Skip to content

snake game ai using Keras implementation of deep-q-learning

Notifications You must be signed in to change notification settings

shinhs0506/snake_game_ai

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

snake_game_ai

snake game ai using Keras implementation of deep-q-learning

Game

  • snake game built using pygame module
  • 20x20 grid with: blue = head, green = body, red = food,

Neural Network

  • 2 conv2d layers with 1 fully connected hidden layer.
  • takes 30x30 resized raw pixel data as input and outputs 4 values, each representing the 4 directions a snake can move
  • predicts the next move that maximizes the reward

Overview of the game mechanics and deep-q-learning

  • (during each move in a game)
  • the neural network evaluates the state and predicts the best possible move
  • the snake receives the following points as rewards: food eaten = 1, moved closer to the food = 0.05, moved further away from the food = -0.05, dead = -1
  • the state, the reward the snake received, the next outcome state, and whether the snake died or not gets recorded in memory
  • the network is trained on batch of size 64 randomly chosed data from the memory
  • (above process repeats until the snake dies)

Run

  • specify whether to load the existing neural network or not, as well as the game iteration count
  • ex) python3 main.py load 500
  • reached max snake length of 20 after about 10,000 games of training
  • training still in progress

About

snake game ai using Keras implementation of deep-q-learning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages