witek3100 / ChromeDinoAI

Training pytorch model to play chrome dino using custom genetic algorithm

ChromeDinoAI

Purpose of the project is to train pytorch model to play chrome dino game using own implementation of genetic algorithm.

Project structure

├── game    # game files
|     ├──  static
|     ├──  game.py 
|     ├──  player.py    
|     ├──  obstacles.py 
├── models     # trained models  
├── genetic_algorithm.py    # custom implementation of genetic algorithm
├── model.py    # pytorch model used in project
├── train.py     # training 
├── play.py     # playing vs trained model

How to run

TODO

Model

As the task is not especially hard model used is quite simple. Input consists of 5 neurons, then fully connected 10-neurons hidden layer with relu activation. At the output there is one neuron (also fully connected), because task is just a binary classification (jump or dodge) it uses sigmoid activation.

Training