ppo-LunarLander-v7

PPO Agent playing LunarLander-v2

This is a trained model of a PPO agent playing LunarLander-v2 using the stable-baselines3 library.

Usage (with Stable-baselines3)

# TODO: Define a PPO MlpPolicy architecture
# We use MultiLayerPerceptron (MLPPolicy) because the input is a vector,
# if we had frames as input we would use CnnPolicy

# https://stackoverflow.com/questions/72483775/stable-baselines3-ppo-how-to-change-clip-range-parameter-during-training
def lrsched():
  def reallr(progress):
    lr = 0.004
    if progress < 0.5:
      lr = 0.0005
    if progress < 0.25:
      lr = 0.00025
    if progress < 0.15:
      lr = 0.0001
    return lr
  return reallr
model = PPO('MlpPolicy', env, n_steps = 512, batch_size = 64, n_epochs = 4, gamma = 0.999, gae_lambda = 0.98, ent_coef = 0.01, verbose=1, learning_rate=lrsched())


model.learn(total_timesteps=1000000)

model_name = "ppo-LunarLander-v2"
model.save(model_name)

# Create a new environment for evaluation
eval_env = Monitor(gym.make("LunarLander-v2"))

# Evaluate the model with 10 evaluation episodes and deterministic=True
mean_reward, std_reward = evaluate_policy(model, eval_env, n_eval_episodes=10, deterministic=True)

print(f"mean_reward={mean_reward:.2f} +/- {std_reward}")

#mean_reward=223.88 +/- 73.34697932613058
...

Diffs

• Changed when the learning rate changes in training
• Continued with a 512 n_steps value.