吴恩达编程作业-L4W2

编程实现一个残差网络，数据集还是之前的数据集
code：

import numpy as np
import tensorflow as tf
from tf_utils import *
import matplotlib.pyplot as plt
# 获取训练过程中的损失
class LossHistory(tf.keras.callbacks.Callback):
    def on_train_begin(self, logs=None):
        self.losses = []

    def on_epoch_end(self, epoch, logs=None):
        self.losses.append(logs['loss'])
def loss_callback(epoch, logs):
    loss = logs['loss']
    print('Epoch {}: Loss = {}'.format(epoch, loss.numpy()))
def convolutional_block(x,f,filters,s):
    filter1,filter2,filter3=filters
    x_shortcut=x

    # main path
    # first
    x=tf.keras.layers.Conv2D(filter1,kernel_size=1,strides=s,padding='valid')(x)
    x=tf.keras.layers.BatchNormalization()(x)
    x=tf.keras.layers.ReLU()(x)
    # second
    x=tf.keras.layers.Conv2D(filter2,kernel_size=f,strides=1,padding='same')(x)
    x=tf.keras.layers.BatchNormalization()(x)
    x=tf.keras.layers.ReLU()(x)
    # third
    x=tf.keras.layers.Conv2D(filter3,kernel_size=1,strides=1,padding='valid')(x)
    x=tf.keras.layers.BatchNormalization()(x)

    # shortcut path
    x_shortcut=tf.keras.layers.Conv2D(filter3,kernel_size=1,strides=s,padding='valid')(x_shortcut)
    x_shortcut=tf.keras.layers.BatchNormalization()(x)


    x=tf.keras.layers.add([x,x_shortcut])
    x=tf.keras.layers.ReLU()(x)
    
    return x
def identity_block(x,f,filters):
    f1,f2,f3=filters
    x_shortcut=x
    
    # main
    x=tf.keras.layers.Conv2D(f1,kernel_size=1,strides=1,padding='valid')(x)
    x=tf.keras.layers.BatchNormalization()(x)
    x=tf.keras.layers.ReLU()(x)

    x=tf.keras.layers.Conv2D(f2,kernel_size=f,strides=1,padding='same')(x)
    x=tf.keras.layers.BatchNormalization()(x)
    x=tf.keras.layers.ReLU()(x)

    x=tf.keras.layers.Conv2D(f3,kernel_size=1,strides=1,padding='valid')(x)
    x=tf.keras.layers.BatchNormalization()(x)

    x=tf.keras.layers.add([x,x_shortcut])
    x=tf.keras.layers.ReLU()(x)

    return x
def build_resnet(input_shape,num_classes=6):
    '''num_classes:分类数量,input_shape:输入数据的形状'''
    input_tensor=tf.keras.layers.Input(shape=input_shape)
    x=tf.keras.layers.ZeroPadding2D(padding=(3,3))(input_tensor)

    # 卷积层
    x=tf.keras.layers.Conv2D(64,kernel_size=7,strides=2)(x)
    x=tf.keras.layers.BatchNormalization()(x)
    x=tf.keras.layers.ReLU()(x)
    x=tf.keras.layers.MaxPooling2D((3,3),strides=2)(x)
    # strage2
    x=convolutional_block(x,f=3,filters=[64,64,256],s=1)
    x=identity_block(x,f=3,filters=[64,64,256])
    x=identity_block(x,f=3,filters=[64,64,256])

    # strage3
    x=convolutional_block(x,f=3,filters=[128,128,512],s=1)
    x=identity_block(x,f=3,filters=[128,128,512])
    x=identity_block(x,f=3,filters=[128,128,512])
    x=identity_block(x,f=3,filters=[128,128,512])

    # strage4
    x=convolutional_block(x,f=3,filters=[256,256,1024],s=2)
    x=identity_block(x,f=3,filters=[256,256,1024])
    x=identity_block(x,f=3,filters=[256,256,1024])
    x=identity_block(x,f=3,filters=[256,256,1024])
    x=identity_block(x,f=3,filters=[256,256,1024])
    x=identity_block(x,f=3,filters=[256,256,1024])

    # strage5
    x=convolutional_block(x,f=3,filters=[512,512,2048],s=2)
    x=identity_block(x,f=3,filters=[256,256,2048])
    x=identity_block(x,f=3,filters=[256,256,2048])
    x=tf.keras.layers.AveragePooling2D(pool_size=(2,2))(x)

    # outer
    x=tf.keras.layers.Flatten()(x)
    x=tf.keras.layers.Dense(num_classes,activation='softmax')(x)

    model=tf.keras.models.Model(inputs=input_tensor,outputs=x)
    return model
# 加载数据
X_train_orig, Y_train_orig, X_test_orig, Y_test_orig, classes = load_dataset()
print(X_train_orig.shape)
print(Y_train_orig.shape)

# 将数据归一化
X_train_orig=X_train_orig/255.0
X_test_orig=X_test_orig/255.0
# 转换标签的形状
Y_train_orig=Y_train_orig.T
Y_train_orig=np.squeeze(Y_train_orig)
Y_test_orig=Y_test_orig.T
Y_test_orig=np.squeeze(Y_test_orig)

# 创建和编译模型
model=build_resnet(input_shape=(64,64,3),num_classes=6)
model.compile(optimizer='adam',loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),metrics=['accuracy'])

loss_history = LossHistory()
# 训练模型
model.fit(X_train_orig,Y_train_orig,epochs=50,validation_data=(X_test_orig,Y_test_orig),callbacks=[loss_history])

# 将损失值转换为 NumPy 数组
losses_np = np.array(loss_history.losses)
plt.plot(losses_np)
plt.show()

# # 保存模型
model.save("model.h5")

实验结果：

很好，过拟合了，而且测试集的loss出现了极大的波动，问题未知。