def predict(self,state): ghost1_state_number=utils.state_2_number(state[2:4]) ghost2_state_number=utils.state_2_number(state[4:6]) ghost1_next=[] ghost1_next_probs=[] ghost2_next=[] ghost2_next_probs=[] for index,x in enumerate(self.model[ghost1_state_number]): if x>0: ghost1_next.append(index) ghost1_next_probs.append(x) for index,x in enumerate(self.model[ghost2_state_number]): if x>0: ghost2_next.append(index) ghost2_next_probs.append(x) li=[] li_probs=[] for index1,g1 in enumerate(ghost1_next): for index2,g2 in enumerate(ghost2_next): g1_state=utils.number_2_state(g1) g2_state=utils.number_2_state(g2) li.append(g1_state+g2_state) li_probs.append(ghost1_next_probs[index1]*ghost2_next_probs[index2]) return li,li_probs
def run(run_ID):
states,next_states=load_synthetic_data(5*49)
ghost_tabular_array=numpy.zeros(49*49).reshape((49,49))
for x,y in zip(states,next_states):
numberS=utils.state_2_number(x)
numberSp=utils.state_2_number(y)
ghost_tabular_array[numberS,numberSp]=ghost_tabular_array[numberS,numberSp]+1
for index in range(ghost_tabular_array.shape[0]):
ghost_tabular_array[index,:]=ghost_tabular_array[index,:]/(numpy.sum(ghost_tabular_array[index,:]))
numpy.savetxt("learn_tabular_models/tabular"+str(run_ID)+".h5",ghost_tabular_array)
for y in range(7):
s = [x, y]
for _ in range(each):
while True:
case = numpy.random.randint(0, 4)
if case == 0:
s_p = [x + 1, y]
elif case == 1:
s_p = [x, y + 1]
elif case == 2:
s_p = [x - 1, y]
elif case == 3:
s_p = [x, y - 1]
if numpy.min(s_p) >= 0 and numpy.max(s_p) <= 6:
break
li_s.append(s)
li_sprime.append(s_p)
return li_s, li_sprime
states, next_states = load_synthetic_data(5 * 49)
ghost_tabular_array = numpy.zeros(49 * 49).reshape((49, 49))
for x, y in zip(states, next_states):
numberS = utils.state_2_number(x)
numberSp = utils.state_2_number(y)
ghost_tabular_array[numberS,
numberSp] = ghost_tabular_array[numberS, numberSp] + 1
for index in range(ghost_tabular_array.shape[0]):
ghost_tabular_array[index, :] = ghost_tabular_array[index, :] / (numpy.sum(
ghost_tabular_array[index, :]))
numpy.savetxt("tabular.h5", ghost_tabular_array)
def predict(self, state, action):
action_array = numpy.array(4 * [0]).reshape(1, 4)
action_array[0, action] = 1
ghost_state = state[2:]
ghost_state_array = numpy.array(ghost_state).reshape(
1, len(ghost_state))
pacman_state = state[0:2]
pacman_state_array = numpy.array(pacman_state).reshape(
1, len(pacman_state))
li_ghosts = self.em_model_object.predict(ghost_state_array)
li_next_states = []
li_rewards = []
li_dones = []
if self.type == 'stochastic':
pacman_next_state = self.other_models_object.pacman_model.predict(
[pacman_state_array, action_array])
for index, gh in enumerate(li_ghosts):
next_state = numpy.concatenate((pacman_next_state, gh), axis=1)
reward = self.other_models_object.reward_model.predict(
next_state)
done = self.other_models_object.done_model.predict(next_state)
li_next_states.append(next_state[0].tolist())
li_rewards.append(reward[0, 0])
if done > 0.5:
li_dones.append(True)
else:
li_dones.append(False)
return li_next_states, li_rewards, li_dones
elif self.type == 'deterministic':
pacman_next_state = self.other_models_object.pacman_model.predict(
[pacman_state_array, action_array])
ghost_next_state = self.other_models_object.ghosts_model.predict(
ghost_state_array)
next_state = numpy.concatenate(
(pacman_next_state, ghost_next_state), axis=1)
reward = self.other_models_object.reward_model.predict(next_state)
done = self.other_models_object.done_model.predict(next_state)
li_next_states.append(next_state[0].tolist())
li_rewards.append(reward[0, 0])
if done > 0.5:
li_dones.append(True)
else:
li_dones.append(False)
return li_next_states, li_rewards, li_dones
elif self.type == 'tabular':
state_number = utils.state_2_number(ghost_state)
li_next_states = []
for j in range(49 * 49):
if self.other_models_object.ghosts_tabular_model[state_number,
j] > 0:
pacman_next_state = self.other_models_object.pacman_model.predict(
[pacman_state_array, action_array])
ghost_next_state = numpy.array(
utils.number_2_state(j)).reshape(1, 4)
next_state = numpy.concatenate(
(pacman_next_state, ghost_next_state), axis=1)
reward = self.other_models_object.reward_model.predict(
next_state)
done = self.other_models_object.done_model.predict(
next_state)
li_next_states.append(next_state[0].tolist())
li_rewards.append(reward[0, 0])
if done > 0.5:
li_dones.append(True)
else:
li_dones.append(False)
#print("hihoo!")
return li_next_states, li_rewards, li_dones
elif self.type == 'random':
'planner is random'
sys.exit(1)