Variational DQN

The implementation of Variational DQN based on Chainer, Tensorflow and Edward. Part of the Chainer code is borrowed from Chainer tutorial on DQN.

Variational DQN leverages variational inference subroutines to update DQN parameters.

Use the code

To run Variational DQN or DQN on Cartpole for 200 episodes

python main_VDQN.py --env CartPole-v1 --episodes 200
python main_DQN.py --env CartPole-v1 --episodes 200

Citations

If you use the code from this repo for academic research, you are very encouraged to cite the following papers.

Tang and Kucukelbir., Variational Deep Q Network. Bayesian Deep Learning Workshop, NIPS, 2017.

Tang and Agrawal., Exploration by Distributional Reinforcement Learning. International Joint Conference on Artificial Intelligence (IJCAI), 2018.