition, agentState.getPosition()) > COLLISION_TOLERANCE):
continue
# If we are a brave ghost or they are a scared ghost, then we will eat them.
# Otherwise, we are being eatten.
if (agentState.isBraveGhost() or otherAgentState.isScaredGhost()):
state.addScore(teamPointModifier * KILL_POINTS)
otherAgentState.respawn()
else:
state.addScore(teamPointModifier * -KILL_POINTS)
agentState.respawn()
def cornersHeuristic(state, problem):
"""
A heuristic for the CornersProblem that you defined.
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.
(You need not worry about consistency for this heuristic to receive full credit.)
"""
# Useful information.
corners = problem.corners # These are the corner coordinates
currentPosition = state[0]
runningPerimeter = 0
unvisitedCorners = set(corners) - state[1]
# the min function chooses the min of the first element in each tuple
# within it is a list comprehendsion of all the possible corner distances
while len(unvisitedCorners) != 0:
(dist, cornerVisited) = min([
(distance.manhattan(currentPosition, corner), corner)
for corner in unvisitedCorners
])
unvisitedCorners.remove(cornerVisited)
runningPerimeter += dist
currentPosition = cornerVisited
return runningPerimeter
def betterEvaluationFunction(currentGameState):
"""
Your extreme ghost-hunting, pellet-nabbing, food-gobbling, unstoppable evaluation function.
DESCRIPTION: <write something here so we know what you did>
Pacman should be penalized based on distance of closest food and number of food
Pacman should get every capsule it can
"""
# Useful information you can extract.
score = currentGameState.getScore()
pacPosition = currentGameState.getPacmanPosition()
foodList = currentGameState.getFood().asList()
n = 0
closestFoodDistance = 10000
for food in foodList:
distanceOfFood = distance.manhattan(pacPosition, food)
if closestFoodDistance > distanceOfFood:
closestFoodDistance = distanceOfFood
n = n + 1
if n != 0:
score = score + 1000 / (1 + float(n))
score = score - (closestFoodDistance * 2)
else:
score = score + 10000
capsuleList = currentGameState.getCapsules()
score = score - (10 * len(capsuleList))
if n == 0:
score = score + 10000
return score
def manhattan(position, problem):
"""
This heuristic is the manhattan distance to the goal.
"""
position1 = position
position2 = problem.goal
return distance.manhattan(position1, position2)
def betterEvaluationFunction(currentGameState):
"""
Your extreme ghost-hunting, pellet-nabbing, food-gobbling, unstoppable evaluation function.
DESCRIPTION: Calculate the sum of distances of foods and the distances of ghosts.
The threat is more important than food. Take the leftover penalty
into account. Minus them from the original score.
"""
position = currentGameState.getPacmanPosition()
foods = currentGameState.getFood().asList()
foodDist = 0
for food in foods:
foodDist += 2 * manhattan(position, food)
ghostDist = 0
for ghost in currentGameState.getGhostPositions():
ghostDist += 4 * manhattan(position, ghost)
penalty = -6 * len(food)
return currentGameState.getScore() - foodDist - ghostDist + penalty
def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current `pacai.bin.pacman.PacmanGameState`
and an action, and returns a number, where higher numbers are better.
Make sure to understand the range of different values before you combine them
in your evaluation function.
"""
successorGameState = currentGameState.generateSuccessor(
self.index, action)
# Useful information you can extract.
newPosition = successorGameState.getAgentPosition(self.index)
oldFood = self.getOppFood(currentGameState)
newGhostStates = self.getOppGhostState(successorGameState)
newGhostPositions = self.getOppGhost(successorGameState)
score = currentGameState.getScore() * 100 # weighted current score
closestDistToFood = float('-inf')
if self.onOppositeSide(newPosition, successorGameState):
score += 100
for food in oldFood.asList():
dist = distance.maze(newPosition, food, currentGameState)
if dist == 0: # check if the new position is food
score += 1000
else: # add reciprocol manhattan dist to food (b/c close food = more points) * weight
score += (1 / dist) * 100
# calculate ghost score
eat_ghost = 1 # coefficient to switch values if ghost are edible
for i in range(len(newGhostPositions)):
if newGhostStates[i].getScaredTimer(
) > 0: # if ghosts edible, change weight to inverse reciprocol
eat_ghost = -0.1
if distance.manhattan(newGhostPositions[i],
newPosition) <= 1: # if near ghost run away
score -= 50 * eat_ghost
else: # subtract reciprocol ghost distance (b/c closer ghost = less points) * weight
score -= (1 / distance.manhattan(newGhostPositions[i],
newPosition)) * 50 * eat_ghost
return score
def isOppClose(self, currentGameState):
myPos = currentGameState.getAgentPosition(self.index)
OppPositions = self.getOppGhost(currentGameState)
OppPacmanPos = self.getOppPacman(currentGameState)
if (len(OppPacmanPos) != 0):
if (self.onMySide(self.teammatePos, currentGameState)
and self.onMySide(myPos, currentGameState)):
for i in OppPositions:
if (distance.manhattan(myPos, i) <=
(distance.manhattan(i, self.teammatePos) + 30)):
return True
else:
return True
for i in OppPositions:
if self.isRed:
if self.mid_x_coord <= (i[0] - 20):
return True
else:
if self.mid_x_coord >= (i[0] + 20):
return True
return False
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 than what 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
`pacai.core.grid.Grid` 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
foodList = foodGrid.asList()
counter = 0
foodToVisit = []
x = 0
if len(foodList) == 0:
return 0
while len(foodList) != x:
foodToVisit.append(x)
x += 1
while len(foodToVisit) != 0:
foodDistances = []
for food in foodToVisit:
foodDistances.append(
(food, distance.manhattan(position, foodList[food])))
for iterator in range(0, len(foodDistances) - 1):
for a in range(0, len(foodDistances) - 1):
temp = foodDistances[a + 1]
if foodDistances[a][1] > temp[1]:
foodDistances[a + 1] = foodDistances[a]
foodDistances[a] = temp
counter += foodDistances[0][1]
position = foodList[foodDistances[0][0]]
foodToVisit.remove(foodDistances[0][0])
return counter - 4
def cornersHeuristic(state, problem):
"""
A heuristic for the CornersProblem that you defined.
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.
(You need not worry about consistency for this heuristic to receive full credit.)
"""
# Useful information.
# corners = problem.corners # These are the corner coordinates
# walls = problem.walls # These are the walls of the maze, as a Grid.
# position = state[0]
# cornersToVisit = []
# cTest = state[1]
# for x in range(0, 4):
# if cTest[x] is False:
# cornersToVisit.append(x)
# counter = 0
# for a in range (0, len(cornersToVisit)):
# counter += distance.manhattan(position, problem.corners[cornersToVisit[a]])
# position = problem.corners[cornersToVisit[a]]
# return counter
counter = 0
cornersToVisit = []
position = state[0]
cTest = state[1]
for x in range(0, 4):
if cTest[x] is False:
cornersToVisit.append(x)
while len(cornersToVisit) != 0:
cornerDistances = []
for corner in cornersToVisit:
cornerDistances.append(
(corner, distance.manhattan(position,
problem.corners[corner])))
for iterator in range(0, len(cornerDistances) - 1):
for a in range(0, len(cornerDistances) - 1):
temp = cornerDistances[a + 1]
if cornerDistances[a][1] >= temp[1]:
cornerDistances[a + 1] = cornerDistances[a]
cornerDistances[a] = temp
counter += cornerDistances[0][1]
position = problem.corners[cornerDistances[0][0]]
cornersToVisit.remove(cornerDistances[0][0])
return counter
def evaluationFunction(self, currentGameState, action):
"""
Design a better evaluation function here.
The evaluation function takes in the current `pacai.bin.pacman.PacmanGameState`
and an action, and returns a number, where higher numbers are better.
Make sure to understand the range of different values before you combine them
in your evaluation function.
"""
successorGameState = currentGameState.generatePacmanSuccessor(action)
# Useful information you can extract.
newPosition = successorGameState.getPacmanPosition()
oldFood = currentGameState.getFood()
newGhostStates = successorGameState.getGhostStates()
# newScaredTimes = [ghostState.getScaredTimer() for ghostState in newGhostStates]
score = 0
closestFoodDistance = -1
for food in oldFood.asList():
distanceOfFood = distance.manhattan(food, newPosition)
if closestFoodDistance == -1 or closestFoodDistance >= distanceOfFood:
closestFoodDistance = distanceOfFood
score += 10 / (1 + float(closestFoodDistance))
if currentGameState.getPacmanPosition() == newPosition:
score -= 5
if newPosition in currentGameState.getCapsules():
score += 100
for ghostState in newGhostStates:
distanceOfGhost = distance.manhattan(ghostState.getPosition(),
newPosition)
if distanceOfGhost <= 2:
score -= 100
return score + successorGameState.getScore()
def getDistribution(self, state):
# Read variables from state.
ghostState = state.getGhostState(self.index)
legalActions = state.getLegalActions(self.index)
pos = state.getGhostPosition(self.index)
isScared = ghostState.isScared()
speed = 1
if (isScared):
speed = 0.5
actionVectors = [
Actions.directionToVector(a, speed) for a in legalActions
]
newPositions = [(pos[0] + a[0], pos[1] + a[1]) for a in actionVectors]
pacmanPosition = state.getPacmanPosition()
# Select best actions given the state.
distancesToPacman = [
distance.manhattan(pos, pacmanPosition) for pos in newPositions
]
if (isScared):
bestScore = max(distancesToPacman)
bestProb = self.prob_scaredFlee
else:
bestScore = min(distancesToPacman)
bestProb = self.prob_attack
zipActions = zip(legalActions, distancesToPacman)
bestActions = [
action for action, distance in zipActions if distance == bestScore
]
# Construct distribution.
dist = counter.Counter()
for a in bestActions:
dist[a] = float(bestProb) / len(bestActions)
for a in legalActions:
dist[a] += float(1 - bestProb) / len(legalActions)
dist.normalize()
return dist
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 than what 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
`pacai.core.grid.Grid` 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 ***
i = 0
max = 0
d = 0
for x in foodGrid:
j = 0
for y in x:
if y:
d = distance.manhattan((i, j), position)
if d > max:
max = d
j += 1
i += 1
return max
def applyAction(state, action):
"""
Edits the state to reflect the results of the action.
"""
legal = PacmanRules.getLegalActions(state)
if (action not in legal):
raise ValueError('Illegal pacman action: ' + str(action))
pacmanState = state.getPacmanState()
# Update position.
vector = Actions.directionToVector(action, PacmanRules.PACMAN_SPEED)
pacmanState.updatePosition(vector)
# Eat.
nextPosition = pacmanState.getPosition()
nearest = nearestPoint(nextPosition)
if (manhattan(nearest, nextPosition) <= 0.5):
# Remove food
PacmanRules.consume(nearest, state)
def getDistance(self, pos1, pos2):
"""
The only function you will need after you create the object.
"""
if (self._distances is None):
return manhattan(pos1, pos2)
if isInt(pos1) and isInt(pos2):
return self.getDistanceOnGrid(pos1, pos2)
pos1Grids = getGrids2D(pos1)
pos2Grids = getGrids2D(pos2)
bestDistance = DEFAULT_DISTANCE
for pos1Snap, snap1Distance in pos1Grids:
for pos2Snap, snap2Distance in pos2Grids:
gridDistance = self.getDistanceOnGrid(pos1Snap, pos2Snap)
distance = gridDistance + snap1Distance + snap2Distance
if bestDistance > distance:
bestDistance = distance
return bestDistance
def cornersHeuristic(state, problem):
"""
A heuristic for the CornersProblem that you defined.
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.
(You need not worry about consistency for this heuristic to receive full credit.)
"""
# Useful information.
# corners = problem.corners # These are the corner coordinates
# walls = problem.walls # These are the walls of the maze, as a Grid.
# *** Your Code Here ***
# Search nodes expanded: 928 #
counter = 0
position = state[0]
cTest = state[1]
cornersToVisit = []
for x in range(0, 4):
if cTest[counter] is False:
cornersToVisit.append(x)
distances = [0, 0, 0, 0]
for a in range(0, len(cornersToVisit)):
for corner in cornersToVisit:
distances[corner] = distance.manhattan(position, problem.corners[corner])
min = -1
for corner in cornersToVisit:
if min == -1 or distances[min] > distances[corner]:
min = corner
cornersToVisit.remove(min)
counter += distances[min]
return counter
def chooseAction(self, state):
"""
alpha-beta pruning algorithm
(esentially minimax with additional alpha, beta conditions)
"""
ghostDists = [
distance.manhattan(self.getPosition(state), ghost)
for ghost in self.getOppGhost(state)
]
if len(ghostDists) > 0 and min(ghostDists) > 2:
legalMoves = state.getLegalActions(self.index)
if Directions.STOP in legalMoves:
legalMoves.remove(Directions.STOP)
# Choose one of the best actions.
scores = [
self.offenseEvaluationFunction(state, action)
for action in legalMoves
]
bestScore = max(scores)
bestIndices = [
index for index in range(len(scores))
if scores[index] == bestScore
]
chosenIndex = random.choice(
bestIndices) # Pick randomly among the best.
return legalMoves[chosenIndex]
print("alpha beta")
# passing in alpha = -inf and beta = inf
value, move = self.maxValue(state, 4, float('-inf'), float('inf'))
# print("move: ", move + " value: ", value)
return move
def chooseAction(self, state):
"""
alpha-beta pruning algorithm
(esentially minimax with additional alpha, beta conditions)
"""
ghostDists = [
distance.manhattan(self.getPosition(state), ghost)
for ghost in self.getOppGhost(state)
]
if len(ghostDists) > 0 and min(ghostDists) > 5:
legalMoves = state.getLegalActions(self.index)
if Directions.STOP in legalMoves:
legalMoves.remove(Directions.STOP)
# Choose one of the best actions.
scores = [
self.offenseEvaluationFunction(state, action)
for action in legalMoves
]
bestScore = max(scores)
bestIndices = [
index for index in range(len(scores))
if scores[index] == bestScore
]
chosenIndex = random.choice(
bestIndices) # Pick randomly among the best.
return legalMoves[chosenIndex]
# print("alpha beta")
# maxDepth = 2
# numAgents = 4
# evalFunc = self.ABEvaluationFunction
#
# def alphabeta(state, depth, agentIndex, alpha, beta):
# legalMoves = state.getLegalActions(agentIndex)
# if Directions.STOP in legalMoves:
# legalMoves.remove(Directions.STOP)
# if depth == maxDepth or len(legalMoves) == 0:
# return evalFunc(state), Directions.STOP
# print(legalMoves)
# v2, move = (None, None)
# nextStates = [state.generateSuccessor(agentIndex, move) for move in legalMoves]
# if agentIndex == 0:
# v = -99999
# for nextState, nextAction in zip(nextStates, legalMoves):
# v2, _ = alphabeta(nextState, depth, 1, alpha, beta)
# if v2 > v:
# v, move = v2, nextAction
# alpha = max(alpha, v)
# if v >= beta:
# return v, move
# return v, move
# elif agentIndex == numAgents - 1:
# v = 99999
# for nextState in nextStates:
# v2, action = alphabeta(nextState, depth + 1, 0, alpha, beta)
# if v2 < v:
# v, move = v2, action
# beta = min(beta, v)
# if v <= alpha:
# return v, move
# return v, move
# else:
# v = 99999
# for nextState in nextStates:
# v2, action = alphabeta(nextState, depth, agentIndex + 1, alpha, beta)
# if v2 < v:
# v, move = v2, action
# beta = min(beta, v)
# if v <= alpha:
# return v, move
# return v, move
#
# alphabetaVal = alphabeta(state, 0, 0, -99999, 99999)
# # print(alphabetaVal)
# print(alphabetaVal)
# return alphabetaVal[1]
# passing in alpha = -inf and beta = inf
value, move = self.maxValue(state, 2, float('-inf'), float('inf'))
# print("move: ", move + " value: ", value)
return move
def betterEvaluationFunction(currentGameState):
"""
Your extreme ghost-hunting, pellet-nabbing, food-gobbling, unstoppable evaluation function.
DESCRIPTION: <write something here so we know what you did>
"""
newPosition = currentGameState.getPacmanPosition()
oldFood = currentGameState.getFood()
newGhostStates = currentGameState.getGhostStates()
newScaredTimes = [ghostState.getScaredTimer() for ghostState in newGhostStates]
# *** Your Code Here ***
x = oldFood.getWidth()
y = oldFood.getHeight()
maxlen = distance.manhattan((x, y), (0, 0))
halflen = maxlen
minfooddist = maxlen
scareddist = maxlen
baddist = maxlen
scaredflag = True
badflag = True
for i in range(x):
for j in range(y):
m = oldFood[i][j]
if m:
d = distance.manhattan(newPosition, (i, j))
if d < minfooddist:
minfooddist = d
for i in range(len(newScaredTimes)):
ghost = newGhostStates[i]
d = distance.manhattan(newPosition, ghost._position)
if newScaredTimes[i] > 0:
scaredflag = False
if d < scareddist:
scareddist = d
else:
badflag = False
if d < baddist:
baddist = d
if baddist == 0:
return -100
if scaredflag:
scareddist = 1
if badflag:
baddist = 1
elif baddist > halflen:
baddist = 1
ev = (1 / scareddist) + (-1 / baddist) + (currentGameState.getScore() / 2)
if minfooddist == 0:
ev += 10
else:
ev += (2 / minfooddist)
if scaredflag:
ev -= 1
if badflag:
ev += 1
elif baddist > halflen:
ev += 1
return ev
def getFurthestCorner(self, pacPos):
poses = [(1, 1), (1, self.height - 2), (self.width - 2, 1),
(self.width - 2, self.height - 2)]
dist, pos = max([(manhattan(p, pacPos), p) for p in poses])
return pos
def canKill(pacmanPosition, ghostPosition):
return manhattan(ghostPosition, pacmanPosition) <= COLLISION_TOLERANCE
def getOtherRespawned(self, gameState):
newTeammatePos = gameState.getAgentPosition(
self.getTeammateIndex(gameState))
return distance.manhattan(newTeammatePos, self.teammatePos) > 1