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 minimax(state, mover: int, t: TTT) -> [Score, Move]: next_mover = -1 if mover is 1 else 1 possible_moves = t.get_available_positions(state) corresponding_scores = [] best_score = 0 best_move = None for index in possible_moves: next_state = state.copy() next_state[index] = mover if t.is_terminated(next_state): score = t.get_score(next_state) corresponding_scores.append(score) else: [score, _] = minimax(next_state, next_mover, t) corresponding_scores.append(score) if mover == 1: best_score = max(corresponding_scores) best_move_index = corresponding_scores.index(best_score) best_move = possible_moves[best_move_index] elif mover == -1: best_score = min(corresponding_scores) best_move_index = corresponding_scores.index(best_score) best_move = possible_moves[best_move_index] return [best_score, best_move]
def minimax_save(state, mover: int, t: TTT, table) -> (Score, Move): encoded_state = encode_state(state) if encode_state in table: return table[encoded_state] next_mover = -1 if mover is 1 else 1 possible_moves = t.get_available_positions(state) corresponding_scores = [] best_score = 0 best_move = None for index in possible_moves: next_state = state.copy() next_state[index] = mover if t.is_terminated(next_state): score = t.get_score(next_state) corresponding_scores.append(score) else: [score, _] = minimax_save(next_state, next_mover, t, table) corresponding_scores.append(score) if mover == 1: best_score = max(corresponding_scores) best_move_index = corresponding_scores.index(best_score) best_move = possible_moves[best_move_index] elif mover == -1: best_score = min(corresponding_scores) best_move_index = corresponding_scores.index(best_score) best_move = possible_moves[best_move_index] table[encoded_state] = (best_score, best_move) return (best_score, best_move)
def test_score1(self): t3 = TTT(3) s = [[1, -1, 0], [-1, 1, 0], [0, 0, 1]] s = np.array(s).reshape(-1) terminated = t3.is_terminated(s) score = t3.get_score(s) self.assertTrue(terminated) self.assertEqual(score, 5)
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_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)