def main():
HEURISTICS = [("Null", null_score),
("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [Agent(CustomPlayer(score_fn=h, **MM_ARGS),
"MM_" + name) for name, h in HEURISTICS]
ab_agents = [Agent(CustomPlayer(score_fn=h, **AB_ARGS),
"AB_" + name) for name, h in HEURISTICS]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
test_agents = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")]
print(DESCRIPTION)
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
if agentUT.name == 'ID_Improved' or agentUT.name == 'Student':
searches = len(agentUT.player.average_depths)
total_depth = sum(agentUT.player.average_depths)
avg = total_depth / searches
print('{} Avg Depth: {}'.format(agentUT.name, avg))
def worker():
# create an isolation board (by default 7x7)
player1 = RandomPlayer()
player2 = GreedyPlayer()
game = Board(player1, player2)
game.apply_move((2, 3))
q.put(game)
sleep(2)
game.apply_move((0, 5))
q.put(game)
sleep(2)
# play the remainder of the game automatically -- outcome can be "illegal
# move" or "timeout"; it should _always_ be "illegal move" in this example
winner, history, outcome = game.play(q)
print("\nWinner: {}\nOutcome: {}".format(winner, outcome))
def main():
# Define two agents to compare -- these agents will play from the same
# starting position against the same adversaries in the tournament
test_agents = [
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved"),
Agent(AlphaBetaPlayer(score_fn=custom_score), "AB_Custom"),
Agent(AlphaBetaPlayer(score_fn=custom_score_2), "AB_Custom_2"),
Agent(AlphaBetaPlayer(score_fn=custom_score_3), "AB_Custom_3"),
Agent(AlphaBetaPlayer(score_fn=custom_score_4), "AB_Custom_4"),
Agent(AlphaBetaPlayer(score_fn=custom_score_5), "AB_Custom_5"),
Agent(AlphaBetaPlayer(score_fn=custom_score_6), "AB_Custom_6"),
Agent(AlphaBetaPlayer(score_fn=custom_score_7), "AB_Custom_7")
]
# Define a collection of agents to compete against the test agents
cpu_agents = [
Agent(RandomPlayer(), "Random"),
Agent(MinimaxPlayer(score_fn=open_move_score), "MM_Open"),
Agent(MinimaxPlayer(score_fn=center_score), "MM_Center"),
Agent(MinimaxPlayer(score_fn=improved_score), "MM_Improved"),
Agent(AlphaBetaPlayer(score_fn=open_move_score), "AB_Open"),
Agent(AlphaBetaPlayer(score_fn=center_score), "AB_Center"),
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved")
]
# test_agents = [
# Agent(AlphaBetaPlayer(score_fn=custom_score), "AB_Custom"),
# Agent(AlphaBetaPlayer(score_fn=custom_score_2), "AB_Custom_2"),
# Agent(AlphaBetaPlayer(score_fn=custom_score_3), "AB_Custom_3"),
# Agent(AlphaBetaPlayer(score_fn=custom_score_4), "AB_Custom_4"),
# Agent(AlphaBetaPlayer(score_fn=custom_score_5), "AB_Custom_5"),
# Agent(AlphaBetaPlayer(score_fn=custom_score_6), "AB_Custom_6"),
# Agent(AlphaBetaPlayer(score_fn=custom_score_7), "AB_Custom_7")
# ]
# cpu_agents = [
# Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved")
# ]
print(DESCRIPTION)
print("{:^74}".format("*************************"))
print("{:^74}".format("Playing Matches"))
print("{:^74}".format("*************************"))
play_matches(cpu_agents, test_agents, NUM_MATCHES)
def main():
# Define two agents to compare -- these agents will play from the same
# starting position against the same adversaries in the tournament
test_agents = [
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved"),
Agent(AlphaBetaPlayer(score_fn=my_moves_heuristic),
"my_moves_heuristic"),
Agent(AlphaBetaPlayer(score_fn=maximise_player_moves),
"maximise_player_moves"),
Agent(AlphaBetaPlayer(score_fn=minimise_opponent_moves),
"minimise_opponent_moves"),
Agent(
AlphaBetaPlayer(
score_fn=maximise_ratio_of_player_to_opponent_moves),
"maximise_ratio_of_player_to_opponent_moves"),
Agent(
AlphaBetaPlayer(
score_fn=minimise_ratio_of_player_to_opponent_moves),
"minimise_ratio_of_player_to_opponent_moves")
]
# Define a collection of agents to compete against the test agents
cpu_agents = [
Agent(RandomPlayer(), "Random"),
Agent(MinimaxPlayer(score_fn=open_move_score), "MM_Open"),
Agent(MinimaxPlayer(score_fn=center_score), "MM_Center"),
Agent(MinimaxPlayer(score_fn=improved_score), "MM_Improved"),
Agent(AlphaBetaPlayer(score_fn=open_move_score), "AB_Open"),
Agent(AlphaBetaPlayer(score_fn=center_score), "AB_Center"),
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved")
]
print(DESCRIPTION)
print("{:^74}".format("*************************"))
print("{:^74}".format("Playing Matches"))
print("{:^74}".format("*************************"))
print()
play_matches(cpu_agents, test_agents, NUM_MATCHES)
def main():
# Define two agents to compare -- these agents will play from the same
# starting position against the same adversaries in the tournament
test_agents = [
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved"),
Agent(AlphaBetaPlayer(score_fn=custom_score), "AB_Custom"),
Agent(AlphaBetaPlayer(score_fn=custom_score_2), "AB_Custom_2"),
Agent(AlphaBetaPlayer(score_fn=custom_score_3), "AB_Custom_3")
]
"""
Uncomment to add grid search function agents to test_agents
"""
# print("{} grid search functions".format(len(gs_funcs)))
# for name, func in gs_funcs.items():
# test_agents.append(Agent(AlphaBetaPlayer(score_fn=func), "AB_{}".format(name)))
# print("{} grid search2 functions".format(len(gs2_funcs)))
# for name, func in gs2_funcs.items():
# test_agents.append(Agent(AlphaBetaPlayer(score_fn=func), "AB_GS2_{}".format(name)))
# Define a collection of agents to compete against the test agents
cpu_agents = [
Agent(RandomPlayer(), "Random"),
Agent(MinimaxPlayer(score_fn=open_move_score), "MM_Open"),
Agent(MinimaxPlayer(score_fn=center_score), "MM_Center"),
Agent(MinimaxPlayer(score_fn=improved_score), "MM_Improved"),
Agent(AlphaBetaPlayer(score_fn=open_move_score), "AB_Open"),
Agent(AlphaBetaPlayer(score_fn=center_score), "AB_Center"),
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved")
]
print(DESCRIPTION)
print("{:^74}".format("*************************"))
print("{:^74}".format("Playing Matches"))
print("{:^74}".format("*************************"))
play_matches(cpu_agents, test_agents, NUM_MATCHES)
def prepare_default_players():
# Define two agents to compare -- these agents will play from the same
# starting position against the same adversaries in the tournament
test_agents = [
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved"),
Agent(AdvancedAlphaBetaNegamaxPlayer(score_fn=custom_score_2),
"TranspositionTable"),
Agent(AdvancedAlphaBetaNegamaxPlayer_V2(score_fn=custom_score_2),
"OrderedMoves"),
Agent(AlphaBetaPlayer(score_fn=custom_score_3), "AB_Custom_3")
]
# # Define a collection of agents to compete against the test agents
cpu_agents = [
Agent(RandomPlayer(), "Random"),
Agent(MinimaxPlayer(score_fn=open_move_score), "MM_Open"),
Agent(MinimaxPlayer(score_fn=center_score), "MM_Center"),
Agent(MinimaxPlayer(score_fn=improved_score), "MM_Improved"),
Agent(AlphaBetaPlayer(score_fn=open_move_score), "AB_Open"),
Agent(AlphaBetaPlayer(score_fn=center_score), "AB_Center"),
Agent(AlphaBetaPlayer(score_fn=improved_score), "AB_Improved")
]
return test_agents, cpu_agents
def main():
for tour_num in range(NUM_TOURNAMENTS):
HEURISTICS = [("Null", null_score),
("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
mm_agents = [Agent(CustomPlayer(score_fn=h, **MM_ARGS),
"MM_" + name) for name, h in HEURISTICS]
ab_agents = [Agent(CustomPlayer(score_fn=h, **AB_ARGS),
"AB_" + name) for name, h in HEURISTICS]
random_agents = [Agent(RandomPlayer(), "Random")]
test_agents = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")]
for agentUT in test_agents:
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio = play_round(agents, NUM_MATCHES)
print("{:>10.2f}".format(win_ratio))
print("-----")
import timeit
from isolation import Board
from sample_players import (RandomPlayer, open_move_score,
improved_score, center_score)
from game_agent import (MinimaxPlayer, AlphaBetaPlayer, custom_score,
custom_score_2, custom_score_3)
# Create the test players
player1 = AlphaBetaPlayer(score_fn=improved_score)
player2 = RandomPlayer()
# Create an isolation board (by default 7x7)
game = Board(player1, player2, 8, 8)
forfeited_match = [[2, 3], [4, 4], [0, 4], [5, 6], [2, 5], [6, 4],
[1, 3], [4, 5], [0, 1], [2, 6], [2, 2]]
for move in forfeited_match:
game.apply_move(move)
# print(game.get_legal_moves())
# print(len(game.get_legal_moves()))
score = custom_score_2(game, player1)
time_millis = lambda: 1000 * timeit.default_timer()
move_start = time_millis()
time_left = lambda: 100 - (time_millis() - move_start)
def setUp(self):
reload(game_agent)
self.player1 = AlphaBetaPlayer()
self.player2 = RandomPlayer()
self.game = isolation.Board(self.player1, self.player2)
Created on Sat Jan 6 20:58:45 2018
@author: Kelvin
"""
from game_agent import (MinimaxPlayer, AlphaBetaPlayer, custom_score,
custom_score_2, custom_score_3, custom_score_4,
custom_score_5)
from sample_players import (RandomPlayer, open_move_score, improved_score,
center_score)
from isolation import Board
import random
import timeit
player1 = RandomPlayer()
player2 = AlphaBetaPlayer(score_fn=improved_score)
player3 = AlphaBetaPlayer(score_fn=custom_score)
player4 = AlphaBetaPlayer(score_fn=custom_score_2)
player5 = AlphaBetaPlayer(score_fn=custom_score_3)
player6 = AlphaBetaPlayer(score_fn=custom_score_4)
player7 = AlphaBetaPlayer(score_fn=custom_score_5)
time_limit = 150
time_millis = lambda: 1000 * timeit.default_timer()
players = [player2, player3, player4, player5, player6, player7]
i = 0
for p in players:
def main():
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
test_agents = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"),
Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")
]
start_time = time.time()
set_score = list()
complete_score = list()
our_win = 0
for i in range(5):
print("\nSet n.{} - Results:".format(i + 1))
print("----------")
for agentUT in test_agents:
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio = play_round(agents, NUM_MATCHES)
set_score.append(win_ratio)
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
print("{!s:<15}{:>10.2f}%".format("Ratio",
(set_score[1] - set_score[0])))
complete_score.append(set_score[1] - set_score[0])
if set_score[1] >= set_score[0]:
our_win = our_win + 1
set_score.clear()
if our_win >= 3:
print("\nYOU WIN! \o/ --> {} - {}".format(our_win, 5 - our_win))
else:
print("\nYOU LOSE. :( --> {} - {}".format(our_win, 5 - our_win))
print("--- Match time: {} minutes ---".format(
(time.time() - start_time) / 60))
def main_mine():
HEURISTICS = [("Null", null_score),
("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [Agent(CustomPlayer(score_fn=h, **MM_ARGS),
"MM_" + name) for name, h in HEURISTICS]
ab_agents = [Agent(CustomPlayer(score_fn=h, **AB_ARGS),
"AB_" + name) for name, h in HEURISTICS]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
print("NEW RUN ******************************************************************")
from game_agent import custom_score_diff_in_free_percent_of_board
from game_agent import custom_score_diff_in_mine_and_double_opponent
from game_agent import custom_score_diff_in_mine_and_double_opponent_chase_incase_of_tie
from game_agent import custom_score_diff_in_mine_and_double_opponent_run_away_incase_of_tie
from game_agent import custom_score_divide_own_by_opponent
from game_agent import custom_score_my_open_moves
from game_agent import custom_score_simple
from game_agent import custom_score_diff_in_mine_and_double_opponent_closest_to_center_tie
from game_agent import custom_score_diff_in_opp_and_double_mine
test_agents = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
Agent(CustomPlayer(score_fn=custom_score_my_open_moves, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_simple, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_diff_in_mine_and_double_opponent, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_diff_in_opp_and_double_mine, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_diff_in_mine_and_double_opponent_chase_incase_of_tie, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_diff_in_mine_and_double_opponent_run_away_incase_of_tie, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_diff_in_mine_and_double_opponent_closest_to_center_tie, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_divide_own_by_opponent, **CUSTOM_ARGS), "Student"),
Agent(CustomPlayer(score_fn=custom_score_diff_in_free_percent_of_board, **CUSTOM_ARGS), "Student")
]
print(DESCRIPTION)
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
def main():
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
PVS_ARGS = {"method": 'pvs', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [
Agent(RandomPlayer(), "random"),
Agent(GreedyPlayer(), "greedy")
]
THE_agent = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved")
]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
#test_agents = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
# Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")]
test_agents = [
#Agent(CustomPlayer(score_fn=my_little_killer, **CUSTOM_ARGS), "my_little_killer"),
#Agent(CustomPlayer(score_fn=you_cant_catch_me_adaptive, **CUSTOM_ARGS), "you_cant_catch_me_adaptive"),
#Agent(CustomPlayer(score_fn=reversed_aggression, **CUSTOM_ARGS), "reversed_aggression"),
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"),
#Agent(CustomPlayer(score_fn=as_far_aggression, **CUSTOM_ARGS), "as_far_aggression"),
#Agent(CustomPlayer(score_fn=super_improve_score, **CUSTOM_ARGS), "super_improve_score"),
#Agent(CustomPlayer(score_fn=random_1, **CUSTOM_ARGS), "random"),
#Agent(CustomPlayer(score_fn=block_in_two, **CUSTOM_ARGS), "block_in_two"),
Agent(CustomPlayer(score_fn=free_in_two, **CUSTOM_ARGS),
"free_in_two"),
Agent(CustomPlayer(score_fn=blind_aggression, **CUSTOM_ARGS), "blind"),
#Agent(CustomPlayer(score_fn=you_cant_catch_me, **CUSTOM_ARGS), "you_cant_catch_me"),
#Agent(CustomPlayer(score_fn=central_knight, **CUSTOM_ARGS), "central_knight"),
#Agent(CustomPlayer(score_fn=division, **CUSTOM_ARGS), "division"),
Agent(CustomPlayer(score_fn=combine, **CUSTOM_ARGS), "The chosen one"),
#Agent(CustomPlayer(score_fn=killer_combine, **CUSTOM_ARGS), "killer_combine")
]
print(DESCRIPTION)
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + [agentUT]
human = THE_agent + [Agent(HumanPlayer(), "human")]
if USE_HUMAN:
win_ratio = play_round(human, NUM_MATCHES)
else:
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
move_cache[str(
np.fliplr(flipud_game_state))] = (depth, best_fliplr_flipud_move,
best_val)
# move_cache[str(np.flipud(fliplr_game_state))] = (depth, best_move, best_val)
move_cache[str(rot_90)] = (depth, best_rot_90_move, best_val)
move_cache[str(rot_180)] = (depth, best_rot_180_move, best_val)
move_cache[str(np.flipud(rot_90))] = (depth, best_flipud_rot_90_move,
best_val)
move_cache[str(np.flipud(rot_180))] = (depth, best_flipud_rot_180_move,
best_val)
from isolation import Board
from sample_players import RandomPlayer
from sample_players import null_score
if __name__ == "__main__":
counter = 0
# (self, search_depth = 3, score_fn = custom_score,
# iterative = True, method = 'minimax', timeout = 10.)
customPlayer = CustomPlayer(3, null_score)
randomPlayer = RandomPlayer()
game = Board(customPlayer, 3, 3)
# player_1 = CustomPlayer(2)
num_trials = orginal_num_trials = 5
while num_trials > 0:
num_trials = num_trials - 1
while True:
game = Board(customPlayer, randomPlayer)
game.play()
def tryall():
# Setup all the permutations
search_depths = [3, 5]
score_functions = [
null_score, open_move_score, improved_score, custom_score
]
methods = ["minimax", "alphabeta"]
iteratives = [False, True]
# Setup the players:
player1_agents = []
player2_agents = [Agent(RandomPlayer(), "random_player")]
for score_function in score_functions:
for method in methods:
for iterative in iteratives:
for search_depth in search_depths:
params = {
"search_depth": search_depth,
"method": method,
"iterative": iterative,
"score_fn": score_function
}
name = "{:16s} / {:9s} / DEPTH({:1d}) / ITER({:1b})".format(
score_function.__name__, method, search_depth,
iterative)
if score_function is improved_score or score_function is custom_score:
player1_agents.append(
Agent(CustomPlayer(**params), name))
else:
player2_agents.append(
Agent(CustomPlayer(**params), name))
# Launch the matches for each pair:
counter = 0
for player1 in player1_agents:
for player2 in player2_agents:
counter += 1
# Play some games:
wins = 0.
num_wins = {player1: 0, player2: 0}
num_timeouts = {player1: 0, player2: 0}
num_invalid_moves = {player1: 0, player2: 0}
print("{:2d}: {:50s}\t--VS--\t {:50s}".format(
counter, player1.name, player2.name),
end=' ')
for _ in range(0, NUM_MATCHES):
games = [
Board(player1.player, player2.player),
Board(player2.player, player1.player)
]
# initialize both games with a random move and response
for _ in range(2):
move = random.choice(games[0].get_legal_moves())
games[0].apply_move(move)
games[1].apply_move(move)
# play both games and tally the results
for game in games:
winner, moves, termination = game.play(
time_limit=TIME_LIMIT)
if player1.player == winner:
num_wins[player1] += 1
if termination == "timeout":
num_timeouts[player2] += 1
else:
num_invalid_moves[player2] += 1
elif player2.player == winner:
num_wins[player2] += 1
if termination == "timeout":
num_timeouts[player1] += 1
else:
num_invalid_moves[player1] += 1
winratio = 100 * (num_wins[player1] /
(num_wins[player1] + num_wins[player2]))
print("==>: Wins {:3.0f} % {:5s} ({:2d} to {:2d}) / Timeouts ({:2d} to {:2d})".\
format(winratio, \
"..| " if winratio <= 50 else " |..", \
int(num_wins[player1]), int(num_wins[player2]), \
int(num_timeouts[player1]), int(num_timeouts[player1])))
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
# test_agents = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
# Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")]
#test_agents = [ Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student"),
# Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved")]
my_agent = Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")
adversary = Agent(RandomPlayer(), "Random")
# game = Board(my_agent.player, adversary.player,7,7)
def main():
HEURISTICS = [("Null", null_score),
("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# w1 = [1]
# custom_agents = [Agent(CustomPlayer(w2=h, **CUSTOM_ARGS),
# "custom_" + str(h)) for h in w2_list]
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [Agent(CustomPlayer(score_fn=h, **MM_ARGS),
"MM_" + name) for name, h in HEURISTICS]
ab_agents = [Agent(CustomPlayer(score_fn=h, **AB_ARGS),
"AB_" + name) for name, h in HEURISTICS]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
ID_Improved_agent = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved")]
Student_agent = [Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")]
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + ID_Improved_agent[0].name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + ID_Improved_agent
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(ID_Improved_agent[0].name, win_ratio))
print(DESCRIPTION)
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + Student_agent[0].name ))
print("*************************")
# agents = ID_Improved_agent + Student_agent
# win_ratio = play_round(agents, NUM_MATCHES)
w2_list = [0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4]
w3_list = [0, 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4]
for w2_ in w2_list:
for w3_ in w3_list:
custom_agent_test = [Agent(CustomPlayerTest(w2=w2_, w3=w3_, **CUSTOM_ARGS), "custom_" +str(w2_) + "_" +str(w3_))]
agents = random_agents + mm_agents + ab_agents + custom_agent_test
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(custom_agent_test[0].name, win_ratio))
print("{!s:<15}{:>10.2f}%".format('likelihood_of_superiority : ',
likelihood_of_superiority(win_ratio, 2 * NUM_MATCHES)))
def tournament_schedule(args):
"args: epochs, matches, self_play_dir start_model, pool_dir, out_pool_dir"
base_model = args.pool_dir + args.start_model
policy = torch.load(base_model)
modelglob = glob.glob(args.pool_dir + '*')
inactive = [["MM_Improved", MinimaxPlayer(score_fn=improved_score)],
["AB_Open", AlphaBetaPlayer(score_fn=open_move_score)],
["AB_Center", AlphaBetaPlayer(score_fn=open_move_score)],
["AB_Improved", AlphaBetaPlayer(score_fn=improved_score)]]
agents = {"Random": RandomPlayer(),
"Random2": RandomPlayer(),
"MM_Open": MinimaxPlayer(score_fn=open_move_score),
"MM_center": MinimaxPlayer(score_fn=center_score)}
def print_round(win_pct, match, opp_name):
print("win_pct:{}, match:{}, vs:{} ---- base:{}".format(win_pct, match, opp_name, args.start_model))
defeated = []
models = modelglob + list(agents.keys())
best_hist = []
start_time = time.strftime("%Y-%m-%d-%H:%M")
if args.q == 0:
game_fn = play_game_rl
else:
game_fn = play_game_questionable
for match in range(args.matches):
# load an Opponent
opp_model_loc = random.choice(models)
if opp_model_loc in agents:
opp_name = opp_model_loc
opponent_player = agents[opp_model_loc]
else:
opp_name = os.path.basename(opp_model_loc)
opponent_player = loadNNPlayer(opp_model_loc)
# play game
policy, win_pct = game_fn(policy, opponent_player, opp_name, args, policy_id='best_in')
print_round(win_pct, match, opp_name)
if win_pct >= 0.99:
print("---------------")
print("{} has been defeated at match {}".format(opp_name, match))
try:
if opp_model_loc in agents:
if len(inactive) > 0:
print("new Search opponent has been created: {}".format(inactive[0][0]))
models.append(inactive[0][0])
agents[inactive[0][0]] = inactive[0][1]
del inactive[0]
else:
add_pool = glob.glob(args.out_pool_dir + '*')
for nn in add_pool:
if nn not in defeated:
models.append(nn)
print("new NN opponent has been created: {}".format(nn))
break
models.remove(opp_model_loc)
defeated.append(opp_model_loc)
except:
print("you are an idiot. remember to write a unittest. f**k state")
if (match + 1) % 50 == 0:
policy_loc = save_model_fmt(policy, args.out_pool_dir, start_time, '-top-', match)
best_hist.append([opp_name, win_pct])
print("---------------Tournament complete------------")
print("---------------defeated opponents--------------")
# (print(x) for x in defeated)
print("---------------FINAL RESULTS------------------")
# running_total = 0
for m in best_hist:
print("win_pct:{}, vs: {}".format(m[1], m[0]))
def mainMod():
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
# test_agents = [Agent(CustomPlayer(score_fn=custom_score_normalizedByBlankSpaces, **CUSTOM_ARGS), "StudentNormalized"),
# Agent(CustomPlayer(score_fn=custom_score_edgeAndCornerLimiting, **CUSTOM_ARGS), "StudentECLimiting"),
# Agent(CustomPlayer(score_fn=custom_score_distaceWeightedPositions, **CUSTOM_ARGS), "StudentDistWeighted")]
test_agents = [
Agent(
CustomPlayer(score_fn=custom_score_distaceWeightedPositions,
**CUSTOM_ARGS), "StudentDistWeighted")
]
idImprovedAgent = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved")
]
#all_agents = test_agents+[idImprovedAgent]
print(DESCRIPTION)
results = dict()
for agentUT in test_agents:
for i in range(50):
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agentName = agentUT.name
agents = idImprovedAgent + [agentUT]
opponent = idImprovedAgent[0].name
key = (opponent + "_" + agentName)
win_ratio = play_round(agents, NUM_MATCHES)
if (key in results):
results.get(key).append(win_ratio)
else:
results[key] = [win_ratio]
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
with open('HeuristicsResults.csv', 'w') as f:
[
f.write('{0},{1}\n'.format(key, value))
for key, v in results.items() for value in v
]
def main(argv):
USAGE = """usage: tournament_mp.py [-m <number of matches>] [-p <pool size>] [-o <outputfile>]
-m number of matches: optional number of matches (each match has 4 games) - default is 5
-p pool size: optional pool size - default is 3
-o output file: optional output file name - default is results.txt"""
# Assumes 2 x dual-core CPUs able to run 3 processes relatively
# uninterrupted (interruptions cause get_move to timeout)
pool_size = 3
outputfilename = 'results.txt'
num_matches = NUM_MATCHES
try:
opts, args = getopt.getopt(argv, "hm:p:o:",
["matches=", "poolsize=", "ofile="])
except getopt.GetoptError as err:
print(err)
print(USAGE)
sys.exit(2)
for opt, arg in opts:
if opt in ["-h", "--help"]:
print(USAGE)
sys.exit()
elif opt in ("-m", "--matches"):
num_matches = int(arg)
elif opt in ("-p", "--poolsize"):
pool_size = int(arg)
elif opt in ("-o", "--ofile"):
outputfilename = arg
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_opponents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_opponents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_opponents = [Agent(RandomPlayer(), "Random")]
all_opponents = random_opponents + mm_opponents + ab_opponents
test_agents = []
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
test_agents.append(
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"))
# Create all the parameterized evaluation function objects
# Then create all the test agents using those eval functions
params = [(a, b, c, d, e, f) for a in range(1, 3) for b in range(0, 3)
for c in range(-1, 2) for d in range(1, 3) for e in range(0, 3)
for f in range(0, 1)]
#params = [(0,0,0,0,0,0)]
#params = [(1, 2, -1, 1, 2, 0)]
for param in params:
eval_obj = ParameterizedEvaluationFunction(param)
test_agents.append(
Agent(CustomPlayer(score_fn=eval_obj.eval_func, **CUSTOM_ARGS),
"Student " + str(param)))
# Put the start time in the output file
with open(outputfilename, mode='a') as ofile:
ofile.write(
'*******************************************************************************************\n'
)
ofile.write(
'Starting Isolation tournament with %d test agents, %d games per round, and %d sub-processes\n'
% (len(test_agents), num_matches * 4, pool_size))
ofile.write('Tournament started at %s\n' %
(datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')))
# Run the tournament!
with Pool(processes=pool_size) as pool:
results = []
for agentUT in test_agents:
results.append(
pool.apply_async(play_round,
args=(all_opponents, agentUT, num_matches)))
# Write the output... flush each time as it takes a long time to run
with open(outputfilename, mode='a') as ofile:
for result in results:
agent, res = result.get()
ofile.write('%s got %2.2f\n' % (agent, res))
ofile.flush()
ofile.write(
'Tournament complete at: %s\n' %
(datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')))
ofile.write(
'*******************************************************************************************\n\n'
)
def main():
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
test_agents = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"),
Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")
]
test_agents = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"),
Agent(CustomPlayer(score_fn=aggressive_move_heuristic, **CUSTOM_ARGS),
"AggressiveMovesStudent"),
Agent(CustomPlayer(score_fn=relaxed_move_heuristic, **CUSTOM_ARGS),
"RelaxedMovesStudent")
]
test_agents = [
Agent(
CustomPlayer(score_fn=relaxed_move_aggressive_distance,
**CUSTOM_ARGS), "relaxed_move_aggressive_distance"),
Agent(
CustomPlayer(score_fn=relaxed_move_relaxed_distance,
**CUSTOM_ARGS), "relaxed_move_relaxed_distance"),
Agent(
CustomPlayer(score_fn=relaxed_move_relaxed_distance_norm,
**CUSTOM_ARGS), "relaxed_move_relaxed_distance_norm"),
Agent(
CustomPlayer(score_fn=relaxed_move_aggressive_distance_norm,
**CUSTOM_ARGS),
"relaxed_move_aggressive_distance_norm"),
Agent(CustomPlayer(score_fn=aggressive_move_heuristic, **CUSTOM_ARGS),
"aggressive_move_heuristic"),
Agent(CustomPlayer(score_fn=relaxed_move_heuristic, **CUSTOM_ARGS),
"relaxed_move_heuristic"),
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved")
]
res = pd.Series(name='player')
agent_names = []
print(DESCRIPTION)
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
res[agentUT.name] = win_ratio
agent_names.append(agentUT.name)
res.reset_index().to_hdf('data/full_run_{}.h5'.format(
"_".join(agent_names)),
'test',
mode='w')
def main():
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
test_agents = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"),
# Agent(CustomPlayer(score_fn=more_improved_score, **CUSTOM_ARGS), "More_Improved"),
Agent(
CustomPlayer(score_fn=linear_ratio_improved_score, **CUSTOM_ARGS),
"Linear_Improved"),
Agent(
CustomPlayer(score_fn=nonlinear_ratio_improved_score,
**CUSTOM_ARGS), "Non_Linear_Improved"),
Agent(CustomPlayer(score_fn=second_moves_score, **CUSTOM_ARGS),
"Second_Moves"),
Agent(
CustomPlayer(score_fn=second_moves_in_middle_game_score,
**CUSTOM_ARGS), "Second_Moves_In_Middle_Game"),
Agent(CustomPlayer(score_fn=all_boxes_can_move_score, **CUSTOM_ARGS),
"All_Boxes_Can_Move"),
]
print(DESCRIPTION)
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
# agents = random_agents + mm_agents + ab_agents + [agentUT]
agents = test_agents + [agentUT]
agents.remove(agentUT)
win_ratio = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
def main():
HEURISTICS = [("Null", null_score),
("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {"search_depth": 5, "method": 'alphabeta', "iterative": False}
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False}
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# For just comparing the correctness of heuristic.. using a shart search depth could help
#CUSTOM_ARGS = {"method": 'minimax', 'iterative': False, 'search_depth': 3}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [Agent(CustomPlayer(score_fn=h, **MM_ARGS),
"MM_" + name) for name, h in HEURISTICS]
ab_agents = [Agent(CustomPlayer(score_fn=h, **AB_ARGS),
"AB_" + name) for name, h in HEURISTICS]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
#test_agents = [Agent(CustomPlayer(score_fn=combined_improved_density_at_end, **CUSTOM_ARGS), "Combined improved"),
# Agent(CustomPlayer(score_fn=diff_density, **CUSTOM_ARGS), "Diff density"),
# Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved")
# Agent(CustomPlayer(score_fn=combined_full, **CUSTOM_ARGS), "Combined full"),
# ]
#test_agents = [Agent(CustomPlayer(score_fn=agrressive_first_then_preserving, **CUSTOM_ARGS), "Agressive first then preserving"),
# Agent(CustomPlayer(score_fn=custom_score_location_using_hash_table, **CUSTOM_ARGS), "Location using hash"),
# Agent(CustomPlayer(score_fn=inverted_to_center, **CUSTOM_ARGS), "Inverted Location using hash"),
# Agent(CustomPlayer(score_fn=custom_score_loser, **CUSTOM_ARGS), "loser"),
# Agent(CustomPlayer(score_fn=preserving_score_with_self, **CUSTOM_ARGS), "Preserving score with self"),
# Agent(CustomPlayer(score_fn=aggressive_score_with_self, **CUSTOM_ARGS), "Agressive score with self"),
# Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Custom score using coefficient"),
# Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved")]
test_agents = [Agent(CustomPlayer(score_fn=combined_improved_density_at_end, **CUSTOM_ARGS), "Combined_improved_and_density"),
Agent(CustomPlayer(score_fn=diff_density, **CUSTOM_ARGS), "Diff_density"),
Agent(CustomPlayer(score_fn=improved_with_sleep, **CUSTOM_ARGS), "ID_Improved_slow"),
Agent(CustomPlayer(score_fn=distance_to_center, **CUSTOM_ARGS), "Distance_to_center"),
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved")
]
print(DESCRIPTION)
full_res = {}
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio, res = play_round(agents, NUM_MATCHES)
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
full_res[agentUT.name] = res
with open('out.json', 'w') as f:
json.dump(full_res, f)
def main():
HEURISTICS = [("Null", null_score), ("Open", open_move_score),
("Improved", improved_score)]
AB_ARGS = {
"search_depth": 5,
"method": 'alphabeta',
"iterative": False
} # 5
MM_ARGS = {"search_depth": 3, "method": 'minimax', "iterative": False} # 3
CUSTOM_ARGS = {"method": 'alphabeta', 'iterative': True}
# Create a collection of CPU agents using fixed-depth minimax or alpha beta
# search, or random selection. The agent names encode the search method
# (MM=minimax, AB=alpha-beta) and the heuristic function (Null=null_score,
# Open=open_move_score, Improved=improved_score). For example, MM_Open is
# an agent using minimax search with the open moves heuristic.
mm_agents = [
Agent(CustomPlayer(score_fn=h, **MM_ARGS), "MM_" + name)
for name, h in HEURISTICS
]
ab_agents = [
Agent(CustomPlayer(score_fn=h, **AB_ARGS), "AB_" + name)
for name, h in HEURISTICS
]
random_agents = [Agent(RandomPlayer(), "Random")]
# ID_Improved agent is used for comparison to the performance of the
# submitted agent for calibration on the performance across different
# systems; i.e., the performance of the student agent is considered
# relative to the performance of the ID_Improved agent to account for
# faster or slower computers.
#test_agents = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
# Agent(CustomPlayer(score_fn=custom_score, **CUSTOM_ARGS), "Student")]
# test_agents = [Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS), "ID_Improved"),
# Agent(CustomPlayer(score_fn=custom_score_mixed_centrality, **CUSTOM_ARGS), "mixed_centrality")]
# FIXME
if True:
test_agents = [
Agent(CustomPlayer(score_fn=improved_score, **CUSTOM_ARGS),
"ID_Improved"),
Agent(
CustomPlayer(score_fn=custom_score_legal_moves_left_balance,
**CUSTOM_ARGS), "Legal Moves\nLeft Balance"),
Agent(
CustomPlayer(score_fn=custom_score_dominating_space,
**CUSTOM_ARGS), "Dominating\nSpace"),
Agent(
CustomPlayer(score_fn=custom_score_dominating_moves,
**CUSTOM_ARGS), "Dominating\nMoves"),
Agent(
CustomPlayer(score_fn=custom_score_schadenfreude,
**CUSTOM_ARGS), "Schadenfreude"),
Agent(
CustomPlayer(score_fn=custom_score_mixed_centrality,
**CUSTOM_ARGS), "Mixed \nCentrality")
]
results_prepared = {}
for run in range(0, 3):
print(DESCRIPTION)
print("============================")
print("Run:", run)
print("============================")
results = []
for agentUT in test_agents:
print("")
print("*************************")
print("{:^25}".format("Evaluating: " + agentUT.name))
print("*************************")
agents = random_agents + mm_agents + ab_agents + [agentUT]
win_ratio = play_round(agents, NUM_MATCHES)
results.append((win_ratio, agentUT.name))
print("\n\nResults:")
print("----------")
print("{!s:<15}{:>10.2f}%".format(agentUT.name, win_ratio))
print("All Results:")
for index, (score, name) in enumerate(results):
if name == "ID_Improved":
break
assert ("ID_Improved" == results[index][1])
baseline_score = results[index][0]
print("baseline_score", baseline_score)
del results[index]
#results_prepared = [[name, score-baseline_score] for (score, name) in sorted(results)]
for (score, name) in results:
results_prepared[name] = results_prepared.get(
name, []) + [score - baseline_score]
#labels, ys = zip(*results_prepared)
import visualize
import numpy as np
print("results_prepared", results_prepared)
visualize.visualize_as_bar(results_prepared, "Absolute Gains to Baseline")
# Create table
averaged_results = {}
for key in results_prepared:
averaged_results[key] = np.mean(results_prepared[key])
print("averaged_results", averaged_results)