I wanted to classify handwritten digits using core Python without any high level frameworks to essentially understand how a neural network works.

What it does

Learn from the MNIST dataset with Adam optimization algorithm and classify the test digit images.

How I built it

I built the vanilla neural network using python, numpy, pandas, and a lot of patience.

Challenges I ran into

Adam was extremely hard to implement! It took a lot of debugging of the inter-modular variables and dictionaries to get it to work.

Accomplishments that I'm proud of

I scored 99% accuracy on training set and 95% accuracy on test set. A good learning curve for me!

What I learned

Adam, deep learning, neural networks, patience.

What's next for Digit Recognizer

Also implement Adagrad on the dataset and graph the costs of different optimization algorithms together.

Built With

Share this project: