def getAction(self, state):
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
#SELECT TARGET
target = api.food(state)[0]
print target
pacman = api.whereAmI(state)
print "Pacman position: ", pacman
if self.backsteps == 0:
if pacman[0] >= target[0]:
if Directions.WEST in legal:
return api.makeMove(Directions.WEST, legal)
else:
if Directions.EAST in legal:
return api.makeMove(Directions.EAST, legal)
if pacman[1] >= target[1]:
if Directions.SOUTH in legal:
return api.makeMove(Directions.SOUTH, legal)
else:
if Directions.NORTH in legal:
return api.makeMove(Directions.NORTH, legal)
self.backsteps = 2 #IT REACHES HERE ONLY ONCE BOTH DIRECTIONS IT WANTS TO GO ARE ILLEGAL, SO: BACKSTEP 2 STOPS TOWARDS RANDOM LEGAL DIRECTION
self.backstep_direction = random.choice(legal)
self.backsteps -= 1
return api.makeMove(self.backstep_direction, legal)
def getAction(self, state):
legal = api.legalActions(state)
if not Directions.WEST in legal:
if Directions.EAST in legal:
legal.remove(Directions.EAST)
return api.makeMove(random.choice(legal), legal)
return api.makeMove(Directions.WEST, legal)
def move_pacman(self, pac_dir, axis, legal, pacman, target):
if pacman[axis] > target[axis]:
if pac_dir in legal:
return api.makeMove(pac_dir, legal)
elif pacman[axis] < target[axis]:
if pac_dir in legal:
return api.makeMove(pac_dir.REVERSE, legal)
def getToCornersOrFood():
# if all corners are visited then visit the food which was left beforehand
if (len(self.visitedCorners) > 4):
value = 1000000
for food in self.allFood:
distance = util.manhattanDistance(pacman, food)
if (distance < value):
self.corner = food
#find the path toward the corner or food
path = breadth_first_search(pacman, self.corner, walls)
if (pacman == self.corner):
self.visitedCorners.append(self.corner)
self.corner = getNearestCorner(pacman, corners)
path = breadth_first_search(pacman, self.corner, walls)
if path:
return path.pop(0)
else:
pick = random.choice(legal)
return api.makeMove(pick, legal)
else:
if path:
return path.pop(0)
else:
pick = random.choice(legal)
return api.makeMove(pick, legal)
def getAction(self, state):
value_list = []
legal = api.legalActions(state)
corner = api.corners(state)
pacman = api.whereAmI(state)
pacman_x = pacman[0]
pacman_y = pacman[1]
legal_width = corner[3][0]
legal_height = corner[3][1]
# policy iteration and evaluation
# get the value of four directions and select the direction corresponding to
# the maximum value as Pacman's decision.
map_effect = gridworld().map_valuegeneration(state, (legal_width, legal_height))
value_list.append(map_effect[(pacman_x-1, pacman_y)])
value_list.append(map_effect[(pacman_x+1, pacman_y)])
value_list.append(map_effect[(pacman_x, pacman_y + 1)])
value_list.append(map_effect[(pacman_x, pacman_y - 1)])
max_value = value_list.index(max(value_list))
# print 'map_effect'
# print map_effect
# print 'value_list'
# print value_list
# print 'max_value'
# print max_value
if max_value == 0:
return api.makeMove(Directions.WEST, legal)
if max_value == 1:
return api.makeMove(Directions.EAST, legal)
if max_value == 2:
return api.makeMove(Directions.NORTH, legal)
if max_value == 3:
return api.makeMove(Directions.SOUTH, legal)
def getAction(self, state):
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
target = (1, 1)
print target
print "Food locations: "
print len(api.food(state))
pacman = api.whereAmI(state)
print "Pacman position: ", pacman
if self.backsteps == 0:
if pacman[0] >= target[0]:
if Directions.WEST in legal:
return api.makeMove(Directions.WEST, legal)
else:
if Directions.EAST in legal:
return api.makeMove(Directions.EAST, legal)
if pacman[1] >= target[1]:
if Directions.SOUTH in legal:
return api.makeMove(Directions.SOUTH, legal)
else:
if Directions.NORTH in legal:
return api.makeMove(Directions.NORTH, legal)
self.backsteps = 2
self.backstep_direction = random.choice(legal)
self.backsteps -= 1
return api.makeMove(self.backstep_direction, legal)
def getAction(self, state):
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.WEST in legal:
return api.makeMove(Directions.WEST, legal)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# Random choice between the legal options.
return api.makeMove(random.choice(legal), legal)
def getAction(self, state):
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# Always go West if it is an option
if Directions.WEST in legal:
return api.makeMove(Directions.WEST, legal)
# Otherwise pick any other legal action
return api.makeMove(random.choice(legal), legal)
def getAction(self, state):
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# Get current location of pacman
pacman = api.whereAmI(state)
# Get list of food locations
food = api.food(state)
# Compute manhattan distance to each food location
foodDistances = []
for i in range(len(food)):
foodDistances.append(util.manhattanDistance(pacman, food[i]))
# print foodDistances
minDistance = min(foodDistances)
# print "Min Distance: ", minDistance
minDistanceIndex = foodDistances.index(minDistance)
# print "Min Food index: ", minDistanceIndex
nearestFood = food[minDistanceIndex]
# print "Pacman: ", pacman
# print "Nearest Food: ", nearestFood
diffX = pacman[0] - nearestFood[0]
diffY = pacman[1] - nearestFood[1]
# print "legal", legal
# print diffX, diffY
moveX = Directions.STOP
moveY = Directions.STOP
# Determine whether to move east or west
if diffX >= 0:
# print "Go left"
moveX = Directions.WEST
elif diffX < 0:
# print "Go right"
moveX = Directions.EAST
# Determine whether to move north or south
if diffY >= 0:
# print "Go down"
moveY = Directions.SOUTH
elif diffY < 0:
# print "Go up"
moveY = Directions.NORTH
# Determine whether to move in X or Y
# print "diffX: ", diffX, " diffY", diffY
if abs(diffX) >= abs(diffY) and moveX in legal:
# print moveX
return api.makeMove(moveX, legal)
elif abs(diffY) >= 0 and moveY in legal:
# print moveY
return api.makeMove(moveY, legal)
elif abs(diffX) >= 0 and moveX in legal:
return api.makeMove(moveX, legal)
else:
# print "Random"
return api.makeMove(random.choice(legal), legal)
def getAction(self,state):
legal = state.getLegalPacmanActions() #get a list of pacman's legal actions
if Directions.STOP in legal:
legal.remove(Directions.STOP) #make it so that pacman won't stop even though it is legal to
#if going west is in the list of legal moves, then go west
#else pick a random direction to move
if Directions.WEST in legal:
return api.makeMove('West', legal)
else:
pick = random.choice(legal)
return api.makeMove(pick, legal)
def findPath(self, state):
#print "findPath"
#initialize the queue for a Depth First Seach
bfsQueue = self.getBFSQueue(state)
#copy of the current state of internal map to mark searched nodes
copyMap = deepcopy(self.map)
#conducts bfs search
while len(bfsQueue) != 0:
#pop the element from the queue
nextCheck = bfsQueue.pop(0)
#get the position stored at element
nextCheckPosition = nextCheck[0]
#extract x position from position
possibleX = nextCheckPosition[0]
#extract y position from position
possibleY = nextCheckPosition[1]
#if the position contains food, a capsule, or is unknown, pacman will visit it
if self.map[possibleX][possibleY] == "F" or self.map[possibleX][possibleY] == "?" or self.map[possibleX][possibleY] == "C":
#print "next move"
#pop the first step of the path stored in the currently searching element
nextMove = nextCheck[1].pop(0)
#set the internal path of pacman to the one found by the BFS
self.path = nextCheck[1]
#mark the position on the map as "P" to mark visited positions
self.map[nextMove[0][0]][nextMove[0][1]] = "P"
#set lastDir as the next move
self.lastDir = nextMove[1]
#return the popped direction stored at the frist step above as the next move
return api.makeMove(self.lastDir, api.legalActions(state))
else:
#if position does not contain food, capsule, or unknown information, continue the BFS search
#print "searching bfs"
#check all possibleMoves from the current popped from the queue in the current iteration of the BFS algorithm
for move in self.possibleMoves:
#get the next position in the search
nextPosition = self.sumPair(move[0], nextCheckPosition)
#if this postion is neither a wall nor a location already searched by the algorithm, add it to the search
if self.map[nextPosition[0]][nextPosition[1]] != "W" and copyMap[nextPosition[0]][nextPosition[1]] != "X":
#mark this location as searched by the BFS algorithm
copyMap[nextPosition[0]][nextPosition[1]] = "X"
#copy the existing path from the search into a new variable
path = deepcopy(nextCheck[1])
#add in the new loation and the direction to move to said location to the path
path.append((nextPosition, move[1]))
#add the path to the bfsQueue
bfsQueue.append((nextPosition, path))
#if no moves are available, pacman will not move
self.lastDir = Directions.STOP
return api.makeMove(Directions.STOP, api.legalActions(state))
def getAction(self, state):
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# If we can repeat the last action, do it. Otherwise make a
# random choice.
if self.last in legal:
return api.makeMove(self.last, legal)
else:
pick = random.choice(legal)
# Since we changed action, record what we did
self.last = pick
return api.makeMove(pick, legal)
def getAction(self, state):
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# Get current location of pacman
pacman = api.whereAmI(state)
# Get list of ghost locations
ghosts = api.ghosts(state)
# Compute manhattan distance to each ghost location
ghostsDistances = []
for i in range(len(ghosts)):
ghostsDistances.append(util.manhattanDistance(pacman, ghosts[i]))
minDistance = min(ghostsDistances)
print "Min Distance: ", minDistance
minDistanceIndex = ghostsDistances.index(minDistance)
print "Min Ghosts index: ", minDistanceIndex
nearestGhost = ghosts[minDistanceIndex]
print "Pacman: ", pacman
print "Nearest Ghost: ", nearestGhost
diffX = pacman[0] - nearestGhost[0]
diffY = pacman[1] - nearestGhost[1]
print "legal", legal
print diffX, diffY
moveX = Directions.STOP
moveY = Directions.STOP
# Determine whether to move east or west
if diffX >= 0:
print "Go right"
moveX = Directions.EAST
elif diffX < 0:
print "Go left"
moveX = Directions.WEST
# Determine whether to move north or south
if diffY >= 0:
print "Go up"
moveY = Directions.NORTH
elif diffY < 0:
print "Go down"
moveY = Directions.SOUTH
# Determine whether to move in X or Y
print "diffX: ", diffX, " diffY", diffY
if abs(diffX) >= abs(diffY) and moveX in legal:
return api.makeMove(moveX, legal)
elif abs(diffY) >= 0 and moveY in legal:
return api.makeMove(moveY, legal)
elif abs(diffX) >= 0 and moveX in legal:
return api.makeMove(moveX, legal)
else:
return api.makeMove(random.choice(legal), legal)
def getAction(self, state):
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = state.getLegalPacmanActions()
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# Get the current score
current_score = state.getScore()
# Get the last action
last = state.getPacmanState().configuration.direction
# If we can repeat the last action, do it. Otherwise make a
# random choice.
if last in legal:
return api.makeMove(last, legal)
else:
pick = random.choice(legal)
return api.makeMove(pick, legal)
def getAction(self, state):
#current possible moves
legal = state.getLegalPacmanActions()
#get current position
pacman = api.whereAmI(state)
pacmanX = pacman[0]
pacmanY = pacman[1]
#get food locations
food = api.food(state)
foodLoc = []
# get Distance
for loc in food:
foodLoc.append((abs(loc[0]-pacmanX + loc[1]-pacmanY),(loc[0]-pacmanX, loc[1]-pacmanY)))
print foodLoc
#Prevent it from stopping
if Directions.STOP in legal:
legal.remove(Directions.STOP)
pick = random.choice(legal)
return api.makeMove(pick, legal)
def getAction(self, state):
#current possible moves
legal = state.getLegalPacmanActions()
#Prevent it from stopping
if Directions.STOP in legal:
legal.remove(Directions.STOP)
#if it can go west, go west
if Directions.WEST in legal:
return api.makeMove(Directions.WEST, legal)
#else pick a random direction
else:
pick = random.choice(legal)
return api.makeMove(pick, legal)
def getAction(self, state):
legal = api.legalActions(state)
corners = api.corners(state)
print(corners)
return api.makeMove(Directions.STOP, legal)
def getAction(self, state):
# Get legal actions
legal = api.legalActions(state)
# Get location of Pacman
pacman = api.whereAmI(state)
# Get location of Ghosts
locGhosts = api.ghosts(state)
#print "locGhosts: ", locGhosts
# Get distance between pacman and the ghosts
for i in locGhosts:
p_g_dist = util.manhattanDistance(pacman, i)
# Get distance between ghosts
g_g_dist = util.manhattanDistance(locGhosts[0], locGhosts[1])
#print "g_g_dist:", g_g_dist
# Get distance between pacman and first Ghost
dist = []
dist.append(locGhosts[0][0] - pacman[0])
dist.append(locGhosts[0][1] - pacman[1])
return api.makeMove(Directions.STOP, legal)
def getMove(self, state):
scores = [0, 0, 0, 0]
#searches all adjacent squares
for i in range(len(self.possibleMoves)):
direction = self.possibleMoves[i][1]
#see if surrounding locations are legal moves. if so, add it to the search
if direction in self.legal:
deltaPosition = self.possibleMoves[i][0]
nextPosition = self.sumPair(self.pacman, deltaPosition)
positionScore = self.map[nextPosition[0]][nextPosition[1]]
scores[i] = positionScore
else:
scores[i] = -1
cumulativeScore = [0, 0, 0, 0]
for i in range(len(scores)):
if self.possibleMoves[i][1] in self.legal:
cumulativeScore[i] = .1 * scores[
(i + 4) % 4] + .8 * scores[i] + .1 * scores[(i + 1) % 4]
else:
cumulativeScore[i] = -100
max = -100
index = 0
for i in range(len(cumulativeScore)):
if cumulativeScore[i] > max:
index = i
max = cumulativeScore[i]
return api.makeMove(self.possibleMoves[index][1], self.legal)
def getMove(self,state):
#print "getmove"
#calculate the scores from all possible moves
scores = self.calcAdjUtility(self.pacman[0], self.pacman[1])
#get the maximum value from calculated utilities
maxUtility = max(scores)
#get the imdex of the move
index = scores.index(maxUtility)
#get the direction that orresponds to the highest utility
direction = self.possibleMoves[index][1]
#if staying still can provide higher utility, then stay still
if self.utility[self.pacman[0]][self.pacman[1]] >= maxUtility:
#print "dont move"
return api.makeMove(Directions.STOP, self.legal)
#otherwise move in the previously found direction
return api.makeMove(direction, self.legal)
def getAction(self, state):
start = api.whereAmI(state)
# generate rewards for map locations in each game cycle to reflect changing conditions
self.rewardMapping(state)
# create the dictionary of states at the start of the game
if not self.stateDict:
self.stateMapping(state)
# get converged utilities from value iteration
gridVals = self.valueIteration(state)
# get and assign the utility values and their associated locations to the Grid object, then print this map
# representation to the console
valKeys = gridVals.keys()
valVals = gridVals.values()
for i in range(len(gridVals)):
y = valKeys[i][0]
x = valKeys[i][1]
val = '{:3.2f}'.format(valVals[i])
self.map.setValue(y, x, val)
self.map.prettyDisplay()
# optimal policy for Pacman: get the location of the optimal utility from his current position
sPrime = self.bestMove(state, gridVals)
# required argument for makeMove()
legal = api.legalActions(state)
# return a move based on the direction returned by singleMove() and sPrime
return api.makeMove(self.singleMove(start, sPrime), legal)
def getAction(self, state):
#if the internal map of the environment has yet to be initialized, initialize it
if not self.init:
self.initialize(state)
#update the legal moves for this move
self.setLegal(state)
#if pacman can detect a ghost nearby pacman needs to run away
if api.ghosts(state):
return self.runAway(state)
#if a route has been found, pacman will follow it instead of searching again
if len(self.path) != 0:
#pop off the first move in the path
nextMove = self.path.pop(0)
#check that the move is legal
if nextMove[1] in self.legal:
#mark that position as visited with "P"
self.map[nextMove[0][0]][nextMove[0][1]] = "P"
self.lastDir = nextMove[1]
#return the move
return api.makeMove(self.lastDir, self.legal)
#if the move is not legal, find a new path
else:
return self.findPath(state)
#otherwise find a path
else:
return self.findPath(state)
def getAction(self, state):
""" This function is used to make pacman move.
Parameter:
state: the state of pacman.
Returns the move of pacman.
"""
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
# Get the size of layout
rows,columns = self.map_size(state)
# Judge the map_type is mediumClass or smallGrid
map_type = 'mediumClass'
if rows >= 10 and columns >= 10:
map_type = 'mediumClass'
else:
map_type = 'smallGrid'
# Update the value map once pacman moved
value_map = self.map_update(state)
# Update the value map with value_iteration
value_map = self.value_iteration(state, value_map, map_type)
# Make move
return api.makeMove(self.where_to_move(state,value_map), legal)
def getAction(self, state):
"""Get Action of pacman
Args:
state: The state of an agent (configuration, speed, scared, etc).
"""
# Get the actions we can try, and remove "STOP" if that is one of them.
legal = api.legalActions(state)
if Directions.STOP in legal:
legal.remove(Directions.STOP)
cols,rows = self.mapSize(state) # Get the layout size
# Judge the layout type medium or small
map_type = 'medium'
if cols >= 10 and rows >= 10:
map_type = 'medium'
else:
map_type = 'small'
# Update value map after pacman moved
updated_map = self.mapUpdate(state)
# Update value map with iteration
updated_map = self.Iteration(state,updated_map,map_type)
# Make move
return api.makeMove(self.whichToMove(state,updated_map), legal)
def getAction(self, state):
# How we access the features.
features = api.getFeatureVector(
state
) # first 4: walls, next 4: food, next 8: ghosts, next 1: ghost infront, next 1: class
# *****************************************************
#
# Here you should insert code to call the classifier to
# decide what to do based on features and use it to decide
# what action to take.
#
# *******************************************************
# from collected 'features' vector, classify as any of 0-3.
# this gives your move
# Get class from model and convert to action
number = self.model.predict(features)
action = self.convertNumberToMove(number)
# Get the actions we can try.
legal = api.legalActions(state)
return api.makeMove(action, legal)
def getAction(self, state):
self.makeMap(state)
self.addWallsToMap(state)
self.addConsumablesToMap(state)
self.updateGhosts(state)
# Get the actions we can try, and remove "STOP" if that is one of them.
pacman = api.whereAmI(state)
legal = api.legalActions(state)
ghosts = api.ghosts(state)
corners = api.corners(state)
layoutHeight = self.getLayoutHeight(corners)
layoutWidth = self.getLayoutWidth(corners)
if (layoutHeight-1)<8 and (layoutWidth-1)<8:
for i in range (100):
self.valIterS(state,0.68,-0.1)
else:
for i in range (50):
self.valIterM(state,0.8,-0.1)
plannedMove = self.plannedMove(pacman[0],pacman[1])
#self.dankMap.prettyDisplay()
#Feel free to uncomment this if you like to see the values generated
if Directions.STOP in legal:
legal.remove(Directions.STOP)
#Input the calculated move for our next move
return api.makeMove(plannedMove, legal)
def getAction(self, state): print "-" * 30 #divider ghosts = api.ghosts(state) #get state of ghosts legal = state.getLegalPacmanActions() #Again, get a list of pacman's legal actions last = state.getPacmanState().configuration.direction #store last move pacman = api.whereAmI(state) #retrieve location of pacman food = api.food(state) #retrieve location of food walls = api.walls(state) #how to call getfoodvalmap method. #In reality, the reward should be the final value-iteration of the grid. foodVal = self.getValueMap(state, 10) print foodVal #example on how to use getPacMEU function currentUtil = self.getPacMEU(pacman[0], pacman[1], foodVal, legal) print "Utility values: " print currentUtil print max(currentUtil.values()) #example on how to use getMEU function foodUtil = self.getMEU((18, 3), foodVal, walls) print "max utility for (18, 3) is: " print foodUtil if Directions.STOP in legal: legal.remove(Directions.STOP) # Random choice between the legal options. return api.makeMove(random.choice(legal), legal) """
def getAction(self, state):
self.updateMap(state)
self.policyIteration()
pacman = api.whereAmI(state)
legal = api.legalActions(state)
move = self.policy[pacman]
return api.makeMove(move, legal)
def getAction(self, state):
# How we access the features.
features = api.getFeatureVector(state)
# *****************************************************
#
# Here you should insert code to call the classifier to
# decide what to do based on features and use it to decide
# what action to take.
#
# *******************************************************
nn = self.nn
# use the trained neural network to predict
y_pred = nn.predict(features)
# transform the output value from a list of 0 and 1 to a single value
n = 0
for i in range(len(y_pred)):
if y_pred[i] > y_pred[n]:
n = i
# Get the actions we can try.
legal = api.legalActions(state)
# getAction has to return a move. Here we pass "STOP" to the
# API to ask Pacman to stay where they are. We need to pass
# the set of legal moves to teh API so it can do some safety
# checking.
return api.makeMove(self.convertNumberToMove(n), legal)
def getAction(self, state):
self.pacman = api.whereAmI(state)
self.legal = api.legalActions(state)
if not self.init:
self.initialize(state)
else:
# if self.reward[self.pacman[0]][self.pacman[1]] < 10:
# self.reward[self.pacman[0]][self.pacman[1]] = self.reward[self.pacman[0]][self.pacman[1]] - 1
# else:
self.reward[self.pacman[0]][self.pacman[1]] = -1
print "reward"
for row in self.reward:
print row
self.updateMap(state)
#print "\nreward"
#for row in self.reward:
# print row
self.bellman(state)
#print "utility"
#for row in self.utility:
# print row
return self.getMove(state)
return api.makeMove(Directions.STOP, self.legal)