/WrittenDigitClassifier

A simple digit classifier built using the MNIST dataset

Primary LanguageJupyter NotebookMIT LicenseMIT

Simple Hand-Written Digit Classifier

A simple hand-written digit classifier which can classify digits with 98% accuracy.

You can visit the website to try this yourself.

Project components

Dataset

  • We are using the MNIST dataset to train and test our model to recognise digits.
  • You can view the dataset here.

Model

  • A dense neural network with architecture:
Layer Activation Neurons
1 ReLU 784
2 ReLU 10
3 Softmax 10
  • Trained on the MNIST dataset for 60 epochs using back propagation.

UI

  • A website made using Vue js to test and showcase the classification prowess of the model.
  • You can visit the website.

To build this on your computer

  1. cd to where you want the project to live.
    • Create a virtual environment and activate it. optional
  2. To set up the project, run these commands:
git clone "git@github.com:Varun487/WrittenDigitClassifier.git"

cd WrittenDigitClassifier

pip install -r requirements.txt
  1. To run the jupyter notebook:
jupter notebook
  1. To build the UI:
npm run build

Credits

This project is inspired by a medium article by Carlos Aguayo.