/AI-Snake

Reinforcement Learning Snake Game

Primary LanguagePythonMIT LicenseMIT

AI-Snake

Reinforcement Learning Snake Game

alt text

Pre-Requirements:

  • python3
  • Google Chrome

Install Python Dependencies

pip3 install -r requirements.txt

Setup on Windows System:

  • Download suitable chrome driver from: https://chromedriver.chromium.org/downloads
  • Extract the .zip file. Copy the chromedriver.exe in the cloned directory folder (AI-snake/chromedriver.exe)
  • Install Pillow(PIL): 
    pip3 install Pillow==2.2.1
  • Open Deep_Reinforcement_Learning_Snake.py, Comment out line 9 and 101 and un-comment line 8 and 100

Setup on Linux System:

  • Install chrome driver: 
    sudo apt-get install chromium-chromedriver
  • Install Pyscreenshot: 
    sudo pip3 install pyscreenshot
  • Open Deep_Reinforcement_Learning_Snake.py, Comment out line 8 and 100 and un-comment line 9 and 101

CNN Network Architecture

alt text

Configuration:

  • You can open Deep_Reinforcement_Learning_Snake.py and change the following parameters: 
    img_rows, img_cols = 64,64
SPEED = 360
WIDTH = 8
HEIGHT = 8
ACTIONS = 4 
GAMMA = 0.99 
OBSERVATION = 65.
EXPLORE = 70
FINAL_EPSILON = 0.1
INITIAL_EPSILON = 1
REPLAY_MEMORY = 50000
BATCH = 64
FRAME_PER_ACTION = 1
LEARNING_RATE = 1e-4
img_channels = 4

Start Training:

python3 Deep_Reinforcement_Learning_Snake.py

Checking the scores:

  • File data/scores_df.csv contains scores of each game played by our model.
  • File data/actions_df.csv contains action at each step taken by the model.
  • File data/loss_df.csv contains loss at each step taken by the model.
  • File data/time_df.csv contains seconds played of each game.

Reference:

https://github.com/yenchenlin/DeepLearningFlappyBird

https://medium.com/acing-ai/how-i-build-an-ai-to-play-dino-run-e37f37bdf153