Keras训练神经网络DEMO——全连接神经网络训练MNIST

#%%导入数据
from keras.datasets import mnist
import matplotlib.pyplot as plt
import numpy as np 

(train_data,train_labels),(test_data,test_labels) = mnist.load_data()

#%%数据预处理
train_data = train_data.reshape([len(train_data),28*28])
test_data = test_data.reshape([len(test_data),28*28]) 
ax = plt.figure().add_subplot()
ax.imshow(train_data[0].reshape([28,28]),cmap = ‘bone‘)
def label_to_one_hot(labels):
    l = np.zeros([len(labels),10])
    for i in range(len(labels)):
        l[i,labels[i]]=1
    return l
train_labels = label_to_one_hot(train_labels)
test_labels = label_to_one_hot(test_labels)
#%%定义模型
from keras import models,layers,losses,optimizers,metrics
def build_model():
    model = models.Sequential()
    model.add(layers.Dense(28*28,activation = ‘relu‘,input_shape = (len(train_data[0]),)))
    #model.add(layers.Dense(100,activation=‘relu‘))
    model.add(layers.Dense(10,activation=‘softmax‘))
    model.compile(optimizer=optimizers.RMSprop(lr=0.001),
        loss = losses.categorical_crossentropy,
        metrics=[metrics.accuracy])
    return model
model = build_model()
#%%测试验证参数和最优迭代次数
x_train = train_data[10000:]
y_train = train_labels[10000:]
x_val = test_data[:10000]
y_val = test_labels[:10000]
history = model.fit(x_train,y_train,512,80,validation_data=(x_val,y_val))
#%%绘制准确率图
print(history.history.keys())
acc = history.history[‘accuracy‘]
acc_val = history.history[‘val_accuracy‘] 
ax = plt.figure().add_subplot() 
ax.plot(range(len(acc)),acc)
ax.plot(range(len(acc)),acc_val)
#%%在整个训练集上重新训练
model_train = build_model()
model_train.fit(train_data,train_labels,512,50)
#%%预测测试集
predict_labels = model_train.predict(test_data)
print("预测标签：")
print(predict_labels)
evaluate = model_train.evaluate(test_data,test_labels)
print("损失值与准确率：")
print(evaluate)