def main(rounds):
global ticTacToe
global wins
global losses
initial_q_value = 0.0
alpha = 0.5 # Step size
gamma = 1.0 # Discount factor
epsilon = 0.2 # Exploration rate
actions = [(i, j) for i in range(3) for j in range(3)]
actions_per_state = {a: initial_q_value for a in actions}
Q_values = {'terminal': actions_per_state.copy()}
Q_values['terminal'][None] = 0.0
first_turn_random_count = 0
# Running Q-Learning Q-value updates for many episodes
for i in range(rounds):
if i % 10000 == 0:
print("Rounds done = {}".format(i), end=" | ")
print("Wins = {}".format(wins), end=" | ")
print("Losses = {}".format(losses))
wins = 0
losses = 0
ticTacToe = TicTacToe()
first_turn = random.choice(['random', 'computer'])
if first_turn == 'random':
first_turn_random_count += 1
# Random player playing one turn
# Q opposition player playing one turn
selected_grid = epsilon_greedy_for_opposition(
Q_values, ticTacToe.get_current_state(), actions_per_state,
epsilon)
if selected_grid not in ticTacToe.get_empty_cells():
selected_grid = random.choice(ticTacToe.get_empty_cells())
ticTacToe.set_one_grid(selected_grid[0], selected_grid[1])
ticTacToe.toggle_turn()
Q_values = Q_Learning(Q_values, alpha, gamma, epsilon,
actions_per_state)
# pprint(Q_values)
policy = find_optimal_policy(Q_values)
print("First turn by random players = {}%".format(first_turn_random_count *
100 / rounds))
filename = "Q_values_{}_episodes_025_epsilon.p".format(rounds)
pickle.dump(Q_values, open(filename, "wb"))
filename = "policy_{}_episodes_025_epsilon.p".format(rounds)
pickle.dump(policy, open(filename, "wb"))
def main(rounds):
global ticTacToe
wins, losses = 0, 0
first_turn_random_count = 0
# Running Q-Learning Q-value updates for many episodes
for i in range(rounds):
if i % 10000 == 0:
print("Rounds done = {}".format(i), end=" | ")
print("Wins = {}".format(wins), end=" | ")
print("Losses = {}".format(losses))
wins = 0
losses = 0
ticTacToe = TicTacToe()
current_state = None
first_turn = random.choice(['random', 'computer'])
if first_turn == 'random':
first_turn_random_count += 1
# Random player playing one turn
selected_grid = random.choice(ticTacToe.get_empty_cells())
ticTacToe.set_one_grid(selected_grid[0], selected_grid[1])
ticTacToe.toggle_turn()
while current_state != 'terminal':
current_state = ticTacToe.get_current_state()
try:
selected_grid = policy[current_state]
if selected_grid not in ticTacToe.get_empty_cells():
selected_grid = random.choice(ticTacToe.get_empty_cells())
except:
selected_grid = random.choice(ticTacToe.get_empty_cells())
ticTacToe.set_one_grid(selected_grid[0], selected_grid[1])
solved, result = ticTacToe.is_solved()
if solved:
if result != 0:
wins += 1
break
ticTacToe.toggle_turn()
selected_grid = random.choice(ticTacToe.get_empty_cells())
ticTacToe.set_one_grid(selected_grid[0], selected_grid[1])
solved, result = ticTacToe.is_solved()
if solved:
if result != 0:
losses += 1
break
ticTacToe.toggle_turn()
print("First turn by random players = {}%".format(first_turn_random_count *
100 / rounds))