Machine Learning Snake

Machine learning implementation of the snake game

Contents

• Overview
• Prerequisites
• Quick run
• Deep dive
  • Data generation
  • Dataset features
  • Modeling and result
  • Model performance

Overview

Creating a snake game using Python and Arcade library in which the snake goes for the apple by itself (without user involvement).
In the meantime data is being collected and later used to construct an Artificial Neural Network model with the help of TensorFlow package.
Then the trained model will be used to guide the snake to its food.

Prerequisites

This project is written in Python language, so first of all you need to have python installed. (My Python version was 3.8.10)

There are several libraries you also need to have installed. Run the following command line to install the required libraries all at once.

pip install -r requirements.txt

The above line will install NumPy, Arcade, Pandas and TensorFlow on your machine, all of which necessary to run the project.

Quick run

If you are already familiar with the project and don't want to go through an in-depth analysis, run the following commands in your terminal:

1. To collect data and create the dataset:

   python SnakeDataset.py
   

   Note 1: Make sure to press Esc button to quit the program. Any other method of quiting will result in not creating the dataset.

   Note 2: The default configuration of this program will create 2 data samples per second.

2. To build the ANN model:

   python TrainModel.py
   

3. To observe the result:

   python TestSnake.py
   

Deep dive

Data generation

So the rule-based game (AutomatedSnake.py) makes the snake move toward the apple on its own. The SnakeDataset.py is basically the rule-based game but it collects the feature information over time. So firstly, you must run this file to create the dataset. It will generate 1 data sample per 0.5 seconds. So forexample if you let it run for 2 minutes, there will be 240 samples in the dataset (which is more than enough).
Pay attention that for quitting, you must press Esc. Otherwise the dataset will not be saved on your computer.

Dataset features

The data obtained from the previous step will look like this:

x_diff	y_diff	directions
160	90	2
15	90	2
0	-35	1
-180	-140	0

There are 2 features:
The difference between snake's position and apple's position on x and y axis.

The label is the direction in which the snake is going. Since it can only go to four directions, there are 4 numbers. (0 to 3)

The snake will prioritize x axis. Meaning that most of the time it first moves toward the apple on x axis and when their x coordinates are equal (x_diff = 0), then it will move through the y axis.

Modeling and result

Now with the dataset being available you can run TrainModel.py to build the neural network model.
At the final step run TestSnake.py to observe the result.

Model performance

Here is a report of the model's performance:

	Accuracy	Loss
Train	0.96	0.15
Validation	0.97	0.16