Disclaimer - This is a repository containing the assignments of 10703 - Deep Reinforcement Learning & Control for Spring'18. If you are currently taking the course, please do NOT refer to this repo. Any kind of code copying or referencing will be treated as plagiarism and results in serious disciplinary action at CMU.
Implementations of RL algorithms-
- DQN
- Advantage Actor-Critic
- Imitation Learning
- REINFORCE
- Dueling DQN
| CartPole | Mountain Car | LunarLander |
|---|---|---|
 |
 |
 |
Requirements-
- TensorFlow
- Keras
- Gym Box2D envs
To run DQN and Dueling DQN -
python DQN_Implementation.py with the following arguments-
| Argument | Description |
|---|---|
--env=ENVIRONMENT_NAME |
CartPole-v0, MountainCar-v0, LunarLander-v2 |
--render=1 OR 0 |
variable to enable render(1) or not(0) |
--train=1 OR 0 |
variable to train(1) the model or not(0) |
--type=MODEL_TYPE |
DQN,Dueling |
--save_folder=FOLDER_DIR |
folder directory to save videos (Optional). Videos are not saved if nothing is given |
--model_file=FILE_DIR |
File directory of saved model(Optional). Nothing is done if not given |
HyperParameters have been sectioned for easy alteration. You should be able to locate them easily by just searching 'Hyper'.
To run -
- Advantage-Actor Critic -
python a2c.py - REINFORCE -
python reinforce.py --render - Imitation -
python imitation --render(Weights are only for LunarLander-v2)