TensorFlow v2.x使用说明[2]-模型构建

# import lib
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

# 载入数据
mnist = tf.keras.datasets.mnist
(train_imgs, train_labels), (test_imgs, test_labels) = mnist.load_data()
train_imgs, test_imgs = train_imgs / 255.0, test_imgs / 255.0
train_imgs = train_imgs.reshape(60000,28,28,1)
test_imgs = test_imgs.reshape(10000,28,28,1)
# 模型构建
# 构建模型中的网络包含的所有内容，例如卷积层，池化层，BN层，全连接层，dropout。
cnn_model = keras.Sequential()
cnn_model.add(layers.Conv2D(input_shape=(28,28,1),filters=32,kernel_size=(3,3),activation='relu'))
cnn_model.add(layers.BatchNormalization())
cnn_model.add(layers.Conv2D(filters=64,kernel_size=(3,3),activation='relu'))
cnn_model.add(layers.BatchNormalization())
cnn_model.add(layers.MaxPool2D((2,2)))
cnn_model.add(layers.Conv2D(filters=128,kernel_size=(3,3),activation='relu'))
cnn_model.add(layers.BatchNormalization())
cnn_model.add(layers.Conv2D(filters=256,kernel_size=(3,3),activation='relu'))
cnn_model.add(layers.BatchNormalization())
cnn_model.add(layers.MaxPool2D((2,2)))
cnn_model.add(layers.Conv2D(filters=512,kernel_size=(3,3),activation='relu'))
cnn_model.add(layers.BatchNormalization())
cnn_model.add(layers.Flatten())
cnn_model.add(layers.Dense(1000,activation='relu'))
cnn_model.add(layers.Dropout(0.2))
cnn_model.add(layers.Dense(100,activation='relu'))
cnn_model.add(layers.Dropout(0.2))
cnn_model.add(layers.Dense(10,activation='softmax'))
cnn_model.summary()
# 模型编译
# 损失函数，优化器选择，精度度量
cnn_model.compile(optimizer='adam',
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])

# 模型训练
cnn_model.fit(train_imgs,train_labels,batch_size=256,epochs=10)
# 模型验证
cnn_model.evaluate(test_imgs,test_labels)

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras.layers import Conv2D, BatchNormalization, MaxPool2D, Flatten, Dense
import numpy as np

# 准备数据
(train_imgs, train_labels), (test_imgs,
                             test_labels) = keras.datasets.mnist.load_data()
train_imgs, test_imgs = train_imgs / 255.0, test_imgs / 255.0

# 增加一个维度，通过numpy方式
# train_imgs = np.expand_dims(train_imgs,axis=-1)
# test_imgs = np.expand_dims(test_imgs,axis=-1)

# 增加一个维度，通过tf方式
# ...代表索引的前面的所有维度
train_imgs = train_imgs[..., tf.newaxis]
test_imgs = test_imgs[..., tf.newaxis]

train_ds = tf.data.Dataset.from_tensor_slices(
    (train_imgs, train_labels)).shuffle(10000).batch(32)
test_ds = tf.data.Dataset.from_tensor_slices((test_imgs, test_labels)).batch(32)
print(type(train_ds))
# 定义模型
class MnistModel(keras.Model):
    def __init__(self):
        super(MnistModel, self).__init__()
        self.conv1 = Conv2D(
            32, (3, 3), activation='relu', input_shape=(28, 28, 1))
        self.bn1 = BatchNormalization()
        self.conv2 = Conv2D(64, (3, 3), activation='relu')
        self.bn2 = BatchNormalization()
        self.maxpool1 = MaxPool2D((2, 2))
        self.conv3 = Conv2D(128, (3, 3), activation='relu')
        self.bn3 = BatchNormalization()

        self.flatten = Flatten()

        self.dense1 = Dense(1000, activation='relu')
        self.dropout1 = keras.layers.Dropout(0.2)
        self.dense2 = Dense(100, activation='relu')
        self.output1 = Dense(10, activation='softmax')

    def call(self, x):
        cnn_block1 = self.bn1(self.conv1(x))
        cnn_block2 = self.maxpool1(self.bn2(self.conv2(cnn_block1)))
        cnn_block3 = self.bn3(self.conv3(cnn_block2))

        dense_block = self.dense2(
            self.dropout1(self.dense1(self.flatten(cnn_block3))))
        return self.output1(dense_block)


# 实例化一个模型
model = MnistModel()

# 定义验证相关的类
train_loss = keras.metrics.Mean()
test_acc = keras.metrics.SparseCategoricalAccuracy()

# 选择优化器
optimizer = keras.optimizers.Adam()

# 定义训练过程
@tf.function
def train_step(x, y):
    with tf.GradientTape() as tape:
        y_ = model(x)
        loss = keras.losses.SparseCategoricalCrossentropy()(y, y_)
        # loss对于所有训练变量求梯度
        gradients = tape.gradient(loss, model.trainable_variables)
        # 梯度下降法更新参数
        optimizer.apply_gradients(zip(gradients, model.trainable_variables))
        train_loss(loss)


#定义测试过程
@tf.function
def test_step(x,y):
    y_ = model(x)
    test_acc(y, y_)


# 训练模型
MAX_STEPS = 5000
# tf.data.Dataset本质是上是一个可迭代对象，用iter()获取它的迭代器，用next则可以获取下个batch的数据。

iter_data = iter(train_ds)
for i in range(MAX_STEPS):
    try:
        img, label = next(iter_data)
        if not (i+1) %100 :
            test_step(img,label)
            print(i, 'test acc:',test_acc.result())
        else:
            train_step(img,label)
            print(i, train_loss.result())
    except:
        pass
for img, label in test_ds:
    test_step(img,label)

print(test_acc.result())