20. Use of the MNIST dataset in TensorFlow

First, the acquisition of data sets

TensorFlow encapsulates the MNIST dataset to make it more convenient for us to use.

from tensorflow.examples.tutorials.mnist import input_data

# 读取数据集，第一次TensorFlow会自动下载数据集到下面的路径中, label 采用 one_hot 形式 
# label 默认采用 0~9 来表示，等价于 one_hot=False, read_data_sets 时会默认把图像 reshape(展平)
# 若想保留图像的二维结构，可以传入 reshape=False
mnist = input_data.read_data_sets('/path/to/MNIST_data', one_hot=True)

---

Second, the division of the data set

1. The dataset will be automatically divided into 3 subsets,train、validation 和 test

# 显示默认数据集的大小
print("Training data size: ", mnist.train.num_examples)
>>> Training data size:  55000

print("Validating data size: ", mnist.validation.num_examples)
>>> Validating data size:  5000

print("Testing data size: ", mnist.test.num_examples)
>>> Testing data size:  10000

---

2. Use train, validation and test to images 和 labels 方法obtain images and class labels

• The data types of images and class labels are bothndarray

# 显示数据集图像和类标的形状
print("Images shape:", mnist.train.images.shape, "Labels shape:", mnist.train.labels.shape)
>>> Images shape: (55000, 784) Labels shape: (55000, 10)

print("Images shape:", mnist.validation.images.shape, "Labels shape:", mnist.validation.labels.shape)
>>> Images shape: (5000, 784) Labels shape: (5000, 10)

print("Images shape:", mnist.test.images.shape, "Labels shape:", mnist.test.labels.shape)
>>> Images shape: (10000, 784) Labels shape: (10000, 10)

---

3. View a picture in the train dataset大小、类标和像素值

# 图片大小(28*28), TensorFlow 默认把它展开了，但这样丢失了图片的二维结构信息！
print("Example training data0: ", mnist.train.images[0].shape)
>>> Example training data0 shape:  (784,)

print("Example training data0 label: ", mnist.train.labels[0])
>>> Example training data0 label:  [ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.]

print("Example training data0: ", mnist.train.images[0])

• TensorFlow normalizes image pixel values ​​($\frac{|x - 255|}{255}$) , so the value range of the elements in the pixel matrix is ​​[0, 1], which represents the depth of the color . where 0 represents a white background and 1 represents a black foreground.

---

4. Use mnist.train.next_batch 随机to take out batch_size pictures and their class labels

batch_size = 100
xs, ys = mnist.train.next_batch(batch_size) 

print("X shape:", xs.shape)
>>> X shape: (100, 784)

print("Y shape:", ys.shape)
>>> Y shape: (100, 10)