def test_inverted_pendulum_v2(self):
env_name = "InvertedPendulum-v2"
env, (policy, optimizer), _ = create_models(env_name,
self.hidden_sizes, self.lr)
baseline = FutureReturnBaseline()
policy_update = REINFORCE(policy, optimizer, baseline)
result = solve(env_name, env, policy_update, logdir)
self.assertEqual(result, True)
def test_lunar_lander_v2(self):
env_name = "LunarLander-v2"
env, (policy, optimizer), _ = create_models(env_name,
self.hidden_sizes, self.lr)
baseline = FullReturnBaseline()
policy_update = REINFORCE(policy, optimizer, baseline)
result = solve(env_name, env, policy_update, logdir, epochs=500)
self.assertEqual(result, True)
def test_cartpole_v1(self):
env_name = "CartPole-v1"
env, (policy, optimizer), _ = create_models(env_name,
self.hidden_sizes, self.lr)
baseline = FutureReturnBaseline()
policy_update = REINFORCE(policy, optimizer, baseline)
result = solve(env_name, env, policy_update, logdir)
self.assertEqual(result, True)
parser.add_argument("--num-envs",
type=int,
default=multiprocessing.cpu_count() - 1)
parser.add_argument("--epochs", type=int, default=100)
parser.add_argument("--batch-size", type=int, default=5000)
parser.add_argument("--gamma", type=float, default=0.99)
parser.add_argument("--lam", type=float, default=0.97)
parser.add_argument("--lr", type=float, default=1e-2)
args = parser.parse_args()
logger.info("Using vanilla formulation of policy gradient.")
hidden_sizes = [100]
lr = 1e-2
env = make_env(args.env_name, args.num_envs)
(policy, optimizer), (value, vopt) = create_models(env, hidden_sizes,
args.lr, args.lr)
baseline = GAEBaseline(value, gamma=args.gamma, lambda_=args.lam)
policy_update = REINFORCE(policy, optimizer, baseline)
vbaseline = DiscountedReturnBaseline(gamma=args.gamma, normalize=False)
value_update = ValueUpdate(value, vopt, vbaseline, iters=1)
update = ActorCriticUpdate(policy_update, value_update)
solve(
args.env_name,
env,
policy_update,
logdir,
epochs=args.epochs,
batch_size=args.batch_size,
)
dist = policy(obs)
log_probs = dist.log_prob(acts)
loss = -((weights * log_probs).mean())
return loss
if __name__ == "__main__":
import argparse
from rl_baselines.core import (
solve,
create_models,
FullReturnBaseline,
FutureReturnBaseline,
)
parser = argparse.ArgumentParser()
parser.add_argument("--env-name", "--env", type=str, default="CartPole-v0")
parser.add_argument("--epochs", type=int, default=100)
parser.add_argument("--render", action="store_true")
parser.add_argument("--lr", type=float, default=1e-2)
args = parser.parse_args()
logger.info("Using vanilla formulation of policy gradient.")
hidden_sizes = [100]
lr = 1e-2
env, (policy, optimizer), _ = create_models(args.env_name, hidden_sizes,
lr)
baseline = FutureReturnBaseline()
policy_update = REINFORCE(policy, optimizer, baseline)
solve(args.env_name, env, policy_update, logdir, epochs=args.epochs)
def test_cartpole_v0(self):
env_name = "CartPole-v0"
env = make_env(env_name, 1)
policy_update = self.get_update_model(env)
result = solve(env_name, env, policy_update, logdir)
self.assertEqual(result, True)
def test_lunar_lander_v2(self):
env_name = "LunarLander-v2"
env = make_env(env_name, 1)
policy_update = self.get_update_model(env)
result = solve(env_name, env, policy_update, logdir, epochs=500)
self.assertEqual(result, True)