easy_neural_network

a single c++ file for learing how data flow graphs work.

1. Introduction

This repository shows you how some deep learning frameworks work. The core parts of main.cpp are class Unit and class Gate. Unit is the node in computation graph if you know reverse-mode automatic differentiation. Gate is the math operation in computation graph. I derive many other gates from class Gate, such as AddGate to do add operation, SigGate to do a sigmod function operation. In the test part, I use mnist database to test if my program works well.

2. How to run this program

I recommend you to compile this program use g++. If you want to run this program, follow these steps:

git clone https://github.com/ucker/easy_neural_network.git
cd easy_neural_network
cd src
g++ main.cpp -o main -std=c++11
./main

After a few minutes, my result of the program looks like that:

3. Development plan

1. Add more annotations.
2. I will add some gates like ConvGate and MaxpoolGate soon, so that I can make a convolutional neural network.