00mjk / mnist-classification-with-neuralnet

MNIST Handwritten Digits Classification using 3 Layer Neural Net 98.7% Accuracy

MNIST-Classification-with-NeuralNet

MNIST Handwritten Digits Classification using 3 Layer Neural Net 98.7% Accuracy

Classifying the MNIST Digits using 3 Layer Neural Networks

Deskewing the Images yields much good accuracy.

Accuracy was 98.7% after deskewing the images before it was 98.4% simple 3 Layer Neural Nets.

Neural Network Model

Our Model has 3 Layers Containing

 1 Input Layer -> 28*28 U
 
 1 Hidden Layer -> 300 HU
 
 1 Output Layer -> 10 U

We have used the Backprop Algorithm for Training using the SGD Optimizer with Momentum . Applied PCA Dimensionality Reduction Technique to reduce the dimension to make dataset smaller, using 324 components to retain 99.78% variance of input data images

Need the Dataset that are for training and testing in one folder

Dependencies Required:

1. Python 2.7xx
2. Numpy, scipy, matplotlib Library Installed
3. OpenCV 3.xx, "MNIST" for reading data. Eg. pip install mnist

Run:

python mnist_nn.py  --path  '/home/......'```