Convolutional Neural Network written by pure numpy.

This CNN is written by pure numpy and support trainning by BP algorithem.

• It follows pytorch's code style(forward, backward).
• Due to numpy limiation, it doesn't use GPU to accelerate. But the training speed is acceptable for the small dataset,(e.g. mnist).

Features:

1. It implements the SGD (Stochastic Gradient Descent)
2. It support multiple channel.

Install:

Downlaod the mnist from website, and extract all the files into data/mnist

Run:

python mnist_cnn.py