ScratchNet - A Neural Network built from Scratch

This is a Neural Network built from Scratch using Python and NumPy. Try it out: https://scratchnet-mnist.streamlit.app/

Dataset: MNIST (70000 handwritten digits) Architecture: [784 -> 128 -> 64 ->] Final Accuracy: 97.61%

How to use

To run this please follow the set of instructions below.

REQUIREMENTS:

1. Python 3.13
2. Pip
3. Git (optional)

First clone the Repository (Alternatively you can download the ZIP file via clicking on the "<> Code" button and then navigating to the "Local" Tab and clicking on the "Download ZIP" option. Then you can extract it onto your Desktop):

git clone https://github.com/PrathamGhaywat/scratchnet.git

OPTIONAL: Creating an Virtual Enviroment will help you seprate the libraries from your main python compiler and mitigates the risk of corrupting it:

python -m venv scratchnet

#Activate the venv: on Linux and MacOS it would be different: source scratchnet/bin/activate
./scratchnet/scripts/activate

Then run:

pip install -r requirements.txt

Then run train_mnist.py

python src/core/train_mnist.py

This will train the neural network. But you can skip that process if you want and directly run the Web GUI:

streamlit run src/main.py

Then you can visit the given localhost address and use it!

License

The license is Apache License 2.0 - For more info see: License