def train_ep_p2(net, policy, render=False):
state = env.reset()
ep_reward = 0
ctr = 0
if render:
env.render()
while True:
# env.render()
if ctr > 0: # skip player1's first turn so that he goes second
action = choose_action('self', 1, state, net, 0.05)
next_state, reward, done, info = env.step(action)
if render:
print('Player 1 moves', action)
env.render()
else:
next_state = state
reward = 0
done = False
info = {'next_player': 2}
# Let player 2 play
p2_reward = 0
while info['next_player'] == 2 and not done:
action2 = choose_action(policy, 2, next_state, net, 0.05)
next_state, reward2, done, info = env.step(action2)
if render:
print('Player 2 moves', action2)
env.render()
p2_reward += reward2
if ctr == 0:
ctr += 1
continue
net.store_transition(state, action, reward - p2_reward, next_state)
ep_reward += reward
net.learn()
if done:
ctr = 0
break
ctr += 1
state = next_state
if ep_reward > 49:
win = 1
else:
win = 0
return ep_reward, win
def test_ep_pvp(net, net2, num_test, eps=0.05, render=False):
test_reward = []
test_reward_p2 = []
test_win = 0
draw = 0
for i in range(num_test):
state = env.reset()
ep_reward = 0
p2_reward = 0
ctr = 0
if render:
env.render()
while True:
# env.render()
action = choose_action('self', 1, state, net, eps)
next_state, reward, done, info = env.step(action)
if render:
print('Player 1 moves', action)
env.render()
while info['next_player'] == 2 and not done:
action2 = choose_action('self', 2, next_state, net2, eps)
next_state, reward2, done, info = env.step(action2)
if render:
print('Player 2 moves', action2)
env.render()
p2_reward += reward2
# state = next_state
ep_reward += reward
if done:
break
state = next_state
ctr += 1
test_reward_p2.append(p2_reward)
test_reward.append(ep_reward)
if ep_reward > 49:
test_win += 1
if ep_reward == 49:
draw += 1
return np.mean(test_reward), np.mean(
test_reward_p2), test_win / num_test, draw / num_test
def test_ep_p2(net, policy, num_test, eps=0.05, render=False):
test_reward = []
test_win = 0
for i in range(num_test):
state = env.reset()
ep_reward = 0
ctr = 0
if render:
env.render()
while True:
# env.render()
if ctr > 0: # skip player1's first turn so that he goes second
action = choose_action('self', 1, state, net, eps)
next_state, reward, done, info = env.step(action)
if render:
print('Player 1 moves', action)
env.render()
else:
next_state = state
reward = 0
done = False
info = {'next_player': 2}
p2_reward = 0
while info['next_player'] == 2 and not done:
action2 = choose_action(policy, 2, next_state, net, eps)
next_state, reward2, done, info = env.step(action2)
if render:
print('Player 2 moves', action2)
env.render()
p2_reward += reward2
# state = next_state
ep_reward += reward
if done:
break
state = next_state
ctr += 1
test_reward.append(ep_reward)
if ep_reward > 49:
test_win += 1
return np.mean(test_reward), test_win / num_test
def train_ep(net, policy, render=False):
state = env.reset()
ep_reward = 0
ctr = 0
if render:
env.render()
while True:
# env.render()
action = choose_action('self', 1, state, net, 0.05)
next_state, reward, done, info = env.step(action)
if render:
print('Player 1 moves', action)
env.render()
# Let player 2 play
p2_reward = 0
while info['next_player'] == 2 and not done:
action2 = choose_action(policy, 2, next_state, net, 0.05)
next_state, reward2, done, info = env.step(action2)
if render:
print('Player 2 moves', action2)
env.render()
p2_reward += reward2
net.store_transition(state, action, reward - p2_reward, next_state)
ep_reward += reward
net.learn()
if done:
ctr = 0
break
ctr += 1
state = next_state
if ep_reward > 49:
win = 1
else:
win = 0
return ep_reward, win