Fashion-MNIST Classification from scratch

Background

This is a repository to reproduce SVM and CNN from scratch in Numpy:

1. A class of SVM with kernel method and SMO optimization in ./fashion_mnist/svm.py.
2. A class of 2-(CONV+POOL)+1FC CNN model with SGD, ConvNet can be found in ./fashion_mnist/conv_net.py. In details, there are 3 classes for layers:
   • ConvLayer in ./fashion_mnist/conv_layer.py
   • MaxPoolLayer in ./fashion_mnist/max_pool_layer.py
   • DenseLayer in ./fashion_mnist/dense_layer.py

Note that there are 3 base classes for you to understand the machinism clearly. The actual classes I uesd perform the same behaviour as them with speeding up in vectorized operation.

Requirement

numpy-1.27.2

Usage

To be updated...