def runEpisode(agent, environment, discount, decision, display, message, pause, episode, testFlag=0, foldN=0):
returns = 0
totalDiscount = 1.0
environment.reset(testFlag, foldN)
if 'startEpisode' in dir(agent): agent.startEpisode()
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
# display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
return returns, reward, AOP, explore # returns: accumulative rewards, reward:
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward, AOP, explore = environment.doAction(action)
if testFlag == 0: # training episode
# UPDATE LEARNER
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action, nextState, reward)
else: # testing episode
pass
returns += reward * totalDiscount
totalDiscount *= discount
def model(self, parameters):
"""Model a year of data for a location.
Args:
parameters: (tuple) capacity (Wh), PV Size (STC).
Returns:
(domain): results from model.
"""
env.reset()
size, pv = parameters
# don't go below 1 negative/division by zero issues
pv = max(pv, 1.)
size = max(size, 1.)
plane = InclinedPlane(Site(self.place), self.tilt, self.azimuth)
load = self.load()
SHS = Gateway([load,
self.cc([SimplePV(pv, plane)]),
IdealStorage(size)])
for r in eere.EPWdata('418830'):
env.update_time(r['datetime'])
SHS()
print SHS.details()
self.foo.write('%s,%s,%s\n' % (size, pv, SHS.merit()))
self.foo.flush()
return SHS
def runEpisode(agent,
environment,
discount,
decision,
display,
message,
pause,
episode,
train=False):
returns = 0
totalDiscount = 1.0
environment.reset()
if 'startEpisode' in dir(agent): agent.startEpisode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
state_copy = copy.deepcopy(state)
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: TOOK " +
str(len(state.history)) + "STEPS; RETURN WAS " +
str(returns) + "\n")
message(str(state.history))
message(str(len(state.history)))
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward = environment.doAction(action)
message("Started in state: " + str(state_copy.x) + str(state_copy.y) +
str(state_copy.direction) + str(state_copy.num_explored) +
" " + str(len(state_copy.history)) + "\nTook action: " +
str(action) + "\nEnded in state: " + str(nextState.x) +
str(nextState.y) + str(nextState.direction) +
str(nextState.num_explored) + " " +
str(len(nextState.history)) + "\nGot reward: " + str(reward) +
"\n")
# UPDATE LEARNER
# print_grid(state_copy.explored_grid)
# print_grid(nextState.explored_grid)
if 'observeTransition' in dir(
agent) and not nextState.is_terminal and train:
agent.observeTransition(state_copy, action, nextState, reward)
print(agent.weights)
print("############################################")
print("############################################\n")
returns += reward * totalDiscount
totalDiscount *= discount
def runEpisode(agent, environment, discount, decision, display, message, pause,
episode):
returns = 0
totalDiscount = 1.0
environment.reset()
message("BEGINNING EPISODE: " + str(episode) + "\n")
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward = environment.doAction(action)
message("Started in state: " + str(state) + "\nTook action: " +
str(action) + "\nEnded in state: " + str(nextState) +
"\nGot reward: " + str(reward) + "\n")
# UPDATE LEARNER
agent.update(state, action, nextState, reward)
returns += reward * totalDiscount
totalDiscount *= discount
def runEpisode(agent, qtype, environment, discount, decision, display, message, pause, episode):
###########################
# GET THE GRIDWORLD
###########################
returns = 0
totalDiscount = 1.0
environment.reset()
#for state in mdp.getStates():
# display(state)
if 'startEpisode' in dir(agent): agent.startEpisode()
message("BEGINNING EPISODE: "+str(episode)+"\n")
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE "+str(episode)+" COMPLETE: RETURN WAS "+str(returns)+"\n")
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward, action2 = environment.doAction(action, qtype)
message("Started in state: "+str(state)+
"\nSpecified action: "+str(action)+
"\nTook action: "+str(action2)+
"\nEnded in state: "+str(nextState)+
"\nGot reward: "+str(reward)+"\n")
# UPDATE LEARNER
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action2, nextState, reward, environment)
if isinstance(reward, tuple):
reward=sum(reward)
returns += reward * totalDiscount
totalDiscount *= discount
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
def runEpisode(agent, environment, discount, decision, display, message, pause,
episode):
returns = 0
totalDiscount = 1.0
environment.reset(agent)
if 'startEpisode' in dir(agent): agent.startEpisode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
timestep = 0
MAX_TIMESTEPS = 20
while True:
if timestep >= MAX_TIMESTEPS:
return returns
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
display(state)
#pause()
if timestep == 0 and episode == 1:
input("")
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward = environment.doAction(action)
#message("Started in state: "+str(state)+
# "\nTook action: "+str(action)+
# "\nEnded in state: "+str(nextState)+
# "\nGot reward: "+str(reward)+"\n")
# UPDATE LEARNER
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action, nextState, reward)
returns += reward * totalDiscount
totalDiscount *= discount
timestep += 1
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
def runEpisode(agent, environment, discount, decision, display, message, pause,
episode):
returns = 0
totalDiscount = 1.0
environment.reset()
if 'startEpisode' in dir(agent): agent.startEpisode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
stateAndValues = -1
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
return (returns, stateAndValues)
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward = environment.doAction(action)
message("Started in state: " + str(state) + "\nTook action: " +
str(action) + "\nEnded in state: " + str(nextState) +
"\nGot reward: " + str(reward) + "\n")
# UPDATE LEARNER
val = -1
if 'observeTransition' in dir(agent):
val = agent.observeTransition(state, action, nextState, reward)
if state == (2, 2):
stateAndValues = [0, val, 0, 0]
returns += reward * totalDiscount
totalDiscount *= discount
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
def runEpisode(agent, environment, discount, decision, display, message, pause, episode):
returns = 0
totalDiscount = 1.0
environment.reset()
if "startEpisode" in dir(agent):
agent.startEpisode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " + str(returns) + "\n")
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise "Error: Agent returned None action"
# EXECUTE ACTION
nextState, reward = environment.doAction(action)
message(
"Started in state: "
+ str(state)
+ "\nTook action: "
+ str(action)
+ "\nEnded in state: "
+ str(nextState)
+ "\nGot reward: "
+ str(reward)
+ "\n"
)
# UPDATE LEARNER
if "observeTransition" in dir(agent):
agent.observeTransition(state, action, nextState, reward)
returns += reward * totalDiscount
totalDiscount *= discount
if "stopEpisode" in dir(agent):
agent.stopEpisode()
def analizar_texto():
response = txt_consultas.get("1.0", "end")
salida_lexico_ast = analizarLex(response)
texto = analizarSin(response)
txt_salida.insert('end', '\n>>>\n')
txt_salida.insert('end', '\n=====SALIDA C3D======\n')
txt_salida.insert('end', texto + '\n\n\n')
txt_salida.insert('end', tab_simbolos())
txt_salida.insert('end', '\n=====REPORTE DE INDEX======')
txt_salida.insert('end', tab_string())
txt_salida.insert('end', '\n=====REPORTE DE FUNCIONES======')
txt_salida.insert('end', tab_func())
txt_salida.insert('end', '\n=====REPORTE DE ERRORES======')
txt_salida.insert('end', get_errores())
txt_salida.insert('end', '\n>>>\n')
reset()
def runEpisode(agent, environment, discount, decision, display, message, pause,
episode):
returns = 0
totalDiscount = 1.0
environment.reset()
if "startEpisode" in dir(agent):
agent.startEpisode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
while True:
# Display current state
state = environment.getCurrentState()
display(state)
pause()
# End if in a terminal state
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
return returns
# Get action (usually from agent)
action = decision(state)
if action == None:
raise Exception("Error: Agent returned None action")
# Execute action
nextState, reward = environment.doAction(action)
message("Started in state: " + str(state) + "\nTook action: " +
str(action) + "\nEnded in state: " + str(nextState) +
"\nGot reward: " + str(reward) + "\n")
# Update learner
if "observeTransition" in dir(agent):
agent.observeTransition(state, action, nextState, reward)
returns += reward * totalDiscount
totalDiscount *= discount
if "stopEpisode" in dir(agent):
agent.stopEpisode()
def run_episode(agent, environment, discount, decision, display, message,
pause, episode):
returns = 0
total_discount = 1.0
environment.reset()
if 'start_episode' in dir(agent):
agent.start_episode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
while True:
# DISPLAY CURRENT STATE
state = environment.get_current_state()
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.get_possible_actions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise Exception('Error: Agent returned None action')
# EXECUTE ACTION
next_state, reward = environment.do_action(action)
message("Started in state: " + str(state) + "\nTook action: " +
str(action) + "\nEnded in state: " + str(next_state) +
"\nGot reward: " + str(reward) + "\n")
# UPDATE LEARNER
if 'observe_transition' in dir(agent):
agent.observe_transition(state, action, next_state, reward)
returns += reward * total_discount
total_discount *= discount
if 'stop_episode' in dir(agent):
agent.stop_episode()
def runEpisode(agent, environment, discount, decision, display, message, pause, episode, rates, offset, pmNum, locks,pllock , q,plinfo):
returns = 0
totalDiscount = 1.0
environment.reset()
if 'startEpisode' in dir(agent): agent.startEpisode()
#message("BEGINNING EPISODE: "+str(episode)+"\n")
pm = environment.baeplatform.pm
while True:
# END IF IN A TERMINAL STATE
if environment.isTerminal():
message("EPISODE "+str(episode)+" COMPLETE: RETURN WAS "+str(returns)+"\n")
return returns, offset
#print rates[offset%len(rates)]
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
pause()
offset += 1
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward = environment.doAction(state, action, pmNum,locks,pllock,q,plinfo)
message("VM " + str(pm.id) + " Started in state: "+str(state)+
"\nTook action: "+str(action)+
"\nEnded in state: "+str(nextState)+
"\nGot reward: "+str(reward)+"\n")
# UPDATE LEARNER
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action, nextState, reward)
returns += reward * totalDiscount
totalDiscount *= discount
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
def watch_play(self):
done = False
board = env.reset()
while not done:
# finds the best action
action = env.process_state(board)
self.drop_piece(action, board)
board, done = env.step(board, *action)
self.board = board.area
self.update()
self.root.update()
time.sleep(self.speed)
def runEpisode(agent, environment, discount, decision, display, message, pause, episode):
returns = 0
totalDiscount = 1.0
environment.reset()
if 'startEpisode' in dir(agent): agent.startEpisode()
message("BEGINNING EPISODE: "+str(episode)+"\n")
f = open("trajectories.txt", "a")
appendString='['
steps=0
grid_width=10
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
display(state)
pause()
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE "+str(episode)+" COMPLETE: RETURN WAS "+str(returns)+"\n")
return returns
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState=state
if action != 'stay':
nextState, reward = environment.doAction(action)
message("Started in state: "+str(state)+
"\nTook action: "+str(action)+
"\nEnded in state: "+str(nextState)+
"\nGot reward: "+str(reward)+"\n")
# UPDATE LEARNER
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action, nextState, reward)
returns += reward * totalDiscount
totalDiscount *= discount
stateNumber= (state[1]*grid_width +state[0])
appendString+=str(stateNumber)
steps+=1
if(steps==stepsLimit):
break;
appendString+=','
appendString+='],'
print("AppendString ",appendString)
f.write(appendString+"\n")
f.close()
runEpisode(agent, environment, discount, decision, display, message, pause, episode);
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
numActions = 10
Actions = np.linspace(0, 1, numActions)
# This is our learning agent
gamma = .95
agent = sarsaAgent(4, numActions, 10, 1, epsilon=5e-2, gamma=gamma, alpha=1e-2)
maxSteps = 1e5
R = []
UpTime = []
step = 0
ep = 0
while step < maxSteps:
ep += 1
x = environment.reset() # initialize the state
C = 0.
done = False
t = 1
while not done:
t += 1
step += 1
a = agent.action(x)
u = Actions[a]
#env.render() # only for visual effects
x_next, c, done = environment.step(u, x)
C += (1. / t) * (c - C)
agent.update(x, a, c, x_next, done)
x = x_next
def handler(signum, frane):
"""Handles CTRL-C to terminate the session early."""
global terminate_early
terminate_early = True
signal.signal(signal.SIGINT, handler)
# Create SNN and Environment objects
snn = net.SpikingNeuralNetwork()
env = env.VrepEnvironment()
# Initialize environment, get initial state and reward
state, reward = env.reset()
# Simulate for training_length steps
for i in range(params.training_length):
# Run network for 50 ms: Get left and right output spikes, get weights
n_l, n_r, weights = snn.simulate(state, reward)
w_l = weights[0]
w_r = weights[1]
# Perform a step
# Get state, distance, pos_data, reward, terminate, steps,
# travelled_distances, vrep_steps
(state, distance, pos_data, reward, t, step, travelled_distances,
vrep_steps) = env.step(n_l, n_r)
STATS_EVERY = 10
# Exploration settings
epsilon = 1 # not a constant, going to be decayed
START_EPSILON_DECAYING = 1
END_EPSILON_DECAYING = EPISODES//2
epsilon_decay_value = epsilon/(END_EPSILON_DECAYING - START_EPSILON_DECAYING)
# For stats
ep_rewards = []
aggr_ep_rewards = {'ep': [], 'avg': [], 'max': [], 'min': []}
q_table = np.load("./qtables/1/800-qtable.npy", allow_pickle=True).item()
for episode in range(EPISODES):
state = env.reset()
if state not in q_table:
q_table[state] = np.random.uniform(
low=-2, high=0, size=env.action_space_n)
episode_reward = 0
done = False
while not done:
valid_actions = env.get_valid_actions(0)
action = max(valid_actions, key=lambda a: q_table[state][a])
new_state, reward, done = env.step(action)
episode_reward += reward
def runEpisode(agent, agent2, environment, discount, decision, decision2,
display, message, pause, episode):
returns = 0
returns2 = 0
totalDiscount2 = 0
totalDiscount = 1.0
a1Done = False
a2Done = False
environment.reset()
if 'startEpisode' in dir(agent): agent.startEpisode()
if 'startEpisode' in dir(agent2): agent2.startEpisode()
message("BEGINNING EPISODE: " + str(episode) + "\n")
while True:
# DISPLAY CURRENT STATE
state = environment.getCurrentState(1)
state2 = environment.getCurrentState(2)
display(state, state2)
pause()
actions = environment.getPossibleActions(state)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
a1Done = True
actions = environment.getPossibleActions(state2)
if len(actions) == 0:
message("EPISODE " + str(episode) + " COMPLETE: RETURN WAS " +
str(returns) + "\n")
a2Done = True
player1NextState = None
player2NextState = None
reward1 = None
reward2 = None
action1 = None
action2 = None
if (a1Done == False and a2Done == False):
action1 = decision(state)
action2 = decision2(state2)
result = env.twoAgentDoAction(action1, action2)
player1NextState = result[0]
player2NextState = result[1]
reward1 = result[2]
reward2 = result[3]
# EXECUTE ACTION
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action1, player1NextState,
reward1)
message("Agent 1 Started in state: " + str(state) +
"\nTook action: " + str(action1) + "\n.Ended in state: " +
str(player1NextState) + "\nGot reward: " + str(reward1) +
"\n")
returns += reward1 * totalDiscount
totalDiscount *= discount
# GET ACTION (USUALLY FROM AGENT)
action = decision2(state2)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
message("Agent 2 Started in state: " + str(state2) +
"\nTook action: " + str(action2) + "\nEnded in state: " +
str(player2NextState) + "\nGot reward: " + str(reward2) +
"\n")
# UPDATE LEARNER
if 'observeTransition' in dir(agent2):
agent2.observeTransition(state2, action, player2NextState,
reward2)
returns2 += reward2 * totalDiscount
totalDiscount2 *= discount
if a1Done or a2Done:
return returns
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
if 'stopEpisode' in dir(agent2):
agent2.stopEpisode()
initial_epsilon = qlearn.epsilon
epsilon_discount = 0.9986
start_time = time.time()
total_episodes = 10000
highest_reward = 0
f = open('q_table.txt', 'a')
f2 = open('q_table_list.pickle', 'wb')
for x in range(total_episodes):
done = False
cumulated_reward = 0 #Should going forward give more reward then L/R ?
observation = environment.reset()
if qlearn.epsilon > 0.05:
qlearn.epsilon *= epsilon_discount
state = ''.join(map(str, observation))
# print("State = ",state," observation = ",observation)
for i in range(1500):
# Pick an action based on the current state
action = qlearn.chooseAction(state)
# Execute the action and get feedback
observation, reward, done, info = environment.step(action)
cumulated_reward += reward
optimizer = chainer.optimizers.Adam(eps = 1e-2)
gamma = 0.95
#explorer = chainerrl.explorers.LinearDecayEpsilonGreedy(start_epsilon = 1.0, end_epsilon = 0.3 , decay_steps = 50000 ,random_action_func = env.random_move)
explorer = chainerrl.explorers.ConstantEpsilonGreedy(epsilon = 1.0, random_action_func = env.random_move)
replay_buffer = chainerrl.replay_buffer.ReplayBuffer(capacity=10 ** 5)
phi = lambda x: x.astype(np.float32, copy = False)
optimizer.setup(q_func)
agent = chainerrl.agents.DQN(
q_func, optimizer, replay_buffer, gamma, explorer,
replay_start_size = 500, update_interval = 1,
target_update_interval = 100, phi = phi)
#学習ループ
obs = env.reset()
r = 0
done = False
x = np.array([])
y = np.array([])
win_array = np.array([])
win_sum = 0
#agent.load('final_agent')
for cnt2 in range(1):
turny = 0
win = 0
print("試行回数" + str(cnt2+1))
for cnt in range(1):
def runEpisode(agent, environment, discount, decision, display, message, pause, episode, update=True, bounded=False):
returns = 0
totalDiscount = 1.0
environment.reset(agent)
if 'startEpisode' in dir(agent): agent.startEpisode(environment.getCurrentState())
#message("BEGINNING EPISODE: "+str(episode)+"\n")
timestep = 0
MAX_TIMESTEPS = 40
while True:
#print("timestep ", timestep)
# DISPLAY CURRENT STATE
state = environment.getCurrentState()
if display is not None:
display(state)
#pause()
#if timestep == 0 and episode == 1:
#if not update:
# input("")
if 'should_end_episode' in dir(agent) and agent.should_end_episode():
#message("EPISODE "+str(episode)+" COMPLETE: RETURN WAS "+str(returns)+"\n")
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
return (timestep, returns)
# END IF IN A TERMINAL STATE
actions = environment.getPossibleActions(state)
if len(actions) == 0 or (bounded and timestep >= MAX_TIMESTEPS):
if update and len(actions) == 0: # reached terminal state but we are using n-step agent
agent.update(state, None, None, None, update) # keep going until n-step agent says stop
continue # for n-step agent
elif not update: # not n-step agent so terminate on goal state or time exceeded
message("EPISODE "+str(episode)+" COMPLETE: RETURN WAS "+str(returns)+"\n")
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
return (timestep, returns)
# GET ACTION (USUALLY FROM AGENT)
action = decision(state)
#print(action)
if action == None:
raise 'Error: Agent returned None action'
# EXECUTE ACTION
nextState, reward = environment.doAction(action)
#message("Started in state: "+str(state)+
# "\nTook action: "+str(action)+
# "\nEnded in state: "+str(nextState)+
# "\nGot reward: "+str(reward)+"\n")
# UPDATE LEARNER
if 'observeTransition' in dir(agent):
agent.observeTransition(state, action, nextState, reward, update)
returns += reward * totalDiscount
totalDiscount *= discount
timestep += 1
if 'stopEpisode' in dir(agent):
agent.stopEpisode()
