def test_alphabeta_vs_penalty(self):
t = TTT(3)
player1 = ABPruning(3)
player2 = ABPruning(3)
player2.set_penalty(0.2)
scores = {-4: 0, -2: 0, 0: 0, 1: 0, 3: 0, 5: 0}
game_played = 0
while game_played < 11:
if t.is_terminated():
score = t.get_score()
scores[score] += 1
game_played += 1
t = TTT(3)
pass
mover = t.get_mover()
if mover == 1:
[_, move] = player1.get(t.get_state(), mover)
t.put(move)
elif mover == -1:
[_, move] = player2.get(t.get_state(), mover)
t.put(move)
pass
print(scores)
wrong_cases = scores[-4] + scores[-2]
self.assertTrue(wrong_cases == 0)
def test_minimax_1(self):
t = TTT(3)
state = [[1, -1, 0], [-1, 1, 0], [0, 0, 0]]
state = np.array(state, dtype='int')
state = state.reshape(-1)
t._state = state
[score, move] = minimax(t.get_state(), 1, t)
self.assertListEqual(list(state), list(t.get_state()))
self.assertEqual(8, move)
self.assertEqual(5, score)
def test_alpha_beta_1(self):
t = TTT(3)
player = ABPruning(3)
state = [[1, -1, 0], [-1, 1, 0], [0, 0, 0]]
state = np.array(state, dtype='int')
state = state.reshape(-1)
t._state = state
[score, move] = player.get(t.get_state(), 1)
self.assertListEqual(list(state), list(t.get_state()))
self.assertEqual(8, move)
self.assertEqual(5, score)
def initialize_minimax(filepath: str, size=3):
table = {}
t = TTT(size)
minimax_save(t.get_state(), t.get_mover(), t, table)
with open(filepath, 'wb') as f:
pickle.dump(table, f)
return
def run_game(self, agent1, agent2, size=3):
t = TTT(size)
for i in range(size * size):
agent = agent1 if i % 2 == 0 else agent2
inferred = agent(t.get_state())
t.put(inferred)
if t.is_terminated():
break
return t.get_result()
def test_penalty_vs_penalty(self):
t = TTT(3)
player1 = ABPruning(3)
player1.set_penalty(0.7)
player2 = ABPruning(3)
player2.set_penalty(0.7)
games_played = 1
scores = set()
case1 = {1, 2, 3, 4}
case2 = {-1, -2, -3}
case3 = {0}
while True:
if t.is_terminated():
score = t.get_score()
scores.add(score)
# check whether if win,draw,lose all happened
wins = case1 & scores
loses = case2 & scores
draw = case3 & scores
if len(wins) > 0:
if len(loses) > 0:
if len(draw) > 0:
break
t = TTT(3)
games_played += 1
pass
mover = t.get_mover()
if mover == 1:
[_, move] = player1.get(t.get_state(), mover)
t.put(move)
elif mover == -1:
[_, move] = player2.get(t.get_state(), mover)
t.put(move)
self.assertTrue(len(scores) > 2)
def initialize_state_indices(filepath: str, size=3):
table = {'current': 0} # store state:index pair
t = TTT(size)
def dfs(state, mover: int, table=table) -> None:
# store if the state is new one :
encoded_state = t.get_encoded_state(state)
if not encoded_state in table:
table[encoded_state] = table['current']
table['current'] += 1
assert type(table[encoded_state]) is int
next_mover = 1 if mover is -1 else -1
available_moves = t.get_available_positions(state)
for i in available_moves:
next_state = state.copy()
next_state[i] = mover
if not t.is_terminated(next_state):
dfs(next_state, next_mover)
return
# indexing start :
initial_mover = t.get_mover()
initial_state = t.get_state()
print('indexing start :')
dfs(initial_state, initial_mover)
# simple validate :
num_visited = table['current']
del (table['current'])
num_stored = len(table)
print(f'visited states : {num_visited}')
print(f'stored states : {num_stored}')
assert num_stored == num_visited
indices = set(table.values())
assert len(indices) == len(table)
sample_index = list(table.values())[1]
assert type(sample_index) is int
# save :
print('saving... ', end='')
with open(filepath, 'wb') as f:
pickle.dump(table, f)
print('done')
return
def test_alphabeta_vs_minimax(self):
t = TTT(3)
minimax_player = minimax_load('results/minimax.pk')
alphabeta_player = ABPruning(3)
moves = 0
print('Moves : 0 ', end='')
while True:
if t.get_mover() == 1:
[_, best_move] = alphabeta_player.get(t.get_state(),
t.get_mover())
elif t.get_mover() == -1:
[_, best_move] = minimax_player(t.get_state())
t.put(best_move)
moves += 1
print(f'{moves} ', end='')
if t.is_terminated():
break
pass
print('final state')
print(t)
self.assertEqual(t.check_winner()['winner'], 0)
def _train_against(self,opponent_agent:Callable[[np.ndarray],int],numOfGames:int)->None:
agent_q_turn = self._is_first_mover
for _ in tqdm(range(numOfGames)):
game = TTT(self._size)
turn = True
# one complete game :
# prev state, action taken are from agent's turn
# next state is from opponent's turn.
# update in opponent's turn
encoded_prev_state = None
move_taken = None
encoded_next_state = None
while True:
if turn is agent_q_turn:
# Q turn :
if game.is_terminated():
break
else:
possible_moves = game.get_available_positions()
encoded_prev_state = game.get_encoded_state()
move_taken = self._epsilon_greedy_train(encoded_prev_state,possible_moves)
game.put(move_taken)
pass
pass
else:
# opponent's turn :
if not game.is_terminated():
state = game.get_state()
# move below is considered as random (sampling procedure) :
move = opponent_agent(state)
game.put(move)
pass
encoded_next_state = game.get_encoded_state()
score = game.get_score()
if encoded_prev_state is not None:
# : to avoid just after first move case ( in case of Q is second mover )
self.update(encoded_prev_state,move_taken,encoded_next_state,score)
pass
turn = not turn
pass
return None
def test_minimax_vs_minimax(self):
size = 3
t = TTT(size)
filepath = 'results/minimax.pk'
minimax_loaded = minimax_load(filepath)
moves = 0
while True:
[_, best_move] = minimax_loaded(t.get_state())
t.put(best_move)
moves += 1
if t.is_terminated():
break
pass
self.assertEqual(t.check_winner()['winner'], 0)
pass
def test_alphabeta_vs_alphabeta(self):
t = TTT(3)
player = ABPruning(3)
moves = 0
print('Moves : 0 ', end='')
while True:
[_, best_move] = player.get(t.get_state(), t.get_mover())
t.put(best_move)
moves += 1
print(f'{moves} ', end='')
if t.is_terminated():
break
pass
print('final state')
print(t)
self.assertEqual(t.check_winner()['winner'], 0)
class GameWindow(tk.Toplevel):
"""Game UI"""
def __init__(self, user_first: bool, size=3, *args, **kwargs):
super().__init__(*args, **kwargs)
# state variables
self._user_first = user_first
self._t = TTT(size)
self._agent: Callable[[np.ndarray], int]
self._num_of_moves = 0
self._state_history = [self._t.get_state()]
# UI accessors
self._history_scale: tk.Scale
self._player_labels: Dict[int, tk.Label] # key : 1,2
self._buttons = []
# UI initialization
self.title(f'TTT')
self._make_top_frame()
self._make_board(size)
self._make_bottom_frame(size)
return
#region Public Methods
def set_agent(self, agent: Callable[[np.ndarray], int], name: str) -> None:
self._agent = agent
return
def get_result(self) -> dict:
return self._t.get_result()
#endregion
#region Put UI Components
def _make_top_frame(self):
frame = tk.Frame(self)
if self._user_first:
text1 = 'O : User'
text2 = 'X : AI'
else:
text1 = 'O : AI'
text2 = 'X : User'
label1 = tk.Label(frame, text=text1)
label2 = tk.Label(frame, text=text2)
label1.pack()
label2.pack()
frame.pack()
return
def _make_board(self, size):
board = tk.Frame(self)
buttons = self._buttons
num_of_buttons = size * size
for i in range(num_of_buttons):
b = tk.Button(board,
width=3,
height=1,
font=('Helvetica', 30),
activebackground='white',
command=lambda num=i: self._on_click_board(num))
buttons.append(b)
b.grid(column=i % size, row=int(i / size))
pass
board.pack()
return
def _make_bottom_frame(self, size):
frame = tk.Frame(self)
history_scale = tk.Scale(frame,
command=self._on_scale_move,
orient='horizontal',
from_=0,
to=0)
history_scale.grid(row=0, columnspan=2)
self._history_scale = history_scale
restart_button = tk.Button(frame,
text="Restart",
command=self._on_click_reset)
exit_button = tk.Button(frame, text="Exit", command=self.destroy)
restart_button.grid(row=1, column=0)
exit_button.grid(row=1, column=1)
frame.pack()
return
#endregion
#region Event Handlers
def _on_click_board(self, position: int):
state_num = int(self._history_scale.get())
is_rewinded = not (self._num_of_moves == state_num)
if is_rewinded:
# reset the game to the rewinded one :
state_to_force = self._state_history[state_num]
self._t.set_state(state_to_force)
self._num_of_moves = self._t._num_moves
self._state_history = self._state_history[0:(self._num_of_moves +
1)]
pass
self._t.put(position)
current_state = self._t.get_state()
self._state_history.append(current_state)
self._num_of_moves += 1
self._history_scale.configure(to=self._num_of_moves)
self._history_scale.set(self._num_of_moves)
"""
[issue] If this procedure is called by button.invoke()
then it doesn't invoke the scale's command _on_scale_move.
So call it manually (and hence, called twice in user's turn) :
"""
self._on_scale_move(self._num_of_moves)
return
def _on_scale_move(self, state_num):
state_num = int(state_num)
first_mover_turn = True if state_num % 2 == 0 else False
user_turn = first_mover_turn == self._user_first
self._set_board(state_num, user_turn)
if self.get_result()['terminated']:
return
if state_num == len(self._state_history) - 1:
if user_turn:
pass
else:
if hasattr(self, '_agent'):
self._on_agent_turn(state_num)
pass
else:
# : agent's turn but it's a previous state
pass
return
def _on_click_reset(self):
self._num_of_moves = 0
self._state_history = self._state_history[0:1]
self._t.set_state(self._state_history[0])
self._history_scale.configure(to=0)
self._history_scale.set(0)
self._set_board(0, self._user_first == True)
return
#endregion
#region Private Methods
def _on_agent_turn(self, state_num: int):
# TODO : async progress bar
state = self._state_history[state_num]
move = self._agent(state)
button = self._buttons[move]
button.configure(state='normal')
button.invoke()
return
def _set_board(self, state_num: int, user_turn: bool):
"""Modify board UI"""
to_state = self._state_history[state_num]
result = self._t.get_result(to_state)
terminated = result['terminated']
lines = result['lines']
lines = sum(lines, []) # flattening
for p in range(len(to_state)):
move = int(to_state[p])
of_line = p in lines
self._modify_button(p, move, user_turn, terminated, of_line)
return
def _modify_button(self,
button_position: int,
mover: int,
move_allowed: bool,
terminated=False,
of_line=False):
button = self._buttons[button_position]
args = {'disabledforeground': 'black', 'state': 'disabled'}
if mover == 1:
args['text'] = '○'
args['state'] = 'disabled'
elif mover == -1:
args['text'] = '×'
args['state'] = 'disabled'
else:
args['text'] = ' '
if move_allowed:
args['state'] = 'normal'
elif not hasattr(self, '_agent'):
args['state'] = 'normal'
if terminated:
args['state'] = 'disabled'
if of_line:
if mover == 1:
args['disabledforeground'] = 'steelblue'
elif mover == -1:
args['disabledforeground'] = 'tomato'
button.config(**args)
return