def look_best_future(board, pf, times=0):
if times == 0:
max_prize_1 = []
possb = [game.up_possible(board),game.down_possible(board),game.left_possible(board),game.right_possible(board)]
for i in range(4):
if possb[i] == 1:
y = copy.deepcopy(board)
y = action(y,i)
max_prize_1.append(get_prize(y, pf))
else:
max_prize_1.append(0)
return max_prize_1
else:
max_prize = 0
move = -1
possb = [game.up_possible(board),game.down_possible(board),game.left_possible(board),game.right_possible(board)]
for i in range(4):
if possb[i] == 1:
y = copy.deepcopy(board)
y = action(y,i)
y = populate_all_possible(y)
max_prize_2 = 0
for j in y:
prize = look_best_future(j, pf, times-1)
prize = max(prize)
if prize > max_prize_2:
max_prize_2 = prize
if max_prize_2 >= max_prize:
max_prize = max_prize_2
move = i
return move
def eval_genomes(genomes, config):
for genome_id, genome in genomes:
net = neat.nn.recurrent.RecurrentNetwork.create(genome, config)
x = game.start_game()
possb = [game.up_possible(x),game.down_possible(x),game.left_possible(x),game.right_possible(x)]
y = [0 for i in range(16)]
y = game.matrix_to_vector(x)
death = 1
death_flag = 0
while possb != [0,0,0,0]:
output = net.activate(format_input(x))
x = action(output, x)
if game.matrix_to_vector(x) == y:
if death_flag == 0:
death = -3
death_flag = 1
else:
death += 1
else:
# print("not equal! \n")
# print(game.matrix_to_vector(x), y)
death = 1
death_flag = 0
y = game.matrix_to_vector(x)
if death == 0:
break
scr = game.score(x)
genome.fitness = scr
possb = [game.up_possible(x),game.down_possible(x),game.left_possible(x),game.right_possible(x)]
def eval_fitness(genomes, config):
for idx, g in genomes:
cppn = neat.nn.FeedForwardNetwork.create(g, config)
network = ESNetwork(substrate, cppn, params)
net = network.create_phenotype_network()
fitnesses = []
for i in range(trials):
x = game.start_game()
possb = [
game.up_possible(x),
game.down_possible(x),
game.left_possible(x),
game.right_possible(x)
]
scr = 0
y = [0 for z in range(16)]
y = game.matrix_to_vector(x)
death = 1
death_flag = 0
net.reset()
for j in range(max_steps):
for k in range(network.activations):
o = net.activate(format_input(x))
x = action(np.argmax(o), x)
if game.matrix_to_vector(x) == y:
if death_flag == 0:
death = -3
death_flag = 1
else:
death += 1
else:
death = 1
death_flag = 0
y = game.matrix_to_vector(x)
if death == 0:
break
scr = game.score(x)
possb = [
game.up_possible(x),
game.down_possible(x),
game.left_possible(x),
game.right_possible(x)
]
if possb == [0, 0, 0, 0]:
break
fitnesses.append(scr)
g.fitness = np.array(fitnesses).mean()
def run(pf):
x = game.start_game()
prev_x = copy.deepcopy(x)
possb = [game.up_possible(x),game.down_possible(x),game.left_possible(x),game.right_possible(x)]
while possb != [0,0,0,0]:
temp = copy.deepcopy(x)
move = look_avg_future(temp, pf, 0)
move = move.index(max(move))
if(possb[move]==1):
x = action(x,move)
x = game.populate(x)
# game.print_board(x)
# print("maxtile: ",game.maxtile(x))
if prev_x == x:
print("breaking ###################################################################################################################################################")
break
else:
prev_x = x
# game.print_board(x)
# print(possb)
possb = [game.up_possible(x),game.down_possible(x),game.left_possible(x),game.right_possible(x)]
print("maxtile: ", game.maxtile(x))
return game.score(x)
def look_avg_future(board, pf, times=0):
if times == 0:
max_prize_1 = []
possb = [game.up_possible(board),game.down_possible(board),game.left_possible(board),game.right_possible(board)]
for i in range(4):
if possb[i] == 1:
y = copy.deepcopy(board)
y = action(y,i)
max_prize_1.append(get_prize(y, pf))
else:
max_prize_1.append(-100000000)
return max_prize_1
else:
max_prize = 0
move = -1
possb = [game.up_possible(board),game.down_possible(board),game.left_possible(board),game.right_possible(board)]
for i in range(4):
if possb[i] == 1:
y = copy.deepcopy(board)
y = action(y,i)
y2 = populate_all_possible2(y)
y4 = populate_all_possible4(y)
prizes = []
for j in y2:
prize = look_avg_future(j, pf, times-1)
prize = max(prize)
prizes.append(prize)
for j in y4:
prize = look_avg_future(j, pf, times-1)
prize = max(prize)/10
prizes.append(prize)
avg_prize = sum(prizes)/len(prizes)
if avg_prize >= max_prize:
max_prize = avg_prize
move = i
return move