def policy(self, state, turn, available, epsilon=0.08):
maxvalue = -99999
minvalue = 99999
action_list = []
if np.random.rand(1) < epsilon:
action_list = available
else:
if turn == 1:
for i in available:
state[i] = turn
state = encode(state)
if self.value[state] > maxvalue:
action_list = []
maxvalue = self.value[state]
action_list.append(i)
elif self.value[state] == maxvalue:
action_list.append(i)
state = decode(state)
state[i] = 0
else:
for i in available:
state[i] = turn
state = encode(state)
if self.value[state] < minvalue:
action_list = []
minvalue = self.value[state]
action_list.append(i)
elif self.value[state] == minvalue:
action_list.append(i)
state = decode(state)
state[i] = 0
return random.choice(action_list)
def policy(self, state, turn, epsilon=0.08):
maxvalue = -99999
minvalue = 99999
encoded_state = encode(state)
available = available_actions(state)
action_list = []
if np.random.rand(1) < epsilon:
action_list = available
else:
if turn == 1:
for action in available:
encoded = encoded_state + str(action)
if self.action_value[encoded] > maxvalue:
action_list = []
maxvalue = self.action_value[encoded]
action_list.append(action)
elif self.action_value[encoded] == maxvalue:
action_list.append(action)
else:
for action in available:
encoded = encoded_state + str(action)
if self.action_value[encoded] < minvalue:
action_list = []
minvalue = self.action_value[encoded]
action_list.append(action)
elif self.action_value[encoded] == minvalue:
action_list.append(action)
return random.choice(action_list)
def init_value(self):
state_list = itertools.product([0, 1, 2], repeat=9)
for state in state_list:
state = list(state)
encoded_state = encode(state)
available = available_actions(state)
for action in available:
encoded = encoded_state + str(action)
self.action_value[encoded] = 0
def init_value(self):
state_list = itertools.product([0, 1, 2], repeat=9)
for state in state_list:
state = list(state)
encoded = encode(state)
done, winner = is_finished(state)
if not done:
self._value[encoded] = random.uniform(-0.5, 0.5)
self._policy[encoded] = random.choice(available_actions(state))
# terminal state value
else:
self._value[encoded] = 0
def init_value(self):
state_list = itertools.product([0, 1, 2], repeat=9)
for state in state_list:
state = list(state)
done, winner = is_finished(state)
encoded = encode(state)
if not done:
self.value[encoded] = 0
elif winner == 1:
self.value[encoded] = 1
elif winner == 2:
self.value[encoded] = -1
else:
self.value[encoded] = 0
def assign_policy(self, state, x):
encoded = encode(state)
self._policy[encoded] = x
def policy(self, state):
encoded = encode(state)
return self._policy[encoded]
def assign_value(self, state, x):
encoded = encode(state)
self._value[encoded] = x
def value(self, state):
encoded = encode(state)
return self._value[encoded]
def update(agent, state, next_state, learning_rate=0.4):
state = encode(state)
next_state = encode(next_state)
agent.value[state] = agent.value[state] + learning_rate * (
agent.value[next_state] - agent.value[state])
def assign_q(self, state, action, x):
encoded = encode(state) + str(action)
self.action_value[encoded] = x
def q(self, state, action):
encoded = encode(state) + str(action)
return self.action_value[encoded]