Tearth / Neural-Snake-2

The second version of NeuralSnake project written in Python and TensorFlow library.

Neural Snake 2

Second version of program which learns how to play Snake game using neural networks and genetic/evolutionary algorithms. Learning process is divided to generations, where every one contains 10 games. A few of neural networks with the highest scores are selected to breeding. After +-50 generations, the snake is capable to play without colliding with the wall and eat multiple feeds.

Neural network parameters

Input (1 = yes, -1 = no):

• is top available to move
• is right available to move
• is bottom available to move
• is left available to move
• is food above snake
• is food to the right of snake
• is food under snake
• is food to the left of snake

Output ([-1, 1]):

• move top
• move right
• move down
• move left

Used libraries:

• TensorFlow: neural networks stuff