liamsi/a2c-cartpole-v1

🚀 A2C CartPole-v1

This model was trained using Advantage Actor-Critic (A2C) from Stable-Baselines3 and logged using Weights & Biases.

🌎 Environment

• Gym ID: CartPole-v1
• Max Episode Length: 500 steps

🧠 Model Configuration

• Algorithm: Advantage Actor-Critic (A2C)
• Policy Network: Multi-Layer Perceptron (MLP)
• Architecture: [64, 64]
• Learning Rate: 5e-3
• Gamma (Discount factor): 0.99
• N Steps per Update: 5
• Max Gradient Norm: 0.5
• Total Training Timesteps: 50,000

📊 Evaluation

Evaluation was performed regularly during training every 100 timesteps. Metrics such as episode reward, episode length, and an accuracy metric (reward normalized by maximum steps per episode) were logged to Weights & Biases.

• Evaluation Frequency: every 100 timesteps
• Evaluation Mode: Deterministic
• Video of the trained agent: Watch the agent 🎬

📈 Logged Metrics

• Episode Reward
• Episode Length
• Accuracy (reward / max_steps_per_episode)

🔗 Links

• Weights & Biases Project

Enjoy exploring the trained agent! 🎬✨