Asynchronous Deep Reinforcement Learning

Intuition, implementation description and training results can be found here.

An attempt to implement asynchronous one-step Q-Learning from Google DeepMind's paper "Asynchronous Methods for Deep Reinforcement Learning", Mnih et al., 2016.

Benchmarks for current implementation of Asynchronous one-step Q-Learning:

Requirements

1. Linux based OS or Mac OS X;

2. Anaconda package (recommended);

   OR manually install python (both 2.7+ and 3.5+ versions are supported), and run in terminal:

   pip install six
   pip install future
   pip install scipy
   pip install pillow

3. TensorFlow;

4. OpenAI Gym.

Usage

To train your own model on 'Atari 2600 SpaceInvaders', simply run:

python run_dqn.py

To specify another environment, use --env flag, e.g:

python run_dqn.py --env Pong-v0

All available environments you can check here. Note, that current implementation supports environments only with raw pixels observations. Tested OpenAI Gym environments:

• SpaceInvaders-v0

• Pong-v0

To change amount of spawned threads, use --threads flag (by default = 8).

To use GPU instead of cpu, pass --gpu flag.

All available flags can be checked by: python run_dqn.py --help

To read TensorBoard logs, use: tensorboard --logdir=path/to/logdir

Trained models

To use pretrained agent, or change log folder, just use --logdir flag:

python run_dqn.py --logdir path/to/checkpoint/folder/

Model, trained on SpaceInvaders, over 80 millions of frames, can be downloaded from here.

Evaluation

To evaluate trained agent, use:

python run_dqn.py --eval --eval_dir folder/for/evaluation/write --logdir path/to/checkpoint/folder/