Introduction

  A small python project to check if I understand elementary neural network
and backprobagation.
  I am using MNIST dataset,cross entropy loss,Relu neuron and simple
  gradient descent.
  this simple neural netword only comprise three layer,
  datalayer,hidden layer and loss layer.
   

Usage

  download MNIST dataset,and unzip them
  https://www.dropbox.com/s/tdv02nqvh8v4nhe/mnist_test.7z?dl=0    https://www.dropbox.com/s/yg96k64ubh01yvm/mnist_train.7z?dl=0  

   1   copy and paste 'main' block in mian.py   you can see the accuracy raising,and it can only reach about 90%

   2   copy and paste 'test' block in mian.py   run just-trained neural network on mnist_test dataset,see overall accuracy

   3    copy and paste 'visualize result' block in mian.py   It will show both image and predicted result

Reference:

http://neuralnetworksanddeeplearning.com/chap2.html

http://cs231n.github.io/

Introduction to Machine Learning | The MIT Press