python神经网络使用Keras进行模型的保存与读取

pip install h5py

model.save("./model.hdf5")

model = load_model("./model.hdf5")

import numpy as np
from keras.models import Sequential,load_model,save_model
from keras.layers import Dense,Activation ## 全连接层
from keras.datasets import mnist
from keras.utils import np_utils
from keras.optimizers import RMSprop
# 获取训练集
(X_train,Y_train),(X_test,Y_test) = mnist.load_data()
# 首先进行标准化 
X_train = X_train.reshape(X_train.shape[0],-1)/255
X_test = X_test.reshape(X_test.shape[0],-1)/255
# 计算cateGorical_crossentropy需要对分类结果进行categorical
# 即需要将标签转化为形如(nb_samples, nb_classes)的二值序列
Y_train = np_utils.to_categorical(Y_train,num_classes= 10)
Y_test = np_utils.to_categorical(Y_test,num_classes= 10)
# 构建模型
model = Sequential([
    Dense(32,input_dim = 784),
    Activation("relu"),
    Dense(10),
    Activation("softmax")
    ]
)
rmsprop = RMSprop(lr = 0.001,rho = 0.9,epsilon = 1e-08,decay = 0)
## compile
model.compile(loss = 'categorical_crossentropy',optimizer = rmsprop,metrics=['accuracy'])
print("\ntraining")
cost = model.fit(X_train,Y_train,nb_epoch = 2,batch_size = 100)
print("\nTest")
# 测试
cost,accuracy = model.evaluate(X_test,Y_test)
print("accuracy:",accuracy)
# 保存模型
model.save("./model.hdf5")
# 删除现有模型
del model
print("model had been del")
# 再次载入模型
model = load_model("./model.hdf5")
# 预测
cost,accuracy = model.evaluate(X_test,Y_test)
print("accuracy:",accuracy)

Epoch 1/2
60000/60000 [==============================] - 6s 104us/step - loss: 0.4217 - acc: 0.8888
Epoch 2/2
60000/60000 [==============================] - 6s 99us/step - loss: 0.2240 - acc: 0.9366
Test
10000/10000 [==============================] - 1s 149us/step
accuracy: 0.9419
model had been del
10000/10000 [==============================] - 1s 117us/step
accuracy: 0.9419