This project is refered from the cs231n-assignment code I finished in August 2018.
- python 3.6
- numpy
- download mnist data and store in the codes/data/
- 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)
mnist_classification.ipynb
The bestParams.pkl in the params directory I didn't upload here.
If you need email me to chenzhengyang117@gmail.com