Handwritten Digit Recognizer

Handwritten Digit Recognizer Github link

Web Application

Neural Network from scratch in Python to recognize handwritten digit achieved 98.45% test accuracy and using Keras CNN(Convolutional neural network) achieved 99.25% test accuracy deployed model to production

Dataset

Deep Neural network (DNN)

Neural Network Implementation from scratch using Python

Model representation

Implemented Neural network has 3 layers an input layer,a hidden layer and an output layer.

Feedforward and Regularized Cross entropy cost function

Implementation of DNN includes code for Feedforward and Regularized Cross entropy cost function

Backpropagation

Implementation of DNN includes code for Backpropagation

Model Performance

  • DNN: 98.45% test accuracy

The CNN model outperformed the DNN on the test and validation sets, hence I used the CNN model for Productionization

Productionization

In this step, I built a flask API endpoint that was hosted on a local webserver

Technologies

  • Python
  • Numpy for data cleaning
  • Matplotlib & Seaborn for data visualization
  • keras for model building
  • Python flask for HTTP server
  • HTML/CSS/Javascript for UI

link