agent_2 = A2C_agent(input_dim=3, lam=0.8, gamma=0.99, lr=1e-4)
param = torch.load('a2c_2_param2.pt')
agent_2.model.load_state_dict(param)
while (1):
for event in pygame.event.get():
# 閉じるボタンが押されたら終了
if event.type == QUIT:
pygame.display.quit()
# Pygameの終了(画面閉じられる)
pygame.quit()
sys.exit()
# black 1 white 2
board, changeable_Pos, Position_Row, Position_Col, Change_Position, done = othello.make(
)
game.updateBoard(board)
for _ in range(100):
#setrow, setcol = agent_1.take_action(board, changeable_Pos, Position_Row, Position_Col, Change_Position)
setrow, setcol = agent_1.take_determ_action(
board, changeable_Pos, Position_Row, Position_Col,
Change_Position)
board, changeable_Pos, Position_Row, Position_Col, Change_Position, done = othello.step(
setrow, setcol)
game.updateBoard(board)
if done:
time.sleep(0.5)
break
def get_play_data(agent_1, agent_2):
othello = Othello()
first_states = deque()
first_rewards = deque()
first_actions = deque()
second_states = deque()
second_rewards = deque()
second_actions = deque()
board, changeable_Pos, Position_Row, Position_Col, Change_Position, done = othello.make(
)
while not done:
if othello.color == 1:
state = getState(board, changeable_Pos, Position_Row, Position_Col,
Change_Position)
first_states.appendleft(state)
reward_1, reward_2 = getReward(board)
first_rewards.appendleft(reward_1)
setrow, setcol = agent_1.take_action(board, changeable_Pos,
Position_Row, Position_Col,
Change_Position)
board, changeable_Pos, Position_Row, Position_Col, Change_Position, done = othello.step(
setrow, setcol)
first_actions.appendleft(8 * setrow + setcol)
else:
state = getState(board, changeable_Pos, Position_Row, Position_Col,
Change_Position)
second_states.appendleft(state)
reward_1, reward_2 = getReward(board)
second_rewards.appendleft(reward_2)
setrow, setcol = agent_2.take_action(board, changeable_Pos,
Position_Row, Position_Col,
Change_Position)
board, changeable_Pos, Position_Row, Position_Col, Change_Position, done = othello.step(
setrow, setcol)
second_actions.appendleft(8 * setrow + setcol)
state = getState(board, changeable_Pos, Position_Row, Position_Col,
Change_Position)
reward_1, reward_2 = getReward(board)
first_states.appendleft(state)
second_states.appendleft(state)
first_states = torch.FloatTensor(first_states)
second_states = torch.FloatTensor(second_states)
first_actions = torch.FloatTensor(first_actions)
second_actions = torch.FloatTensor(second_actions)
first_rewards.appendleft(reward_1)
second_rewards.appendleft(reward_2)
first_rewards = torch.FloatTensor(list(first_rewards)[:-1])
second_rewards = torch.FloatTensor(list(second_rewards)[:-1])
discount_rate = 1
if reward_1 > reward_2:
first_values = torch.FloatTensor(
[discount_rate**i for i in range(len(first_rewards))])
second_values = torch.FloatTensor(
[-discount_rate**i for i in range(len(second_rewards))])
elif reward_1 < reward_2:
first_values = torch.FloatTensor(
[-discount_rate**i for i in range(len(first_rewards))])
second_values = torch.FloatTensor(
[discount_rate**i for i in range(len(second_rewards))])
else:
first_values = torch.FloatTensor(
[0 for i in range(len(first_rewards))])
second_values = torch.FloatTensor(
[0 for i in range(len(second_rewards))])
data_first = {
'states': first_states,
'rewards': first_rewards,
'actions': first_actions,
'values': first_values,
}
data_secound = {
'states': second_states,
'rewards': second_rewards,
'actions': second_actions,
'values': second_values,
}
return data_first, data_secound
