def game_solved_basic(self,trainer: Engine):
passed = trainer.state.metrics['time_passed']
print("Game solved in %s, after %d episodes "
"and %d iterations!" % (
timedelta(seconds=int(passed)),
trainer.state.episode, trainer.state.iteration))
trainer.should_terminate = True
def game_solved(trainer: Engine):
print('Game solved in %s, after %d episode and %d iterations' % (
timedelta(seconds=trainer.state.metrics['time_passed']),
trainer.state.episode, trainer.state.iteration
))
trainer.should_terminate = True
def game_solved(trainer: Engine):
print("Game solved in %s, after %d episodes and %d iterations!" % (
timedelta(seconds=trainer.state.metrics["time_passed"]),
trainer.state.episode,
trainer.state.iteration,
))
trainer.should_terminate = True
def episode_completed(trainer: Engine):
passed = trainer.state.metrics.get('time_passed', 0)
avg_steps = trainer.state.metrics.get('avg_steps', 50)
avg_reward = trainer.state.metrics.get('avg_reward', 0.0)
print("Episode %d: reward=%.0f (avg %.2f), "
"steps=%s (avg %.2f), speed=%.1f f/s, "
"elapsed=%s" %
(trainer.state.episode, trainer.state.episode_reward, avg_reward,
trainer.state.episode_steps, avg_steps,
trainer.state.metrics.get('avg_fps',
0), timedelta(seconds=int(passed))))
if avg_steps < 15 and trainer.state.episode > 100:
print("Average steps has fallen below 10, stop training")
trainer.should_terminate = True
def test_terminate_on_nan_and_inf():
torch.manual_seed(12)
def update_fn(engine, batch):
pass
trainer = Engine(update_fn)
trainer.state = State()
h = TerminateOnNan()
trainer.state.output = 1.0
h(trainer)
assert not trainer.should_terminate
trainer.state.output = torch.tensor(123.45)
h(trainer)
assert not trainer.should_terminate
trainer.state.output = torch.asin(torch.randn(10, ))
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
trainer.state.output = np.array([1.0, 2.0])
h._output_transform = lambda x: x.tolist()
h(trainer)
assert not trainer.should_terminate
h._output_transform = lambda x: x
trainer.state.output = torch.asin(torch.randn(4, 4))
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
trainer.state.output = (10.0, 1.0 /
torch.randint(0, 2, size=(4, )).type(torch.float),
1.0)
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
trainer.state.output = (1.0, torch.tensor(1.0), "abc")
h(trainer)
assert not trainer.should_terminate
trainer.state.output = 1.0 / torch.randint(0, 2, size=(4, 4)).type(
torch.float)
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
trainer.state.output = (float("nan"), 10.0)
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
trainer.state.output = float("inf")
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
trainer.state.output = [float("nan"), 10.0]
h(trainer)
assert trainer.should_terminate
trainer.should_terminate = False
def game_solved(trainer: Engine):
passed = trainer.state.metrics["time_passed"]
print(
f"Game solved in {timedelta(seconds=int(passed))} after {trainer.state.episode}"
f" episodes and {trainer.state.iteration} iterations!")
trainer.should_terminate = True