Tensorflow implementation of noisy DQN
This is a tensorflow implementation of the paper Noisy Networks For Exploration arXiv preprint arXiv:1706.10295 (2018) over the Open AI Atari gym environment.
To start training, simply run :-
python noisy_dqn.py
This will start the noisy DQN training for the pong atari game.
To do vanilla DQN trainining run:-(Note - epsiolon greedy exploration will not happen even for the vanilla DQN case)
python noisy_dqn.py --NoisyDQN False
Below is the plot showing the evolution of loss and reward with training for the noisy DQN case over the pong environment.
Code borrowed from RL-Adventure