/dueling-DQN-pytorch

very easy implementation of dueling DQN in pytorch

Primary LanguagePython

dueling-DQN-pytorch

very easy implementation of dueling DQN in pytorch

(update implementation in tensorflow v1(tf1) & v2(tf2))

all things are in one file, easy to follow~~

Architecture

requirements

  • tensorflow (for tensorboard logging)
  • pytorch (>=1.0, 1.0.1 used in my experiment)
  • gym
  • ViZDoom

CartPole-v0

for training dueling DQN in CartPole, just run

python dueling_dqn.py

common, no description~

in CartPole-v0 the network will convergence to 200 episode reward very quickly~~

Visual doom

for training dueling DQN in Visual doom, just run

python visual_doom.py

for testing dueling DQN in Visual doom, just run

python visual_doom_test.py

use the basic environment of Visual doom. The agent could play very experienced after about 1000 times of games~~

also provide the dqn code for comparision.

reference

Dueling Network Architectures for Deep Reinforcement Learning (arxiv)