mnist_classify-handwriten-NN-network

This project is refered from the cs231n-assignment code I finished in August 2018.

Require：

• python 3.6
• numpy
• download mnist data and store in the codes/data/

Function:

• Define arbitrary hiddens layers as you want
• Choose activation function between sigmoid and relu
• BatchNorm
• Dropout
• Model storing(FullyConnectedNet.storeModel())
• Model loading(FullyConnectedNet.loadModel())
• Choose different update rules among sgd/sgd_momentum/rmsprop/adam
• Seperated model definition module(codes.classifiers.fc_net.FullyConnectedNet) and updating module(codes.solver.Solver)

Demo

mnist_classification.ipynb

Postscript

The bestParams.pkl in the params directory I didn't upload here.
If you need email me to [email protected]