NNforMNIST
A fully connected neural network for MNIST classification, implemented by Python 3.5 and Numpy.
Author: Mingqi Gao, Chongqing University
Homepage: https://mingqigao.com
Email: gaomingqi@cqu.edu.cn, im.mingqi@gmail.com
Figure. The architecture of implemented neural network, where dim is the number of the pixels in a MNIST image (i.e. width x height).
Requirements:
- Python 3.5
- Numpy
Description:
| Filename | Description |
|---|---|
dataloader.py |
Dataloader for MNIST dataset |
main.py |
Entry point for this project |
network.py |
Implementation for the proposed network |
model |
Parameters obtained by training process |
TRAIN_DATA.npy |
Meta-data created by training process (epoch, iteration, accuracy, loss) |
Training:
Uncomment and run train() in main.py. The updated weights and parameters will be saved in 'model' folder.
Testing:
Uncomment and run test() in main.py to obtain the classification accuracy on test set.
You can also run test10RandomImgs() to check classification results for 10 random images through graphic interface.