vjg28 / Lunar-Lander-RL

This code base is an attempt to design an agent using Monte-Carlo Policy Gradient algorithm without / with baseline function.

Lunar-Lander-RL

This code base is an attempt to design an agent using Monte-Carlo Policy Gradient algorithm with baseline function.

How to use

1. git clone the repository.
2. Enter the directory of the repository. cd Lunar-Lander-RL and open two terminal windows.
3. Training: Terminal window 1 command python3 reinforce_baseline.py --task train
4. To visualize live plots: Terminal window 2 command python plotting.py
5. You can change almost all variables by changing its value in reinforce_baseline.py.
6. After completion, the directory will have two new .txt files, two image files and two directories with saved models. Do not change the location of those files before testing.
7. Testing: python reinforce_baseline.py --task train --exp_no 1 . You need to change the experiment number 1 to any integer k as it represents the folder in Saved_Exps/Test_k where all the results will be saved.

Results

Around 20 experiments were done with varyng hyperparameters and the best result in terms of solving the task - getting 200+ reward for 100 consequtive episodes in the least number of episodes is Test_11. There were some tests which scored more rewards on 100 consequtive episodes than the model in Test_11. Each folder has a logfile.txt representing the hyperparams value, test_log.txt for test results, reward_log.txt for episode logs during training.

Dependencies

• Tensorflow
• Gym
• Numpy
• MatPlotLib

Install them using pip

Contributing

Please feel free to create a Pull Request for any suggested improvements or error in the code. If you are a beginner, you can refer to this for getting started.

Support

If you found this useful, please consider starring(★) the repo so that it can reach a broader audience.

License

This project is licensed under the MIT License - see the LICENSE file for details.

References

• Sutton RL Book
• David Silver Lectures