Playing around with OpenAI gym using the SpaceInvader Environment. Just experimenting, no serious competition intended.
Run
python Spaceinvaders.py
for simple use. You can change the parameters within the setup.py
-parameters dictionary.
Includes the 'main thread' of the program. Very simple abstraction level.
Builds the tensorflow graph, as suggested by this famous paper by Mnih, Silver, Kavukcuooglu, Graves, Antanoglou, Wierstra and Riedmiller (DeepMind) Not all details are implemented yet, such as using the last 4 images as input.
Includes helper functions for saving the model features, saving figures in matlab plots, and preprocess images (to apply efficient squared convolution)
Creates environment and sets up parameters to be used during training
Implementation of a basic DeepQ-Learning algorithm. This is the heart of the program.
Feel free to contact me for suggestions on the style, implementation etc.