image_maze

Simple maze with augmented state to try RL agent

Step :

• Intall dependancies (python3.6, numpy, matplotlib, pytorch 0.3.0, keras). Although the code should work with newer pytorch, the "volatile" keyword has been removed and some parts will work less efficiently.
• Clone repository and move to newly created "image_maze" directory
• Optional (if next step fails because of imports): add src to path

  export "PYTHONPATH=$PYTHONPATH:/home/insert_your_path_here/image_maze/src/"
  

• Generate image datasets :

cd src/
python3 pretraining/preprocess.py
cd ..

You can now launch experiments either one at a time :

python3 main.py -config config/base.json

or in parallel and with a queue on (possibly several) GPU :

python3 main.py -config config/base.json

NB : using the parallel launcher can cause material damage; please make sure you understand it before launching it.