Shulou(Shulou.com)06/03 Report--

This article mainly introduces the example analysis of MNIST data set in TensorFlow, which is very detailed and has certain reference value. Friends who are interested must read it!

Introduction to MNIST dataset

MNIST contains 0-9 handwritten numbers, a total of 60000 training sets and 10000 test sets. The format of the data is a grayscale image of a single channel 28 to 28.

Introduction of LeNet Model

LeNet network was first proposed by Yann LeCun et al of New York University in 1998, also known as LeNet5. LeNet is the ancestor of neural network and is known as the "Hello World" of convolution neural network.

Convolution

Pooling (downsampling)

Activation function (ReLU)

LeNet layer by layer analysis 1. The first convolution layer

two。 The first pooled layer

3. The second convolution layer

4. The second pooled layer

5. Fully connected convolution layer

6. Fully connected layer

7. Full connection layer (output layer)

Code implementation guide package from tensorflow.keras.datasets import mnistfrom matplotlib import pyplot as pltimport numpy as npimport tensorflow as tf read & view data #-1. Read & View data-# read data (X_train, y_train), (X_test, y_test) = mnist.load_data () # dataset View print (X_train.shape) # (60000, 28, 28) print (y_train.shape) # (60000,) print (X_test.shape) # (10000, 28) 28) print (y_test.shape) # (10000,) print (type (X_train)) # # Image display plt.imshow (X_train [0], cmap= "Greys") # View the first picture plt.show () data preprocessing #-2. Activation=tf.nn.relu) # first pooled layer pool_layer_1 = tf.keras.layers.MaxPool2D (pool_size= (2,2), padding= "same") # second convolution layer conv_layer_2 = tf.keras.layers.Conv2D (filters=16, kernel_size= (5,5), padding= "valid" Activation=tf.nn.relu) # second pooling layer pool_layer_2 = tf.keras.layers.MaxPool2D (padding= "same") # flattened flatten = tf.keras.layers.Flatten () # first fully connected layer fc_layer_1 = tf.keras.layers.Dense (units=120, activation=tf.nn.relu) # second fully connected layer fc_layer_2 = tf.keras.layers.Dense (units=84, activation=tf.nn.softmax) # output layer output_layer = tf.keras.layers.Dense (units=10 Activation=tf.nn.softmax)

Usage of convolution Conv2D:

Filters: number of convolution kernels

Kernel_size: convolution kernel size

Strides = (1,1): step size

Padding = "vaild": valid for discard, same for make-up

Activation = tf.nn.relu: activate function

Data_format = None: default channels_last

Usage of pooled AveragePooling2D:

Pool_size: the size of the pool

Strides = (1,1): step size

Padding = "vaild": valid for discard, same for make-up

Activation = tf.nn.relu: activate function

Data_format = None: default channels_last

Usage of fully connected Dense:

Units: dimension of output

Activation: activation function

Strides = (1,1): step size

Padding = "vaild": valid for discard, same for make-up

Activation = tf.nn.relu: activate function

Data_format = None: default channels_last

# Model instantiation model = tf.keras.models.Sequential ([tf.keras.layers.Conv2D (filters=6, kernel_size= (5,5), padding='valid', activation=tf.nn.relu, input_shape= (32,32,1)), # relu tf.keras.layers.AveragePooling2D (pool_size= (2,2), strides= (2,2), padding='same'), tf.keras.layers.Conv2D (filters=16, kernel_size= (5,5)) Padding='valid', activation=tf.nn.relu), tf.keras.layers.AveragePooling2D (pool_size= (2,2), strides= (2,2), padding='same'), tf.keras.layers.Flatten (), tf.keras.layers.Dense (units=120, activation=tf.nn.relu), tf.keras.layers.Dense (units=84, activation=tf.nn.relu), tf.keras.layers.Dense (units=10, activation=tf.nn.softmax)]) # Model display model.summary ()

Output result:

Training model #-4. Training model-# set hyperparameter num_epochs = 10 # training rounds batch_size = 1000 # batch size learning_rate = 0.001 # Learning rate # define optimizer adam_optimizer = tf.keras.optimizers.Adam (learning_rate) model.compile (optimizer=adam_optimizer,loss=tf.keras.losses.sparse_categorical_crossentropy,metrics= ['accuracy'])

The usage of complie:

Optimizer: optimizer

Loss: loss function

Metrics: evaluation

With tf.Session () as sess: # initialize all the variables init = tf.global_variables_initializer () sess.run (init) model.fit (x = model.evaluate (X_test, y_test)) # loss value & metrics values

Output result:

The usage of fit:

X: training set

Y: test set

Batch_size: batch size

Enpochs: number of training times

Save the model #-5. Save the model-model.save ('lenet_model.h6') process summary

Complete code from tensorflow.keras.datasets import mnistfrom matplotlib import pyplot as pltimport numpy as npimport tensorflow as tf#-1. Read & View data-# read data (X_train, y_train), (X_test, y_test) = mnist.load_data () # dataset View print (X_train.shape) # (60000, 28, 28) print (y_train.shape) # (60000,) print (X_test.shape) # (10000, 28) 28) print (y_test.shape) # (10000,) print (type (X_train)) # # Image display plt.imshow (X_train [0], cmap= "Greys") # View the first picture plt.show () #-2. Data preprocessing-# format conversion (expanding the picture from 28x28 to 32x32) X_train = np.pad (X_train, ((0,0), (2,2), (2,2)), "constant", constant_values=0) X_test = np.pad (X_test, (0,0), (2,2), (2,2)), "constant" Constant_values=0) print (X_train.shape) # (60000, 32, 32) print (X_test.shape) # (10000, 32, 32) # dataset format transformation X_train = X_train.astype (np.float32) X_test = X_test.astype (np.float32) # data regularization X _ train / = 255X_test / = 25 data dimension transformation X_train = np.expand_dims (X_train Axis=-1) X_test = np.expand_dims (X_test, axis=-1) print (X_train.shape) # (60000, 32, 32, 1) print (X_test.shape) # (10000, 32, 32, 1) #-3. Model building-# first convolution layer conv_layer_1 = tf.keras.layers.Conv2D (filters=6, kernel_size= (5,5), padding= "valid", activation=tf.nn.relu) # first pooled layer pool_layer_1 = tf.keras.layers.MaxPool2D (pool_size= (2,2), padding= "same") # second convolution layer conv_layer_2 = tf.keras.layers.Conv2D (filters=16, kernel_size= (5) 5), padding= "valid", activation=tf.nn.relu) # second pooled layer pool_layer_2 = tf.keras.layers.MaxPool2D (padding= "same") # flattened flatten = tf.keras.layers.Flatten () # first fully connected layer fc_layer_1 = tf.keras.layers.Dense (units=120, activation=tf.nn.relu) # second fully connected layer fc_layer_2 = tf.keras.layers.Dense (units=84 Activation=tf.nn.softmax) # output layer output_layer = tf.keras.layers.Dense (units=10, activation=tf.nn.softmax) # Model instantiation model = tf.keras.models.Sequential ([tf.keras.layers.Conv2D (filters=6, kernel_size= (5,5), padding='valid', activation=tf.nn.relu, input_shape= (32,32,1)), # relu tf.keras.layers.AveragePooling2D (pool_size= (2,2), strides= (2)) ), padding='same'), tf.keras.layers.Conv2D (filters=16, kernel_size= (5,5), padding='valid', activation=tf.nn.relu), tf.keras.layers.AveragePooling2D (pool_size= (2,2), strides= (2,2), padding='same'), tf.keras.layers.Flatten (), tf.keras.layers.Dense (units=120, activation=tf.nn.relu), tf.keras.layers.Dense (units=84, activation=tf.nn.relu) Tf.keras.layers.Dense (units=10, activation=tf.nn.softmax)]) # Model display model.summary () #-4. Training model-# set hyperparameter num_epochs = 10 # training rounds batch_size = 1000 # batch size learning_rate = 0.001 # Learning rate # define optimizer adam_optimizer = tf.keras.optimizers.Adam (learning_rate) model.compile (optimizer=adam_optimizer,loss=tf.keras.losses.sparse_categorical_crossentropy Metrics= ['accuracy']) with tf.Session () as sess: # initialize all variables init = tf.global_variables_initializer () sess.run (init) model.fit. Save the model-model.save ('lenet_model.h6') above is all the content of the article "sample Analysis of MNIST datasets in TensorFlow". Thank you for reading! Hope to share the content to help you, more related knowledge, welcome to follow the industry information channel!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.