/DNN-from-scratch

Basic implementation of neural network from scratch

Primary LanguagePython

DNN-from-scratch

This is basic deep neural network implementation from scratch in python. This NN desigen for hand written digits classification.

Datasets and Training

The neural network is trained on 60000 MNIST hand written digits and tested on 10000 MNIST hand written digits. The overall accuracy of the model is 99.3% on training data and 98.7% on test data.

Architecture

The architecture of the model is divided as -

  • input layer with 784 input nodes
  • one hidden layer with 300 nodes
  • final output layer with 10 nodes Internally the model uses sigmoid activation function