Deep Q-Network implementation using Keras to solve the CartPole problem in OpenAI's gym. Implemented using a feed forward neural network as a non-linear value function approximator.

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.

• CartPole-v0 defines "solving" as getting average reward of 195.0 over 100 consecutive trials.
• The DQN is very susceptible to the learning rate of the neural network.
• Parameters used:

  params = {
      "num_episodes": 300,
      "epsilon": 1,
      "rep_size": 20,
      "discount_factor": 0.95,
      "eps_decay": 0.995,
      "min_eps": 0.01
  }    
  

1. Run pip install -r requirements.txt.
2. Tune hyperparameters in DQN-Code.ipynb, run the model.
3. Run the last cell in DQN-Code.ipynb to save the model files.
4. Run render.py to visualise the solution.