Q-Learning SanDbox (QLSD)

Files

./CORE/

• fSandDataGetter.py - loop which produces one step in the sandbox world; !!!Run it for train!!!
• fSandBox.py - main class contains all sandbox world and creatures;
• fSandController - Q controller;
• fSandFun - features extraction and support functions;
• fSandDemo* - final model working demonstration; !!!Run it for demo!!!

./CORE/models/

• W_iterations#_features# - model in numPy format;

./test/ - folder with experiments; ./img/ - folder with screenshots and media;

./CORE/deepmode/ - deep learning model based on neural networks;

Dependencies

• NumPy
• PyGame
• Pillow
• • libjpeg-dev (before Pillow on Ubuntu)
• Common sense (=

Actions

|0|1|2| |3|4|5| |6|7|8|

|9| Stop

Final models

W_final_wd_stoper_3.npy - 7M iterations | Max fixed | Direct control | Without inertia | Final W_final_wd_stoper_4.npy - 2M iterations | Max fixed | Direct control | Without inertia | Final W_final_wd_stoper_2.npy - 7M iterations | Max NOT fixed | Direct control | Without inertia | Final

All other models are not final and just for comparing