def ModifiedMinMax(self, board, limit, player, checker = 0):
global depth
depth = depth + 1
'''
a recursive min-max function
the function changes min/max utility based on the player passed
the algorithm is modified such that the selection is made based on maximizing one score and minimizing a different score at each step
'''
#if the board is half empty the game is over
if board.isHalfEmpty():
return 0
max_score = float("-inf")
max_move = 0
moves = board.generateOptions(player)
isCloneMove = True
#check if the game state is in the level just above the limit in the min max tree
#only check for maximum profit of the current player and return the best move
if ((depth == (limit - 1)) or ((depth == limit) and (limit == 1))):
for move_option in moves:
new_board = board.clone()
new_board_score = new_board.move(move_option, isCloneMove)
if new_board_score > max_score:
max_score = new_board_score
max_move = move_option
#on any depth other than limit - 1 recursively call the function and get the best score of the opponent in the resulting stage of the current move
#calculate the ratio of the addition to current player score by the move, to best potential addition to the opponent score of the resulting game state by the move
#choose the move with maximum ratio
else:
max_ratio = float("-inf")
for move_option in moves:
new_board = board.clone()
new_player = (player + 1) % 2
score = new_board.move(move_option, isCloneMove)
opponent_score = self.ModifiedMinMax(new_board, limit, new_player, depth)
if opponent_score != 0:
if float(score/opponent_score) > max_ratio:
max_score = score
max_ratio = float(score/opponent_score)
max_move = move_option
else:
max_score = score
max_move = move_option
if checker == 1:
return max_move
else:
return max_score
def pickMaxMin(self, board, marker):
moves = board.getMoves()
pairs = []
for (x, y) in moves:
temp = board.clone()
temp.move(x, y, marker)
if temp.finished():
score = self.evalBoard(temp)
else:
pick = self.pickMaxMin(temp, self.flip(marker))
score = pick[1]
pairs.append(((x, y), score))
pairs.sort(key=lambda x: x[1])
if marker == self.marker:
index = -1
else:
index = 0
return pairs[index]
def pickMaxMin(self, board, marker):
moves = board.getMoves()
pairs = []
for (x,y) in moves:
temp = board.clone()
temp.move(x,y,marker)
if temp.finished():
score = self.evalBoard(temp)
else:
pick = self.pickMaxMin(temp, self.flip(marker))
score = pick[1]
pairs.append(((x,y),score))
pairs.sort(key=lambda x:x[1])
if marker == self.marker:
index = -1
else:
index = 0
return pairs[index]
def greedyPlay(self):
'''
plays the game
heuristics: look ahead for one step and select the move with maximum score
'''
#choosing a pit
max_score = float("-inf")
max_move = 0
moves = board.generateOptions(player)
for move_option in moves:
new_board = board.clone()
new_board_score = new_board.move(move_option)
if new_board_score > max_score:
max_score = new_board_score
max_move = move_option
pit = max_move
#make moves and update score
score = self.game_board.move(pit)
self.updateScore(score)
self.switchPlayer()
def mainaction(board):
score = []
movescore = [1]
allmoves = moves()
for lists in allmoves:
sandbox = board.clone()
oldscore = sandbox.score
hole = 0
height = 24
allheights, diff_in_height, var, num_of_blocks, allholes = [], [], [], [], []
for move in lists:
if sandbox.falling is None: continue
elif isinstance(move, Rotation):
sandbox.rotate(move)
elif isinstance(move, Direction):
sandbox.move(move)
else:
continue
holes = 0
for x in range(sandbox.width):
min_y_seen = 24
temp = 0
for y in range(sandbox.height):
if (x, y) in sandbox:
if y <= min_y_seen:
# OVERALL HEIGHT
min_y_seen = y
allheights.append((height - min_y_seen))
# HEIGHT DIFFERENCES
ma = max(y for (x, y) in sandbox.cells)
mi = min(y for (x, y) in sandbox.cells)
differences = (ma - mi)
diff_in_height.append(differences)
mean = (height - min_y_seen) / 10
difsq = (y - mean) * (y - mean)
var.append(difsq)
num_of_blocks.append(len(sandbox.cells))
if (x, y) in sandbox.cells:
temp = temp + 1
if (x, y) not in sandbox.cells and temp > 0:
holes = holes + 1
allholes.append(holes)
# # HOLES IN THE BOARD
# if y >= min_y_seen:
# if (x, y) in sandbox.cells and (x, y + 1) not in sandbox.cells and (x, y + 2) in sandbox.cells:
# hole = hole + 10
# allholes.append(hole)
# else:
# allholes.append(0)
# VARIANCE
act_var = sum(var) / 10
# STANDARD DIVIATION
sd = []
sd.append(statistics.stdev(allheights))
# BUMPINESS
bump = []
for i in range(len(allheights) - 1):
l = abs(allheights[i + 1] - allheights[i])
bump.append(l)
# GAME
if sandbox.score - oldscore > 99:
movescore.append(10)
elif sandbox.score - oldscore > 199:
movescore.append(5)
else:
movescore.append(0)
# movescore = [float(i)/max(movescore) for i in movescore]
sumofscores = sum(movescore)
sumofheights = sum(allheights)
sumofdifferences = sum(diff_in_height)
sumofbump = sum(bump)
sumofsd = sum(sd)
sumofvar = sum(var) / 1000
sumofblocks = sum(num_of_blocks) / 10
sumofholes = sum(allholes)
ok = [-0.5, 0.25, 0.29, 0.05, 0.35, 0, 0, 0.35]
# Normalise the values of individual weights: sums up to 1
s = [
float(i) / sum(ok) for i in ok
] # from https://stackoverflow.com/questions/26785354/normalizing-a-list-of-numbers-in-python
heuristics = (s[0] * sumofheights, s[1] * sumofdifferences,
s[2] * sumofscores, s[3] * sumofbump, s[4] * sumofsd,
s[5] * sumofvar, s[6] * sumofblocks, s[7] * sumofholes)
sum_of_heuro = sum(heuristics)
# 2, 16, 1, 8 tuple move - score: 4440
# 2, 0, 0, 8 tuples move - score: 5376
# 10, 1, 1, 8 tuple move - score: 6874
# 10, 1, 1, 10 tuple move - score: 8805
# 10, 1, 10, 10 tuple move - score: 9699
# 10, 1, 10, 1, 10 tuple move score: 10404 (with four heuristics)
# 10, 1, 10, 1, 1, 0, 0, 0, 10 tuple move score: 17632 (left, right, left, right, clock, anti-clock)
# 1.5, 0.5, 0.5, 0.1, 0, 0, 0,
score.append(sum_of_heuro)
lowestscoore = score.index(min(score))
return allmoves[lowestscoore]
def set_board(self, board):
self.board = board.clone()
self.preview.board_drawer.set_board(self.board)
def set_board(self, board):
self.board = board.clone()
self.preview.board_drawer.set_board(self.board)
def GetAlphaBetaScore(self, board, limit, player, alpha, beta, checker=0):
# The function changes min/max utility with alpha beta prunning based on the player passed
# Getting class depth value
global depth
depth = depth + 1
# If the board is half empty the game is over
if board.isEmptyBoard():
return 0
# Gets current choices for this agent
max_score = float("-inf")
max_move = 0
moves = board.getCurrentChoices(player)
isCloneMove = True
# The algorithm take moves such that the selection is made based on maximizing one score and minimizing a different score at each step
# Also it prunes the some of the tree to avoid bigger space requirement and to decrease time complexity
# First of all check if the game state is in the level just above the limit in the min max tree
# Then Only check for maximum profit of the current player and return the best move
if ((depth == (limit - 1)) or ((depth == limit) and (limit == 1))):
for move_option in moves:
new_board = board.clone()
new_board_score = new_board.move(move_option, isCloneMove,
self.current_score)
if new_board_score > max_score:
max_score = new_board_score
max_move = move_option
# On any depth other than limit - 1 recursively call the function and get the best score of the opponent in the resulting stage of the current move
# Calculate the ratio of the addition to current player score by the move, to best potential addition to the opponent score of the resulting game state by the move
# Choose the move with maximum ratio
else:
max_ratio = float("-inf")
# Iterate over all of the choices
for move_option in moves:
new_board = board.clone()
new_player = (player + 1) % 2
score = new_board.move(move_option, isCloneMove,
self.current_score)
# if new score is smaller than alpha value then prune rest of tree branch
if (score <= alpha):
return score
# if new score is greater than beta value then prune rest of tree branch
if (score >= beta):
return score
# This is the recursion call for another player
# This time will be minimizing the score for oppenents
opponent_score = self.GetAlphaBetaScore(
new_board, limit, new_player, alpha + score, beta + score)
if opponent_score != 0:
if float(score / opponent_score) > max_ratio:
max_score = score
max_ratio = float(score / opponent_score)
max_move = move_option
else:
max_score = score
max_move = move_option
if checker == 1:
return max_move
else:
return max_score
