def basicSetup(self, resample_prob=0.0, explore=None):
pbt = PopulationBasedTraining(time_attr="training_iteration",
perturbation_interval=10,
resample_probability=resample_prob,
hyperparam_mutations={
"id_factor": [100],
"float_factor": lambda: 100.0,
"int_factor": lambda: 10,
},
custom_explore_fn=explore)
runner = _MockTrialRunner(pbt)
for i in range(5):
trial = _MockTrial(
i, {
"id_factor": i,
"float_factor": 2.0,
"const_factor": 3,
"int_factor": 10
})
runner.add_trial(trial)
trial.status = Trial.RUNNING
self.assertEqual(
pbt.on_trial_result(runner, trial, result(10, 50 * i)),
TrialScheduler.CONTINUE)
pbt.reset_stats()
return pbt, runner
def schedulerSetup(self, num_trials):
sched = PopulationBasedTraining()
runner = _MockTrialRunnerPBT()
for i in range(num_trials):
t = _MockTrialPBT("__parameter_tuning")
t.config = {'test': 1, 'test1': 1, 'env': 'test'}
t.experiment_tag = str(i)
runner._launch_trial(t)
return sched, runner
def basicSetup(self, resample_prob=0.0, explore=None):
pbt = PopulationBasedTraining(
time_attr="training_iteration",
perturbation_interval=10,
resample_probability=resample_prob,
hyperparam_mutations={
"id_factor": [100],
"float_factor": lambda: 100.0,
"int_factor": lambda: 10,
},
custom_explore_fn=explore)
runner = _MockTrialRunner(pbt)
for i in range(5):
trial = _MockTrial(
i,
{"id_factor": i, "float_factor": 2.0, "const_factor": 3,
"int_factor": 10})
runner.add_trial(trial)
trial.status = Trial.RUNNING
self.assertEqual(
pbt.on_trial_result(runner, trial, result(10, 50 * i)),
TrialScheduler.CONTINUE)
pbt.reset_stats()
return pbt, runner
# ensure we collect enough timesteps to do sgd
if config["timesteps_per_batch"] < config["sgd_batchsize"] * 2:
config["timesteps_per_batch"] = config["sgd_batchsize"] * 2
# ensure we run at least one sgd iter
if config["num_sgd_iter"] < 1:
config["num_sgd_iter"] = 1
return config
pbt = PopulationBasedTraining(
time_attr="time_total_s",
reward_attr="episode_reward_mean",
perturbation_interval=120,
resample_probability=0.25,
# Specifies the mutations of these hyperparams
hyperparam_mutations={
"lambda": lambda: random.uniform(0.9, 1.0),
"clip_param": lambda: random.uniform(0.01, 0.5),
"sgd_stepsize": [1e-3, 5e-4, 1e-4, 5e-5, 1e-5],
"num_sgd_iter": lambda: random.randint(1, 30),
"sgd_batchsize": lambda: random.randint(128, 16384),
"timesteps_per_batch": lambda: random.randint(2000, 160000),
},
custom_explore_fn=explore)
ray.init()
run_experiments(
{
"pbt_humanoid_test": {
"run": "PPO",
"env": "Humanoid-v1",
"repeat": 8,
},
"stop": {
"mean_accuracy": 0.80,
"timesteps_total": 300,
},
"config": {
"epochs": 1,
"batch_size": 64,
"lr": grid_search([10**-4, 10**-5]),
"decay": lambda spec: spec.config.lr / 100.0,
"dropout": grid_search([0.25, 0.5]),
},
"repeat": 4,
}
if args.smoke_test:
train_spec["config"]["lr"] = 10**-4
train_spec["config"]["dropout"] = 0.5
ray.init()
pbt = PopulationBasedTraining(time_attr="timesteps_total",
reward_attr="mean_accuracy",
perturbation_interval=10,
hyperparam_mutations={
"dropout":
lambda _: np.random.uniform(0, 1),
})
run_experiments({"pbt_cifar10": train_spec}, scheduler=pbt)
register_trainable("my_class", MyTrainableClass)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--smoke-test",
action="store_true",
help="Finish quickly for testing")
args, _ = parser.parse_known_args()
ray.init()
pbt = PopulationBasedTraining(
time_attr="training_iteration",
reward_attr="episode_reward_mean",
perturbation_interval=10,
hyperparam_mutations={
# Allow for scaling-based perturbations, with a uniform backing
# distribution for resampling.
"factor_1": lambda: random.uniform(0.0, 20.0),
# Allow perturbations within this set of categorical values.
"factor_2": [1, 2],
})
# Try to find the best factor 1 and factor 2
run_experiments(
{
"pbt_test": {
"run": "my_class",
"stop": {
"training_iteration": 2 if args.smoke_test else 99999
},
"repeat": 10,
if config["num_sgd_iter"] < 1:
config["num_sgd_iter"] = 1
return config
pbt = PopulationBasedTraining(
time_attr="training_iteration",
reward_attr="episode_reward_mean",
perturbation_interval=75,
resample_probability=0.25, #0.25,
# Specifies the mutations of these hyperparams
hyperparam_mutations={
"lambda": lambda: random.uniform(0.9, 1.0),
"gamma": lambda: random.uniform(0.99, 0.999),
"horizon": lambda: random.randint(256, 2048),
"clip_param": lambda: random.uniform(0.01, 0.4),
"sgd_stepsize": [1e-3, 5e-4, 1e-4, 5e-5, 1e-5],
"num_sgd_iter": lambda: random.randint(2, 10),
#"sgd_batchsize": lambda: random.randint(128, 16384),
"timesteps_per_batch": lambda: random.randint(16, 256),
"vf_loss_coeff": lambda: random.uniform(0.3, 1),
"entropy_coeff": lambda: random.uniform(0.0, 0.01),
#"kl_coeff": [0.0,0.2,1.0],
"kl_target": lambda: random.uniform(0.003, 0.03)
}
#,custom_explore_fn=explore
)
ray.init()
run_experiments(
{