def test_add_step():
logger = Logger()
list_records = [Record(1.0), Record(1.0), Record(1.0), Record(1.0)]
for ite, record in enumerate(list_records):
logger.add_steps(record)
assert ite + 1 == len(logger.current_steps)
assert record == logger.current_steps[-1]
def test_record_init():
test_fail_values = [None, "deeee", [], {}, object()]
test_values = [1, 1.0, 0.0, 0, 125, 125.125]
for value in test_fail_values:
with pytest.raises(TypeError):
Record(value=value)
for value in test_values:
record = Record(value=value)
assert record.value == value
def do_step(self, observation, learn=True, logger=None, render=True):
"""
:param observation:
:param learn:
:param logger:
:param render: if show env render
:type render: bool
:return:
"""
if render:
self.render()
action = self.agent.get_action(observation=observation)
next_observation, reward, done, info = self.environment.step(action)
if learn:
self.agent.learn(observation, action, reward, next_observation, done)
if logger:
logger.add_steps(Record(reward))
return next_observation, done, reward
def test_max_records():
list_records = [[Record(1), Record(1),
Record(1), Record(1)],
[Record(2), Record(2),
Record(2), Record(2)],
[Record(1.0),
Record(1.0),
Record(1.0),
Record(1.0)],
[Record(2.0),
Record(2.0),
Record(2.0),
Record(2.0)],
[Record(1), Record(2),
Record(3), Record(4)], [Record(-1), Record(1)],
[Record(-10),
Record(-15),
Record(-20),
Record(-15)]]
list_value = [1, 2, 1.0, 2.0, 4, 1, -10]
for records, value in zip(list_records, list_value):
assert value == Record.max_records(records)
list_fail = ["dsdzs", ["dzdqzdq"], [1548, 1548], 1254]
for records in list_fail:
with pytest.raises(TypeError):
Record.max_records(records)
assert 0 == Record.max_records([])
def test_write_log():
list_steps = [[Record(1), Record(1),
Record(1), Record(1)],
[Record(2), Record(2),
Record(2), Record(2)],
[Record(1.0),
Record(1.0),
Record(1.0),
Record(1.0)],
[Record(2.0),
Record(2.0),
Record(2.0),
Record(2.0)], [Record(1),
Record(2),
Record(3),
Record(4)], [Record(-1),
Record(1)],
[Record(-10),
Record(-15),
Record(-20),
Record(-15)]]
for ite, records in enumerate(list_steps):
Logger.write_log("./runs", records, ite)
def test_log_episode():
summary_writer = FakeSummaryWriter()
list_steps = [[Record(1), Record(1),
Record(1), Record(1)],
[Record(2), Record(2),
Record(2), Record(2)],
[Record(1.0),
Record(1.0),
Record(1.0),
Record(1.0)],
[Record(2.0),
Record(2.0),
Record(2.0),
Record(2.0)], [Record(1),
Record(2),
Record(3),
Record(4)], [Record(-1),
Record(1)],
[Record(-10),
Record(-15),
Record(-20),
Record(-15)]]
for ite, records in enumerate(list_steps):
Logger.log_episode(summary_writer, records, ite)
assert (ite + 1) * 4 == len(summary_writer.add_scalar_call)
assert ite == summary_writer.add_scalar_call[-1][2]
def test_evaluate():
logger = Logger()
list_steps = [[Record(1), Record(1),
Record(1), Record(1)],
[Record(2), Record(2),
Record(2), Record(2)],
[Record(1.0),
Record(1.0),
Record(1.0),
Record(1.0)],
[Record(2.0),
Record(2.0),
Record(2.0),
Record(2.0)], [Record(1),
Record(2),
Record(3),
Record(4)], [Record(-1),
Record(1)],
[Record(-10),
Record(-15),
Record(-20),
Record(-15)]]
for ite, steps in enumerate(list_steps):
logger.current_steps = steps
logger.evaluate()
assert 0 == len(logger.episodes)
def test_add_episode():
logger = Logger()
list_episodes = [[Record(1), Record(1),
Record(1), Record(1)],
[Record(2), Record(2),
Record(2), Record(2)],
[Record(1.0),
Record(1.0),
Record(1.0),
Record(1.0)],
[Record(2.0),
Record(2.0),
Record(2.0),
Record(2.0)],
[Record(1), Record(2),
Record(3), Record(4)], [Record(-1),
Record(1)],
[Record(-10),
Record(-15),
Record(-20),
Record(-15)]]
for ite, episode in enumerate(list_episodes):
logger.add_episode(episode)
assert ite + 1 == len(logger.episodes)
assert episode == logger.episodes[-1]
from blobrl import Trainer, Record
from blobrl.agents import CategoricalDQN, DQN, DoubleDQN
import gym
if __name__ == "__main__":
for agent in [CategoricalDQN, DQN, DoubleDQN]:
env = gym.make("CartPole-v1")
a = agent(env.observation_space, env.action_space)
trainer = Trainer(environment=env, agent=agent)
for i in range(100):
trainer.train(max_episode=50, render=False, nb_evaluation=0)
m = max([Record.sum_records(e) for e in trainer.logger.episodes])
print(agent.__name__, i, m)
if m > 200:
break
print("####### ", agent.__name__, i, m, " #######")