Machine learning - artificial neural network （机器学习-人工神经网络）

1.Introduction 简介

This project refers to a book "Python neural network programming"[author:Tariq Rashid]. I build a BP artificial neural network with python. In addition, I add some Chinese annotation for the code , and add something new: support for Deep BP neural network , read & write dataset , Import and export model , picture test, etc.

用Python实现的人工神经网络源代码（参考书《Python 神经网络编程》[英.Tariq Rashid]）, 对书中的代码做了一些中文注释和优化，增加：支持深度BP、操作数据集、导入导出模型、图片测试等功能

code is written in Python3（代码基于: Python3）

I build an example web for Handwritten digital recognition with mnist data set, you can visit this website (我搭建了一个网站，上面有用BP神经网络识别手写数字的例子)

2.Getting started 下面介绍如何使用这个项目

1）use PIP to install required libs (先导入库)

pip install -r requirements.txt

2）run the code (运行)

python3 test_nn.py

3）if you use VSCode as your development tools ， you will meet this error ： (如果你用VSCode作为开发工具，可能会遇到这个错误)

[pylint] E1101:Module 'scipy.special' has no 'expit' member

you can modify your config file in VSCode (请更改你的设置)：

"python.linting.pylintArgs": ["--extension-pkg-whitelist=scipy"]

3.Project Structure 代码目录结构

item(目录)	introduction(说明)
`kdd/kdd.zip`	kdd cup 99 dataset , unzip to current folder (kdd数据集，直接解压kdd.zip到当前目录即可)
`mnist_dataset/`	100 taining data, 10 test data (100组数据的训练集，10组数据的测试集)
`mnist_dataset/dataset.url`	full mnist dataset download url (完整的mnist 数据集的下载地址)：mnist_train.csv ,mnist_test.csv
`mnist_dataset/w_hidden_output.txt,w_input_hidden.txt`	exported model data , can be imported by program (导出后的权重矩阵，可直接用来测试)
`img/mnist/`	100 images, which is exported from mnist dataset , used for testing (从mnist数据集中导出的图片，有100多张，可以用来测试),,
`dataset.py`	the code to read dataset (操作数据集代码)
`neural_network.py`	BP core code (神经网络代码)
`deep_neural_network.py`	Deep BP core code (深度神经网络代码)
`query.py`	test images code (图像测试代码)
`test_nn.py`	test BP in mnist code , can run (训练并测试神经网络代码 > 可运行)
`test_dnn.py`	test Deep BP in mnist code , can run (训练并测试多隐含层神经网络代码 > 可运行)
`test_kdd.py`	test BP used in kdd99 code BP , can run (神经网络在kdd上的应用 > 可运行)

Don't worry about 100 sets of training data are too little, run the code and you'll find that you can get [60%] correct with only 100 sets of test data !! (不用担心100组训练数据太少 ,运行代码你会发现只用100组测试数据就能达到【60%】的正确率！！)

Any question contact me with E-mail newham.cn@gmail.com (任何问题给我发邮件吧！)

newham / machine_learning