aadi-mishra / reinforcementlearningprojects Goto Github PK

Here you can code to understand basics of solving a RL based problem using Tensorflow.

• Tensorflow (keras)
• keras-rl
• OpenAI gym environment
• numpy

In each of the projects you can find

• environment.py - file which which provides state observations and reward based on certain actions. I can also be run independently with some randomized actions.

• train.py - Trains the model based on defined agent and policy.

• test.py - Performance can be visualized after training.

• Folder cartpole_DQN consists pendulum on a cart experiment with trained weights

• Folder spaceInvaders consists an RL agent playing classic atari Space Invader game. Unfortunately, I couldn't upload the weights which I had trained for almost 36 hours due to its large size.

• Folder customEnv consists a dqn agent trained on a custom shower environment, where the agent tries to control the optimal temperature(Discrete Action Space).