def training(self, episodes):
self.env.set_speed_mode(self.env.my_car_id, 0)
state = None
steps = 0
# reward_type = "collision"
# reward_type = "horizon"
reward_type = "security_distance"
speed_limit = True
plt_data = {
"collisions": [],
"space_headway": [],
"relative_speed": [],
"speed": [],
"steps": 0
}
while True:
print(state)
if state:
plt_data["space_headway"].append(state.get("space_headway"))
plt_data["relative_speed"].append(
round(state.get("relative_speed") * 3.6, 0))
plt_data["speed"].append(round(state.get("speed") * 3.6, 0))
d_t, ds_t, s_t = \
self.framing(state.get('space_headway'), self.i_dict_space_headway), \
self.framing(state.get('relative_speed'), self.i_dict_relative_speed), \
self.framing(state.get('speed'), self.i_dict_speed)
a = self.e_greedy_policy(d_t, ds_t, s_t)
q_t = self.q[self.i_dict_space_headway.get(d_t),
self.i_dict_relative_speed.get(ds_t),
self.i_dict_speed.get(s_t),
self.i_dict_action.get(self.action[a])]
new_speed = self.new_speed(self.action[a], state.get('speed'))
self.env.set_speed(self.env.my_car_id, new_speed)
self.env.simulation_step()
next_state = self.env.get_state(self.env.my_car_id)
q_max_t1 = None
if self.env.is_collided(self.env.my_car_id):
self.set_reward_collision(reward_type)
self.env.set_speed(self.env.my_car_id, 0)
q_max_t1 = 0
state = None
plt_data["collisions"].append(steps)
elif next_state:
"""REWARD"""
"""
if reward_type == "horizon":
self.set_reward_horizon_speed(next_state.get('space_headway'), next_state.get('speed'), speed_limit)
"""
if reward_type == "security_distance":
self.set_reward_security_dist_speed(
next_state.get('space_headway'),
next_state.get('speed'), speed_limit)
print(f"reward {self.reward}")
d_t1, ds_t1, s_t1 = \
self.framing(next_state.get('space_headway'), self.i_dict_space_headway), \
self.framing(next_state.get('relative_speed'), self.i_dict_relative_speed), \
self.framing(next_state.get('speed'), self.i_dict_speed)
q_max_t1 = np.max(
self.q[self.i_dict_space_headway.get(d_t1),
self.i_dict_relative_speed.get(ds_t1),
self.i_dict_speed.get(s_t1)])
state = next_state
if q_max_t1 is not None:
self.q[
self.i_dict_space_headway.get(d_t),
self.i_dict_relative_speed.get(ds_t),
self.i_dict_speed.get(s_t),
self.i_dict_action.get(self.action[a])] = \
(1 - self.alpha) * q_t + self.alpha * (self.reward + self.gamma * q_max_t1)
""" PRINT Q"""
print(
f"q: {self.q[self.i_dict_space_headway.get(d_t), self.i_dict_relative_speed.get(ds_t), self.i_dict_speed.get(s_t)]}"
)
steps += 1
self.epsilon_decay(steps)
# print(steps)
# print(f"time: {self.env.get_current_time()}")
else:
self.env.simulation_step()
state = self.env.get_state(self.env.my_car_id)
self.env.set_speed(self.env.my_car_id, 0)
if steps > (episodes * 10000):
time.sleep(.1)
if steps == episodes * 10000:
plt_data["steps"] = steps
plotting = Plotting(self, plt_data)
plotting.plot_()
