def test_get_winner(self):
model = PatchworkModel()
self.assertEqual(model.p1_win(), 0)
model.p1.buttons = 30
self.assertEqual(model.p1_win(), 1)
model.p2.buttons = 60
self.assertEqual(model.p1_win(), -1)
model.p1.buttons = 60
self.assertEqual(model.p1_win(), 0)
class PatchworkControllerAIvAI():
INTERVAL_SIZE = 5
def __init__(self):
self.model = PatchworkModel()
self.ai = PatchworkAI()
self.feature_weights = FeatureWeights()
self.p1_feature_state = FeatureStateModel(self.model.p1, self.model.p2)
self.p2_feature_state = FeatureStateModel(self.model.p2, self.model.p1)
self.running = True
def mainloop(self, num_samples):
#run "num_samples" games and then calculate the win percentage
p1_win = 0
p1_running_sum = 0
p2_running_sum = 0
p1_win_count = 0
p1_game_avgs = []
p2_game_avgs = []
p1_win_counts = []
for i in range(num_samples):
#reset the model for next game
self.model = PatchworkModel()
self.running = True
while self.running:
if self.model.p1_turn():
player = self.model.p1
other_player = self.model.p2
else:
player = self.model.p2
other_player = self.model.p1
#p1 is smart, p2 is dumb
if self.model.p1_turn():
turn = self.ai.choose_turn_hand_craft(self.model)
else:
turn = self.ai.choose_turn_random(self.model)
if isinstance(turn, BuyTurn):
#TEMPORARY HANDLING FOR IF THE PIECE CANT BE PLACED, NEED NEW SOLUTION FOR THIS
if not self.ai.can_place(
self.model.patch_list[turn.patch_idx],
player.quilt):
turn = JumpTurn()
else:
row, col, patch_orientation = self.ai.choose_placement(
self.model.patch_list[turn.patch_idx],
player.quilt)
self.model.place_patch(player, patch_orientation, row,
col)
#run the turn (buy piece for buy, jump for jump), and check if patch is passed on time track
passed_patch, passed_button_gen = turn.run(self.model)
if passed_patch:
row, col, patch_orientation = self.ai.choose_placement(
Patch(34, [[1]], 0, 0, 0), other_player.quilt)
self.model.place_patch(other_player,
Patch(34, [[1]], 0, 0, 0), row, col)
#if game is over, exit running loop and update p1_win counter
if self.model.game_over():
if self.model.p1_win() == 1:
p1_win += 1
#keep track of p1/p2 end scores to calculate averages at end (hoping to see progress)
if ((i + 1) % self.INTERVAL_SIZE) == 0:
p1_game_avgs.append(p1_running_sum /
self.INTERVAL_SIZE)
p2_game_avgs.append(p2_running_sum /
self.INTERVAL_SIZE)
p1_running_sum = 0
p2_running_sum = 0
p1_win_count = 0
p1_running_sum += self.model.p1.get_score()
p2_running_sum += self.model.p2.get_score()
if self.model.p1_win() == 1:
p1_win_count += 1
self.running = False
#If there's only 1 sample just print out the basic output
if num_samples == 2 or num_samples == 1:
print("P1 score: " + str(self.model.p1.get_score()) +
", P2 score: " + str(self.model.p2.get_score()))
print("P1 buttons: " + str(self.model.p1.buttons) +
", P2 buttons: " + str(self.model.p2.buttons))
print("-------------------------------")
print("P1 board")
for row in range(9):
for col in range(9):
print(self.model.p1.quilt.board_array[row][col],
end="")
print()
print("-------------------------------")
print("P2 board")
for row in range(9):
for col in range(9):
print(self.model.p2.quilt.board_array[row][col],
end="")
print()
print("-------------------------------")
if i == 0:
print("Simulating", end="")
sys.stdout.flush()
elif (i % 4) == 0:
sys.stdout.write("\b\b\b \b\b\b")
sys.stdout.flush()
else:
sys.stdout.write(".")
sys.stdout.flush()
print("Games run: " + str(num_samples))
print("P1 wins: " + str(p1_win) + ", P2 wins: " +
str(num_samples - p1_win))
print("P1 win %: " + str((p1_win / num_samples) * 100) +
", P2 win %: " +
str(((num_samples - p1_win) / num_samples) * 100))
print()
for i in range(len(self.ai.feature_weights["patch_weights_early"])):
print("Patch ID: " + str(i + 1) + ", Early Weight: " +
str(self.ai.feature_weights["patch_weights_early"][i]) +
", Mid Weight: " +
str(self.ai.feature_weights["patch_weights_mid"][i]) +
", Late Weight: " +
str(self.ai.feature_weights["patch_weights_late"][i]))
print()
print("GAME AVGS: ")
for i in range(len(p1_game_avgs)):
print(
str(i * self.INTERVAL_SIZE) + " - " +
str((i * self.INTERVAL_SIZE) + self.INTERVAL_SIZE) +
" | Player 1: " + str(p1_game_avgs[i]) + ", Player 2: " +
str(p2_game_avgs[i]) + ", Difference: " +
str(p1_game_avgs[i] - p2_game_avgs[i]))
def mainloop_learning(self, num_samples):
#run "num_samples" games and then calculate the win percentage
p1_win = 0
p1_running_sum = 0
p2_running_sum = 0
p1_win_count = 0
p1_game_avgs = []
p2_game_avgs = []
p1_win_counts = []
p1_button_gen_sum = 0
p1_buttons_sum = 0
p1_board_util_sum = 0
p1_button_gen_sums = []
p1_buttons_sums = []
p1_board_util_sums = []
p1_quilts = []
p2_button_gen_sum = 0
p2_buttons_sum = 0
p2_board_util_sum = 0
p2_button_gen_sums = []
p2_buttons_sums = []
p2_board_util_sums = []
p1_jump_ct = 0
p2_jump_ct = 0
p1_buy_ct = 0
p2_buy_ct = 0
p1_delayed_jump_ct = 0
p2_delayed_jump_ct = 0
for i in range(num_samples):
#reset the model for next game
self.model = PatchworkModel()
self.running = True
self.p1_feature_state = FeatureStateModel(self.model.p1,
self.model.p2)
self.p2_feature_state = FeatureStateModel(self.model.p2,
self.model.p1)
while self.running:
if self.model.p1_turn():
player = self.model.p1
state_model = self.p1_feature_state
other_player = self.model.p2
else:
player = self.model.p2
state_model = self.p2_feature_state
other_player = self.model.p1
#p1 is learning, p2 is hand crafted ai
if self.model.p1_turn():
#picks turn based on learned weights and updates those weights
turn = self.ai.choose_turn_learning(
self.model, self.p1_feature_state,
self.feature_weights)
else:
turn = self.ai.choose_turn_random(self.model)
#update weights based on p2 moves as well
future_state = self.ai.find_future_state(
turn, self.p2_feature_state)
reward = self.p2_feature_state.calculate_reward(
turn.passes_player, turn.passes_econ, 0)
self.feature_weights.update_feature_weights(
reward,
future_state.get_state_utility(self.feature_weights),
self.p2_feature_state)
if isinstance(turn, JumpTurn):
#DEBUGGING
if self.model.p1_turn():
p1_jump_ct += 1
else:
p2_jump_ct += 1
#handling turn running
if isinstance(turn, BuyTurn):
#DEBUGGING
if self.model.p1_turn():
p1_buy_ct += 1
else:
p2_buy_ct += 1
#check if it's possible for the player to place the selected patch
if not self.ai.can_place(
self.model.patch_list[turn.patch_idx],
player.quilt):
passes_patch, passes_econ, passes_player = player.will_pass_tile(
other_player.position - player.position + 1,
self.model.time_track, other_player)
turn = JumpTurn(passes_patch, passes_econ,
passes_player)
if self.model.p1_turn():
p1_buy_ct -= 1
p1_delayed_jump_ct += 1
else:
p2_buy_ct -= 1
p2_delayed_jump_ct += 1
else:
row, col, patch_orientation = self.ai.choose_placement(
self.model.patch_list[turn.patch_idx],
player.quilt)
self.model.place_patch(player, patch_orientation, row,
col)
#run the turn (buy piece for buy, jump for jump), and update passed_patch/passed_button_gen booleans
passed_patch, passed_button_gen = turn.run(
self.model, state_model)
#placing 1x1 patch if it was passed
if passed_patch:
row, col, patch_orientation = self.ai.choose_placement(
Patch(34, [[1]], 0, 0, 0), other_player.quilt)
self.model.place_patch(other_player,
Patch(34, [[1]], 0, 0, 0), row, col)
#if game is over, exit running loop and update p1_win counter
if self.model.game_over():
p1_rewards = self.p1_feature_state.calculate_reward(
False, True, self.model.p1_win())
p2_rewards = self.p2_feature_state.calculate_reward(
False, True, self.model.p1_win())
self.feature_weights.update_feature_weights(
p1_rewards,
self.p1_feature_state.get_state_utility(
self.feature_weights), self.p1_feature_state)
self.feature_weights.update_feature_weights(
p2_rewards,
self.p2_feature_state.get_state_utility(
self.feature_weights), self.p2_feature_state)
p1_quilts.append(self.model.p1.quilt.board_array)
#CALCULATING OUTPUTS FOR PROGRESS TRACKING
#keep track of p1/p2 end scores to calculate averages at end (hoping to see progress)
if ((i + 1) % self.INTERVAL_SIZE) == 0:
p1_game_avgs.append(p1_running_sum /
self.INTERVAL_SIZE)
p2_game_avgs.append(p2_running_sum /
self.INTERVAL_SIZE)
p1_win_counts.append(p1_win_count)
p1_running_sum = 0
p2_running_sum = 0
p1_win_count = 0
p1_button_gen_sums.append(p1_button_gen_sum /
self.INTERVAL_SIZE)
p1_buttons_sums.append(p1_buttons_sum /
self.INTERVAL_SIZE)
p1_board_util_sums.append(p1_board_util_sum /
self.INTERVAL_SIZE)
p1_button_gen_sum = 0
p1_buttons_sum = 0
p1_board_util_sum = 0
p2_button_gen_sums.append(p2_button_gen_sum /
self.INTERVAL_SIZE)
p2_buttons_sums.append(p2_buttons_sum /
self.INTERVAL_SIZE)
p2_board_util_sums.append(p2_board_util_sum /
self.INTERVAL_SIZE)
p2_button_gen_sum = 0
p2_buttons_sum = 0
p2_board_util_sum = 0
if self.model.p1_win() == 1:
p1_win_count += 1
p1_win += 1
p1_running_sum += self.model.p1.get_score()
p2_running_sum += self.model.p2.get_score()
p1_button_gen_sum += self.model.p1.quilt.button_gen
p1_buttons_sum += self.model.p1.buttons
p1_board_util_sum += self.ai.get_quilt_utility(
self.model.p1.quilt)
p2_button_gen_sum += self.model.p2.quilt.button_gen
p2_buttons_sum += self.model.p2.buttons
p2_board_util_sum += self.ai.get_quilt_utility(
self.model.p2.quilt)
#exit while loop, game is over
self.running = False
#If there's only 1 sample just print out the basic output
if num_samples == 2 or num_samples == 1:
print("P1 score: " + str(self.model.p1.get_score()) +
", P2 score: " + str(self.model.p2.get_score()))
print("P1 buttons: " + str(self.model.p1.buttons) +
", P2 buttons: " + str(self.model.p2.buttons))
print("-------------------------------")
print("P1 board")
for row in range(9):
for col in range(9):
print(self.model.p1.quilt.board_array[row][col],
end="")
print()
print("-------------------------------")
print("P2 board")
for row in range(9):
for col in range(9):
print(self.model.p2.quilt.board_array[row][col],
end="")
print()
print("-------------------------------")
if i == 0:
print("Simulating", end="")
sys.stdout.flush()
elif (i % 4) == 0:
sys.stdout.write("\b\b\b \b\b\b")
sys.stdout.flush()
else:
sys.stdout.write(".")
sys.stdout.flush()
print()
print("Games run: " + str(num_samples))
print("P1 wins: " + str(p1_win) + ", P2 wins: " +
str(num_samples - p1_win))
print("P1 win %: " + str((p1_win / num_samples) * 100) +
", P2 win %: " +
str(((num_samples - p1_win) / num_samples) * 100))
print()
print("WEIGHTS: ")
print("Button Gen Weight: " +
str(self.feature_weights.button_gen_weight))
print("Board Coverage Weight: " +
str(self.feature_weights.board_coverage_weight))
print("Button Weight: " + str(self.feature_weights.buttons_weight))
print("Player Distance Weight: " +
str(self.feature_weights.player_distance_weight))
print()
#f1 = open("ai_button_gen.txt", "w")
#f2 = open("ai_button_total.txt", "w")
#f3 = open("ai_quilt_coverage.txt", "w")
#f4 = open("ai_score.txt", "w")
#f5 = open("ai_score_diff.txt", "w")
print("GAME AVGS: ")
for i in range(len(p1_game_avgs)):
print(
str(i * self.INTERVAL_SIZE) + " - " +
str((i * self.INTERVAL_SIZE) + self.INTERVAL_SIZE) +
" | Player 1: " + str(p1_game_avgs[i]) + ", Player 2: " +
str(p2_game_avgs[i]) + ", Difference: " +
str(p1_game_avgs[i] - p2_game_avgs[i]) + ", P1 Wins: " +
str(p1_win_counts[i]) + ", P2 Wins: " +
str(self.INTERVAL_SIZE - p1_win_counts[i]) + ", P1 Win %: " +
str((p1_win_counts[i] / self.INTERVAL_SIZE) * 100))
#f.write(str((p1_win_counts[i]/5)*100) + ", ")
print("AI TENDENCIES: ")
print("P1 (LEARNED AI) FEATURE AVERAGES")
for i in range(len(p1_button_gen_sums)):
print(
str(i * self.INTERVAL_SIZE) + " - " +
str((i * self.INTERVAL_SIZE) + self.INTERVAL_SIZE) +
" | Button Gen: " + str(p1_button_gen_sums[i]) +
", Buttons: " + str(p1_buttons_sums[i]) + ", Board Util: " +
str(p1_board_util_sums[i]))
#f1.write(str(p1_button_gen_sums[i]) + ", ")
#f2.write(str(p1_buttons_sums[i]) + ", ")
#f3.write(str(p1_board_util_sums[i]) + ", ")
#f4.write(str(p1_game_avgs[i]) + ", ")
#f5.write(str(p1_game_avgs[i] - p2_game_avgs[i]) + ", ")
print()
#print("QUILTS: ")
#for i in range(len(p1_quilts)):
# print(str(i) + ":")
# for row in range(len(p1_quilts[i])):
# for col in range(len(p1_quilts[i][row])):
# print(p1_quilts[i][row][col], end = "")
# print()
# print()
# print()
print("P2 (RANDOM AI) FEATURE AVERAGES")
for i in range(len(p2_button_gen_sums)):
print(
str(i * self.INTERVAL_SIZE) + " - " +
str((i * self.INTERVAL_SIZE) + self.INTERVAL_SIZE) +
" | Button Gen: " + str(p2_button_gen_sums[i]) +
", Buttons: " + str(p2_buttons_sums[i]) + ", Board Util: " +
str(p2_board_util_sums[i]))