The classic game of Super Mario. But make AI play it. Using pygame and NEAT.
Ensure you have the following installed on your local machine:
- Python 3 or above
- Suitable IDE
-
Clone the repository to your local machine:
git clone https://github.com/harini-1597/NEAT-Super-Mario.git
-
Navigate to the project directory:
cd NEAT-Super-Mario -
Create a Python Virtual Environment (if you use VSCode, you can create one using the Command palette itself)
python3 -m venv <myenvname>
-
Install requirements.txt:
pip install -r requirements.txt
NEAT-Super-Mario-Demo.mp4
NEAT, is an advanced Neural Network. Usually, neural networks are of this form:
NN's modify and optimize the functionality by modifying their weights, based on the result of the activation function.
NEAT on the other hand, has the ability to mutate, meaning, it can either:
- add nodes
- add connections to other nodes thus being able to vary its topology to identify the most suitable topology.
This is more suitable as the algorithm can tailor itself to be as complex or simple as the problem requires. For ex, Super Mario, a simple game, the NEAT can formulate a very simple Topology
But for more complex problems like User Based Recommendation systems, where each user is a different human, NEAT can evolve to be more complex.
To evolve a suitable solution to the problem, the user must provide a fitness_function something which tells the indivisula genome, in this case Mario, how well it performs.
In this case the best Mario is someone who reaches the higest level. Through a user specified amount of generations, wherein each generation of a Mario is produced by mutation of the most fit individuals of the previous generations.
The reproduction and mutation operations may add nodes and/or connections to the NN. When the preset number of generations is reached, or when at least one individual's fitness value exceeds the user-specified threshold, the algorithm terminates.
Read more about NEAT here: NEAT Official Docs