def run_one_test_episode(self, env) -> dict:
state = env.reset()
p_actual, p_reference = state
state = np.array(state)
eps = self.min_eps
action_index = self.agent.get_action(state, eps)
action = self.actions[action_index]
output = {'p_actual': [p_actual], 'p_reference': [p_reference], 'action': [action]}
for i in range(self.max_n_steps):
next_state, reward, done, _ = env.step(action)
p_actual, p_reference = next_state
output['p_actual'].append(p_actual)
output['p_reference'].append(p_reference)
output['action'].append(action)
state = np.array(next_state)
action_index = self.agent.get_action(state, eps)
action = self.actions[action_index]
if done:
break
mse = calc_mse(output['p_actual'], output['p_reference'])
output['mse'] = mse
return output
def run_one_train_episode(self, env) -> dict:
state = env.reset()
p_actual, p_reference = state
state = np.array(state)
eps = epsilon_annealing(self.episode_counter, self.eps_decrease_last_episode, self.min_eps)
action_index = self.agent.get_action(state, eps)
action = self.actions[action_index]
output = {'p_actual': [p_actual], 'p_reference': [p_reference], 'action': [action]}
for i in range(self.max_n_steps):
next_state, reward, done, _ = env.step(action)
p_actual, p_reference = next_state
output['p_actual'].append(p_actual)
output['p_reference'].append(p_reference)
output['action'].append(action)
next_state = np.array(next_state)
self.replay_memory.push(state, action_index, reward, next_state, done)
if len(self.replay_memory) > self.batch_size:
minibatch = self.replay_memory.pop(self.batch_size)
train_helper(self.agent, minibatch, self.gamma)
state = next_state
action_index = self.agent.get_action(state, eps)
action = self.actions[action_index]
if done:
break
mse = calc_mse(output['p_actual'], output['p_reference'])
output['mse'] = mse
self.episode_counter += 1
return output
def run_one_test_episode(self, env) -> dict:
state = env.reset()
delta = 0
p_actual, p_reference = state
action = self.agent.use([*state, delta])
output = {
'p_actual': [p_actual],
'p_reference': [p_reference],
'action': [action]
}
for i in range(self.max_n_steps):
next_state, reward, done, _ = env.step(action)
next_delta = next_state[0] - state[0]
p_actual, p_reference = next_state
output['p_actual'].append(p_actual)
output['p_reference'].append(p_reference)
output['action'].append(action)
action = self.agent.use([*next_state, next_delta])
state = next_state
if done:
break
mse = calc_mse(output['p_actual'], output['p_reference'])
output['mse'] = mse
return output
def run_one_train_episode(self, env) -> dict:
state = env.reset()
p_actual, p_reference = state
output = {
'p_actual': [p_actual],
'p_reference': [p_reference],
'action': [self.k]
}
for i in range(self.max_steps):
obs, rew, done, _ = env.step(self.k)
p_actual, p_reference = obs
output['p_actual'].append(p_actual)
output['p_reference'].append(p_reference)
output['action'].append(self.k)
if done:
break
mse = calc_mse(output['p_actual'], output['p_reference'])
output['mse'] = mse
return output
def run_one_train_episode(self, env) -> dict:
state = env.reset()
p_actual, p_reference = state
action = self.agent.use(state)
output = {
'p_actual': [p_actual],
'p_reference': [p_reference],
'action': [action]
}
for i in range(self.max_n_steps):
next_state, reward, done, _ = env.step(action)
p_actual, p_reference = next_state
output['p_actual'].append(p_actual)
output['p_reference'].append(p_reference)
output['action'].append(action)
action = self.agent.learn(state, action, reward, next_state)
state = next_state
if done:
break
mse = calc_mse(output['p_actual'], output['p_reference'])
output['mse'] = mse
return output