/cartpole

OpenAI's cartpole env solver.

Primary LanguagePythonMIT LicenseMIT

Cartpole

Reinforcement Learning solution of the OpenAI's Cartpole.

Check out corresponding Medium article: Cartpole - Introduction to Reinforcement Learning (DQN - Deep Q-Learning)

About

A pole is attached by an un-actuated joint to a cart, which moves along a frictionless track. The system is controlled by applying a force of +1 or -1 to the cart. The pendulum starts upright, and the goal is to prevent it from falling over. A reward of +1 is provided for every timestep that the pole remains upright. The episode ends when the pole is more than 15 degrees from vertical, or the cart moves more than 2.4 units from the center. source

DQN

Standard DQN with Experience Replay.

Hyperparameters:

  • GAMMA = 0.95
  • LEARNING_RATE = 0.001
  • MEMORY_SIZE = 1000000
  • BATCH_SIZE = 20
  • EXPLORATION_MAX = 1.0
  • EXPLORATION_MIN = 0.01
  • EXPLORATION_DECAY = 0.995

Model structure:

  1. Dense layer - input: 4, output: 24, activation: relu
  2. Dense layer - input 24, output: 24, activation: relu
  3. Dense layer - input 24, output: 2, activation: linear
  • MSE loss function
  • Adam optimizer

Performance

CartPole-v0 defines "solving" as getting average reward of 195.0 over 100 consecutive trials. source

Example trial gif

Example trial chart

Solved trials chart

Author

Greg (Grzegorz) Surma

PORTFOLIO

GITHUB

BLOG

Support via PayPal