def test_on_policy_trainer():
logger = Logger()
env = gym.make("CartPole-v1")
algo = PPO1("mlp", env)
trainer = OnPolicyTrainer(algo, env, logger, epochs=1)
assert trainer.off_policy == False
trainer.train()
def test_on_policy_trainer():
env = VectorEnv("CartPole-v1", 2)
algo = PPO1("mlp", env)
trainer = OnPolicyTrainer(algo, env, ["stdout"], epochs=1, evaluate_episodes=2)
assert not trainer.off_policy
trainer.train()
trainer.evaluate()
def test_evaluate(self):
"""
test evaluating trained algorithm
"""
env = gym.make("CartPole-v0")
algo = PPO1("mlp", env, epochs=1)
algo.learn()
evaluate(algo, num_timesteps=10)
def test_ppo1(self):
env = gym.make("Pendulum-v0")
algo = PPO1("mlp", env, layers=[1, 1])
logger = Logger("./logs", ["csv"])
trainer = OnPolicyTrainer(algo, env, logger, epochs=1, render=False)
trainer.train()
shutil.rmtree("./logs")
def test_save_params(self):
"""
test saving algorithm state dict
"""
env = gym.make("CartPole-v0")
algo = PPO1("mlp", env, epochs=1, save_model="test_ckpt")
algo.learn()
assert len(os.listdir("test_ckpt/PPO1_CartPole-v0")) != 0
def test_ppo1(self):
env = gym.make("Pendulum-v0")
algo = PPO1("mlp", env, layers=[1, 1])
trainer = OnPolicyTrainer(
algo, env, log_mode=["csv"], logdir="./logs", epochs=1, evaluate_episodes=2
)
trainer.train()
trainer.evaluate()
shutil.rmtree("./logs")
def test_ppo1():
env = VectorEnv("CartPole-v0", 1)
algo = PPO1("mlp", env)
trainer = OnPolicyTrainer(algo,
env,
log_mode=["csv"],
logdir="./logs",
epochs=1)
trainer.train()
shutil.rmtree("./logs")
def test_ppo1_cnn():
env = VectorEnv("Pong-v0", 1, env_type="atari")
algo = PPO1("cnn", env)
trainer = OnPolicyTrainer(algo,
env,
log_mode=["csv"],
logdir="./logs",
epochs=1)
trainer.train()
shutil.rmtree("./logs")
def test_load_params(self):
"""
test loading algorithm parameters
"""
env = gym.make("CartPole-v0")
algo = PPO1("mlp",
env,
epochs=1,
load_model="test_ckpt/PPO1_CartPole-v0/0-log-0.pt")
rmtree("test_ckpt")
def test_load_params(self):
"""
test loading algorithm parameters
"""
env = VectorEnv("CartPole-v0", 1)
algo = PPO1(
"mlp",
env,
epochs=1,
load_model="test_ckpt/PPO1_CartPole-v0/0-log-0.pt",
)
rmtree("logs")
def test_load_params(self):
"""
test loading algorithm parameters
"""
env = VectorEnv("CartPole-v0", 1)
algo = PPO1("mlp", env)
trainer = OnPolicyTrainer(
algo,
env,
epochs=0,
load_model="test_ckpt/PPO1_CartPole-v0/0-log-0.pt")
trainer.train()
rmtree("logs")
def test_save_params(self):
"""
test saving algorithm state dict
"""
env = VectorEnv("CartPole-v0", 1)
algo = PPO1("mlp", env)
trainer = OnPolicyTrainer(algo,
env, ["stdout"],
save_model="test_ckpt",
save_interval=1,
epochs=1)
trainer.train()
assert len(os.listdir("test_ckpt/PPO1_CartPole-v0")) != 0