MNIST数据集比较小，一般入门机器学习都会采用这个数据集来训练

下载地址：yann.lecun.com/exdb/mnist/

有4个有用的文件：
train-images-idx3-ubyte: training set images
train-labels-idx1-ubyte: training set labels
t10k-images-idx3-ubyte: test set images
t10k-labels-idx1-ubyte: test set labels

The training set contains 60000 examples, and the test set 10000 examples. 数据集存储是用binary file存储的，黑白图片。知道这些基本上就够了，更多的请移步这里。

下面给出load数据集的代码：

import os
import struct
import numpy as np
import matplotlib.pyplot as plt

def load_mnist():
    '''
    Load mnist data
    http://yann.lecun.com/exdb/mnist/
    
    60000 training examples
    10000 test sets
    
    Arguments:
        kind: 'train' or 'test', string charater input with a default value 'train'
        
    Return:
        xxx_images: n*m array, n is the sample count, m is the feature number which is 28*28
        xxx_labels: class labels for each image, (0-9)
    '''
    
    root_path = '/home/cc/deep_learning/data_sets/mnist'
    
    train_labels_path = os.path.join(root_path, 'train-labels.idx1-ubyte')
    train_images_path = os.path.join(root_path, 'train-images.idx3-ubyte')
    
    test_labels_path = os.path.join(root_path, 't10k-labels.idx1-ubyte')
    test_images_path = os.path.join(root_path, 't10k-images.idx3-ubyte')
    
    with open(train_labels_path, 'rb') as lpath:
        # '>' denotes bigedian
        # 'I' denotes unsigned char
        magic, n = struct.unpack('>II', lpath.read(8))
        #loaded = np.fromfile(lpath, dtype = np.uint8)
        train_labels = np.fromfile(lpath, dtype = np.uint8).astype(np.float)
    
    with open(train_images_path, 'rb') as ipath:
        magic, num, rows, cols = struct.unpack('>IIII', ipath.read(16))
        loaded = np.fromfile(train_images_path, dtype = np.uint8)
        # images start from the 16th bytes
        train_images = loaded[16:].reshape(len(train_labels), 784).astype(np.float)
        
    with open(test_labels_path, 'rb') as lpath:
        # '>' denotes bigedian
        # 'I' denotes unsigned char
        magic, n = struct.unpack('>II', lpath.read(8))
        #loaded = np.fromfile(lpath, dtype = np.uint8)
        test_labels = np.fromfile(lpath, dtype = np.uint8).astype(np.float)
    
    with open(test_images_path, 'rb') as ipath:
        magic, num, rows, cols = struct.unpack('>IIII', ipath.read(16))
        loaded = np.fromfile(test_images_path, dtype = np.uint8)
        # images start from the 16th bytes
        test_images = loaded[16:].reshape(len(test_labels), 784)    
    
    return train_images, train_labels, test_images, test_labels

再看看图片集是什么样的：

def test_mnist_data():
    '''
    Just to check the data
    
    Argument:
        none
        
    Return:
        none
    '''
    train_images, train_labels, test_images, test_labels = load_mnist()
    fig, ax = plt.subplots(nrows = 2, ncols = 5, sharex = True, sharey = True)
    ax =ax.flatten()
    for i in range(10):
        img = train_images[i][:].reshape(28, 28)
        ax[i].imshow(img, cmap = 'Greys', interpolation = 'nearest')
        print('corresponding labels = %d' %train_labels[i])

if __name__ == '__main__':
    test_mnist_data()

跑出的结果如下：

image.png