/dqn

Primary LanguageJupyter NotebookMIT LicenseMIT

Deep Q-Networks(DQN)

This repo contains a very simple implementation of an agent playing LunarLander in the OpenAI-Gym environment, the training has been accomplished by using the DQN method of Deep Reinforcement Learning and TensorFlow(CPU).

after_training.mp4

Tensorboard

To inspect in-real-time logs, run the code below in the project directory to start tensorboard:

tensorboard --logdir "log/"

then, connect to its interface which, by default, should be https://localhost:6006

Dependencies

  • Python(3.8.8)
  • Numpy(1.19.3)
  • TensorFlow(2.6.0)
  • Matplotlib(3.4.2)
  • Gym(0.18.3)

Results

1