This repository contains the implementation of a simple two-layer neural network for digit recognition using the MNIST dataset. It's designed as an instructional example to help you understand the underlying math of neural networks.

To run this code, you'll need the following:

• Python 3.x
• NumPy
• Pandas
• Matplotlib

You can install the required dependencies using the following commands:

pip install numpy pandas matplotlib

To run the code, simply clone the repository and execute the following command:

python3 mnist_nn.py

This will train the neural network on the MNIST dataset and print the accuracy of the predictions.

The code is divided into several functions, each responsible for a specific step in the neural network training process:

• init_params(): Initializes the neural network parameters.
• ReLU(): Implements the ReLU activation function.
• softmax(): Implements the softmax activation function.
• forward_prop(): Performs forward propagation through the neural network.
• one_hot(): Converts labels to one-hot vectors.
• deriv_ReLu(): Computes the derivative of the ReLU activation function.
• back_prop(): Performs backpropagation to compute the gradients.
• update_params(): Updates the neural network parameters based on the gradients.
• get_predictions(): Converts network outputs to predicted labels.
• get_accuracy(): Calculates the accuracy of the predictions.
• gradient_descent(): Trains the neural network using gradient descent.

If you have any questions or suggestions, feel free to open an issue or pull request. Your contributions are welcome!

I hope this README file is helpful and informative. Please let me know if you have any other questions.