# 1.MNIST dataset Profile
1.MNIST data set that is MNIST handwritten collection of digital images, machine learning is a very classic data sets.
2.MNIST constituted by a digital image data set from 0 to 9. There are sixty thousand training image, test images have ten thousand. These images can be used for learning and reasoning.
3.MNIST are grayscale image data 28px * 28px, the value of each pixel between 0 and 255. Each image respectively labeled "0", "1", "2", "3", "4", "5" ... other labels.
General use 4.MNIST data sets:
first image of learning by training, and then learning to be able to measure the model test images can be classified correctly to what extent.
Download and use # 2.MNIST datasets
"Getting deep learning - theory and implementation of Python-based" (Saito with Kang Yi, Lu Yujie translated) book provides a download URL of the source code (http://www.ituring.com.cn/book/1921) . MNIST download the code data set in the dataset / mnist.py file, the following example (source code ch03 / mnist_show) is a simple to use and analysis of the data set MNIST.
# coding: utf-8
# 这里是对便利脚本dataset/mnist.py中load_mnist()函数的使用
import sys, os # os模块用于提供系统级别的操作，sys模块用于提供对解释器相关的操作
sys.path.append(os.pardir) # sys.path.append()用于添加系统环境变量，os.pardir就是当前目录的父目录
import numpy as np
from dataset.mnist import load_mnist # 因为已添加父目录为系统环境变量，所以此处可顺利导入父目录下dataset.mnist中的函数
from PIL import Image
# 用于显示图片的函数
def img_show(img):
pil_img = Image.fromarray(np.uint8(img)) # 将数组转换成图像
pil_img.show() # 显示图像
(x_train, t_train), (x_test, t_test) = load_mnist(flatten=True, normalize=False)
# 这里通过load_mnist函数读入了MNIST数据
# 该函数以（训练图像，训练标签），（测试图像，测试标签）的形式返回读入的MNIST数据
# 该函数可设置三个参数：normalize,flatten,one_hot_label
# normalize=true表示将输入图像像素正规化为0.0到1.0的值;反之，图像像素保持原来的0到255
# flatten=true表示输入图像会保存为由784个元素构成的一维数组;反之，图像保存为1*28*28的三维数组
# one_hot_label=true表示标签保存为one_hot形式（仅正确解标签为1其余为0的数组形式：[0,0,1,0,0,0,0,0,0,0]）;反之，标签简单保存。（如：1、2、3...）
img = x_train[0] # 将第一张训练图像存至img变量
label = t_train[0] # 将第一个训练标签存至label变量
print(label) # 打印出该训练标签
print(img.shape) # 打印出该训练图像的存储形状（像素数组存储格式）
img = img.reshape(28, 28) # 利用reshape函数把图像（像素数组）的形状变为原来的尺寸
print(img.shape) # (28, 28)
img_show(img) # 显示图像Results are as follows:
# This blog reference to the "deep learning portal - based on theory and implementation of Python" (Saito with Kang Yi, Lu Yujie translation), especially in this statement.
