DRL_Examples

Deep Learning Course Examples

Baylor University, Department of Electrical and Computer Engineering

Deep Reinforcement Learning (DRL)

Course Examples in TensorFlow and Keras:

1. Use gym environments for developing and comparing reinforcement learning algorithms. http://gym.openai.com/

   gym_display.py

   Note: Gym environment display - Random operation of the selected gym environment.

2. Q-learning for an agent to play the cartpole game.

   cartpole.py

   Note: The training-data collection phase and the network training phase are separate.

3. Deep Q Learning to play the cartpole game.

   DQN_run_cartpole.py (Main program - Tensorflow 1.0)

   RL_DeepQNetwork.py (Deep Q Network Class)

   Note: (1) Train DNN while playing games and increasing experience; (2) Single-step update of DNN parameters; (3) Experience replay with low correlation; (4) Fixed Q-targets.

   Algorithm for Deep Q Learning with Experience Replay:

   

4. Policy Gradient Method to play the cartpole game.

   PolicyGradient_run_cartpole.py (Main program - Tensorflow 1.0)

   RL_PolicyGradient.py (Policy Gradient Class)

   Note: (1) Full-episode update of DNN parameters.

   Algorithm for REINFORCE Policy Gradients:

   

5. Actor-Critic Method to play the cartpole game.

   ActorCritic_run_cartpole.py (Tensorflow 1.0)

   Actor-Critic Algorithm:

   

6. Deep Deterministic Policy Gradient to play the pendulum game (continuous-valued actions)

   DDPG_pendulum.py (Tensorflow 1.0)

   DDPG Algorithm: