def Play_Random_Test(model, env):
env.reset()
model.double()
model.load_state_dict(torch.load('model.pth'))
#AI plays 1
#You play -1
TD = TDAgent(model)
action_dict = dict()
player = -1
rolls = env.roll_dice()
fake_board = copy.deepcopy(env.board)
moves = env.all_possible_moves(player, fake_board, rolls)
for i, move in enumerate(moves):
action_dict[i] = move
#Alwats do 0
action = action_dict[0]
for a in action:
reward, done = env.step(a, env.board, player)
while True:
player = env.change_player(player)
rolls = env.roll_dice()
fake_board = copy.deepcopy(env.board)
actions = env.all_possible_moves(player, fake_board, rolls)
if actions != None:
fake = copy.deepcopy(env)
best_action, win_prob = TD.select_best_action(
actions, fake, player)
if len(best_action) != 0:
for a in best_action:
reward, done = env.step(a, env.board, player)
if done:
winner = reward
break
player = env.change_player(player)
rolls = env.roll_dice()
action_dict = dict()
fake_board = copy.deepcopy(env.board)
moves = env.all_possible_moves(player, fake_board, rolls)
if len(moves) != 0:
for i, move in enumerate(moves):
action_dict[i] = move
#print(action_dict)
#select = int(input('Select Action:'))
#Alwats do 0
#action = action_dict[select]
action = action_dict[0]
for a in action:
reward, done = env.step(a, env.board, player)
if done:
winner = reward
break
return reward
model.double()
model.load_state_dict(torch.load('model.pth'))
#AI plays 1
#You play -1
env.reset()
print(' 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17')
print('_________________________________________________________________________')
print(env.board)
print('_________________________________________________________________________')
print(' 18 19 20 21 22 23 24 25 26 27')
print('\n')
# time.sleep(2)
TD = TDAgent(model.double())
print('You play First')
action_dict = dict()
step = 1
player = -1
rolls = env.roll_dice()
fake_board = copy.deepcopy(env.board)
moves = env.all_possible_moves(player,fake_board,rolls)
for i,move in enumerate(moves):
action_dict[i] =move
# print(action_dict)
# print(moves)
def train(self, iters):
network = self
TD = TDAgent(network)
count = 0
gen_count_new = 0
gen_count_old = 0
for eps in range(iters):
env.reset()
self.init_eligiblity_trace()
rolls = env.roll_dice()
player = env.random_player()
env.player = player
count += 1
print("Calculating Weights: {:.5f}".format(count), end="\r")
step = 0
#Saving Model Every 10 steps
if count % 100 == 0:
torch.save(model.state_dict(), 'model.pth')
#See Who wins Every 100 steps
fake_1 = copy.deepcopy(env)
winner_random = Play_Random_Test(self, fake_1)
# if count%50 ==0:
# print('\n Random Agent Match Winner: ',winner_random)
# clear_output()
while True:
print('\t\t\t\t Working on Step: ', step, end="\r")
step += 1
features = env.board_features()
features = torch.tensor(features)
p = self.forward(features)
fake_board = copy.deepcopy(env.board)
actions = env.all_possible_moves(player, fake_board, rolls)
# print(actions)
if actions != None:
fake = copy.deepcopy(env)
action, win_prob = TD.select_best_action(
actions, fake, player)
if len(action) != 0:
for a in action:
reward, done = env.step(a, env.board, player)
features = env.board_features()
features = torch.tensor(features)
p_next = self.forward(features)
if done:
loss = self.update_weights(p, reward)
# print(loss)
break
else:
loss = self.update_weights(p, p_next)
player = env.change_player(player)
rolls = env.roll_dice()
def Play_Agent_VS_Old_Agent(env):
model = Network()
step = 0
old_model = Network()
model.double()
old_model.double()
print('Model_New_Main is 1 ')
print('Model_colab_old is -1')
model.load_state_dict(torch.load('model.pth'))
old_model.load_state_dict(torch.load('model_weak.pth'))
TD_new = TDAgent(model)
TD_old = TDAgent(old_model)
env.reset()
player = np.random.choice([-1, 1])
rolls = env.roll_dice()
while True:
step += 1
print('Player: ', player)
print('Rolls: ', rolls)
print(
' 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17'
)
print(
'_________________________________________________________________________'
)
print(env.board)
print(
'_________________________________________________________________________'
)
print(' 18 19 20 21 22 23 24 25 26 27')
print('\n')
fake_board = copy.deepcopy(env.board)
actions = env.all_possible_moves(player, fake_board, rolls)
if actions != None:
fake = copy.deepcopy(env)
if player == 1:
best_action, win_prob = TD_new.select_best_action(
actions, fake, player)
elif player == -1:
best_action, win_prob = TD_old.select_best_action(
actions, fake, player)
if len(best_action) != 0:
for a in best_action:
reward, done = env.step(a, env.board, player)
if done:
print('Won in %s moves!' % (step))
winner = reward
break
player = change_player(player)
rolls = env.roll_dice()
return winner