Easy to set up...
Set up tesorflow or tensorflow-gpu:
For setting up tensorflow-gpu a on fresh Ubuntu installation:
- Install nvidia drivers after:
sudo apt-add-repository ppa:graphics-drivers/ppa
sudo apt-get update
and then go to 'Software & Updates', 'Additional Drivers' and choose a driver. - Download cuda-8.0 linux_86_64x.deb and cudnn5.1.deb and install both... restart your PC for them to work.
- Install Anaconda x86_64.sh, follow the Tensorflow tutorial and use: https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.2.1-cp36-cp36m-linux_x86_64.whl to install the tensorflow-gpu version in the conda environment.
Add to ".bashrc":
export PATH=/home/carl/anaconda2/bin:$PATH
export PATH=/usr/local/cuda-8.0/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH
Other dependecies:
-
Of course, OpenAI Universe: https://github.com/openai/universe
-
Install go-vncdiver with OpenGL support. Look here: https://github.com/openai/go-vncdriver
-
(Ignore if you have Anaconda) Python dependencies :
sudo pip install numpy
sudo pip install Pillow
sudo pip install scipy -
The python file must be run as root. For running applications as root from UnityLaucher (Ubuntu's bar on the left) follow this: https://askubuntu.com/questions/118822/how-to-launch-application-as-root-from-unity-launcher
-
(Might need for PyCharm users) To use TensorFlow in PyCharm IDE edit project configuration and add
LD_LIBRARY_PATHwith'/usr/local/cuda-8.0/lib64'to the environment variables.
Run RL.py to start training the agent. TensorFlow checkpoint saved every 10 runs of the game by default.
This attempt was inspired from a tutorial by Hvass-Labs who's author is Magnus Erik Hvass Pedersen. https://github.com/Hvass-Labs/TensorFlow-Tutorials/blob/master/16_Reinforcement_Learning.ipynb
This is a very plain implementation with just the minimum of code required to run the Q-Learning Algorithm.