Trained a linear Q model using mean-squared error loss, to predict movements of a snake, based on inputs of orientation of snake, position of food, etc.
The model learns how to play at a reasonable standard after about 100 games.
Linear Q-net model trained to play the classic game 'Snake'. Can play at superhuman levels after about 100 games/10 minutes of training.
Trained a linear Q model using mean-squared error loss, to predict movements of a snake, based on inputs of orientation of snake, position of food, etc.
The model learns how to play at a reasonable standard after about 100 games.
Linear Q-net model trained to play the classic game 'Snake'. Can play at superhuman levels after about 100 games/10 minutes of training.