A digit recognition neural network made from scratch (no high-level libraries were used, only NumPy and Pandas) trained on the MNIST dataset. I created this to grasp the mathematics behind a neural network, partially inspired by 3Blue1Brown's video series.
Simply install the dependencies for this project using pip and run
python visualizer.py
to activate the visualizer. This file extracts weights and biases from saved_nn.npz which stores parameters from previously trained neural networks using train.py. The neural network implementation itself is contained in main.py
Some free resources on neural networks that helped me in completing this project:
- 3Blue1Brown's neural network video series: https://www.youtube.com/watch?v=aircAruvnKk&list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
- The Coding Train's neural network visualizer implementation: https://www.youtube.com/watch?v=XJ7HLz9VYz0&list=PLRqwX-V7Uu6aCibgK1PTWWu9by6XFdCfh
- Samson Zhang's brief introduction to neural networks: https://www.youtube.com/watch?v=w8yWXqWQYmU&t=41s