MNIST Character Recognition Neural Net

---

Three Layer Neural Network Trained on the MNIST dataset to recognise human handwritten numbers

---

1. Intro

This neural network is written following lessons from a great ML book for beginners, Make Your Own Neural Network, by Tariq Rashid.

This simple but powerful three-layer neural network is trained on data from the MNIST Dataset.

2. Dataset

The dataset used is the MNIST Dataset reformated to .csv. This project contains only a hundred training examples and ten test cases in the mnist_train_100.csv and mnist_test_10.csv files respectively. These are located in the mnist_dataset directory.

However, the code in the main.py was last trained on the full 60,000 training examples file and tested on the 10,000 test case file that you can download from here and here respectively.

You should place these files in the mnist_dataset folder before running the code. Adjust the learning rate, number of epoch, hidden nodes or other variables to experiment.

3. Results

I managed to attain an accuracy rate of around 97.48% with the following parameters: hidden_nodes = 200, learning_rate = 0.1 and epochs = 5.