import numpy as np import tensorflow as tf from tensorflow import keras import matplotlib.pyplot as plt ## Let us define a plt function for simplicity def plt_loss(x,training_metric,testing_metric,ax,colors = ['b']): ax.plot(x,training_metric,'b',label = 'Train') ax.plot(x,testing_metric,'k',label = 'Test') ax.set_xlabel('Epochs') ax.set_ylabel('Accuarcy')# ax.set_ylabel('Categorical Crossentropy Loss') plt.legend() plt.grid() plt.show() tf.keras.utils.set_random_seed(1) ## We import the Minist Dataset using Keras.datasets (train_data, train_labels), (test_data, test_labels) = keras.datasets.mnist.load_data() ## We first vectorize the image (28*28) into a vector （784） train_data = train_data.reshape(train_data.shape[0],train_data.shape[1]train_data.shape[2]) # 60000784 test_data = test_data.reshape(test_data.shape[0],test_data.shape[1]test_data.shape[2]) # 10000784 ## We next change label number to a 10 dimensional vector, e.g., 1->[0,1,0,0,0,0,0,0,0,0] train_labels = keras.utils.to_categorical(train_labels,10) test_labels = keras.utils.to_categorical(test_labels,10) ## start to build a MLP model N_batch_size = 5000 N_epochs = 100 lr = 0.01 ## we build a three layer model, 784 -> 64 -> 10 MLP_4 = keras.models.Sequential([ keras.layers.Dense(128, input_shape=(784,),activation='relu'), keras.layers.Dense(64,activation='relu'), keras.layers.Dense(10,activation='softmax') ]) MLP_4.compile( optimizer=keras.optimizers.Adam(lr), loss= 'categorical_crossentropy', metrics = ['accuracy'] ) History = MLP_4.fit(train_data[:10000],train_labels[:10000], batch_size = N_batch_size, epochs = N_epochs,validation_data=(test_data,test_labels), shuffle=False) train_acc = History.history['accuracy'] test_acc = History.history['val_accuracy']在该模型的每一层（包括输出层）都分别加入L1，L2正则项训练，分别汇报测试数据准确率

可以使用以下代码在模型的每一层（包括输出层）都加入L1，L2正则项训练，并汇报测试数据准确率：

import numpy as np 
import tensorflow as tf 
from tensorflow import keras 
import matplotlib.pyplot as plt 

## Let us define a plt function for simplicity 
def plt_loss(x,training_metric,testing_metric,ax,colors = ['b']): 
    ax.plot(x,training_metric,'b',label = 'Train') 
    ax.plot(x,testing_metric,'k',label = 'Test') 
    ax.set_xlabel('Epochs') 
    ax.set_ylabel('Accuarcy')# ax.set_ylabel('Categorical Crossentropy Loss') 
    plt.legend() 
    plt.grid() 
    plt.show() 

tf.keras.utils.set_random_seed(1) 

## We import the Minist Dataset using Keras.datasets 
(train_data, train_labels), (test_data, test_labels) = keras.datasets.mnist.load_data() 

## We first vectorize the image (28*28) into a vector （784） 
train_data = train_data.reshape(train_data.shape[0],train_data.shape[1]*train_data.shape[2]) # 60000*784 
test_data = test_data.reshape(test_data.shape[0],test_data.shape[1]*test_data.shape[2]) # 10000*784 

## We next change label number to a 10 dimensional vector, e.g., 1->[0,1,0,0,0,0,0,0,0,0] 
train_labels = keras.utils.to_categorical(train_labels,10) 
test_labels = keras.utils.to_categorical(test_labels,10) 

## start to build a MLP model 
N_batch_size = 5000 
N_epochs = 100 
lr = 0.01 

## we build a three layer model, 784 -> 64 -> 10 
MLP_4 = keras.models.Sequential([ 
    keras.layers.Dense(128, input_shape=(784,),activation='relu',kernel_regularizer=keras.regularizers.l1_l2(l1=0.01, l2=0.01)), 
    keras.layers.Dense(64,activation='relu',kernel_regularizer=keras.regularizers.l1_l2(l1=0.01, l2=0.01)), 
    keras.layers.Dense(10,activation='softmax',kernel_regularizer=keras.regularizers.l1_l2(l1=0.01, l2=0.01)) 
]) 

MLP_4.compile( 
    optimizer=keras.optimizers.Adam(lr), 
    loss= 'categorical_crossentropy', 
    metrics = ['accuracy'] 
) 

History = MLP_4.fit(train_data[:10000],train_labels[:10000], 
                    batch_size = N_batch_size, 
                    epochs = N_epochs, 
                    validation_data=(test_data,test_labels), 
                    shuffle=False) 
train_acc = History.history['accuracy'] 
test_acc = History.history['val_accuracy'] 

print("MLP model with L1 and L2 regularization, test accuracy: ", test_acc[-1])

在该模型的每一层都加入L1和L2正则化项训练，并且L1和L2正则化项的系数都设置为0.01。最后输出模型在测试数据上的准确率。