深度学习04-keras实例iris数据集+模型保存调用及修改+在sklearn框架内使用Keras模型+交叉验证

此篇代码接着上一篇，这里查看https://blog.csdn.net/qq_42871249/article/details/104456690

模型的保存与载入

先将模型保存的函数调出来，存贮为checkpoint

from keras.callbacks import ModelCheckpoint

checkpoint = ModelCheckpoint(filepath = r'F:\learning_kecheng\deenlearning\NEW\all_zjwj\iris_best.hdf5',
                             monitor = 'val_acc',
                             save_best_only = True,
                             verbose = 1)

建立模型

先建立个简单的模型

model = Sequential()
model.add(Dense(5,input_dim = 4,activation = 'relu'))
model.add(Dense(3,activation = 'softmax'))#softmax保证输出结果为[0,1]
model.compile(loss = 'categorical_crossentropy',
              optimizer = 'rmsprop',
              metrics = ['accuracy'])
model.fit(irisZX,y,epochs = 50,
          validation_data = (irisZX,y),
          callbacks = [checkpoint])

储存模型

第一个是存模型，包括模型结构。第二个是只存权重，缺少模型结构。

model.save(r'F:\learning_kecheng\deenlearning\NEW\all_zjwj\iris_best.hdf5')
#只保存权重，缺少模型结构
model.save_weights(r'F:\learning_kecheng\deenlearning\NEW\all_zjwj\iris_best_weights.hdf5')

载入模型

载入模型，后面是载入模型的权重

from keras.models import load_model
model = load_model(r'F:\learning_kecheng\deenlearning\NEW\all_zjwj\iris_best.hdf5')

#载入权重，需要已设定相对应的模型框架
model.load_weights(r'F:\learning_kecheng\deenlearning\NEW\all_zjwj\iris_best_weights.hdf5')

模型的修改

原模型

先看一下模型
model.summary()

将模型的第二层权重赋值给a1

a1 = model.layers[1].get_weights()
type(a1)
a1

权重如下

添加模型

在原有模型基础上增加网络层，添加一层

model.add(Dense(3,activation = 'softmax'))
model.summary()

结果如下

拟合匹配

model.compile(loss = 'categorical_crossentropy',
              optimizer = 'rmsprop',
              metrics = ['accuracy'])
model.layers[1].get_weights()

model.fit(irisZX,y,epochs = 50)

在sklearn框架内使用Keras模型

拆分训练集和测试集

from sklearn.model_selection import train_test_split

#拆分训练集和测试集
X_train,X_test,y_train,y_test = train_test_split(irisZX,y,
                                                 test_size = 0.3,random_state = 111)
#默认shuffle = True

定义可被调用的模型函数

def M_model():
    model = Sequential()
    model.add(Dense(5,input_dim = 4,activation = 'relu'))
    model.add(Dense(3,activation = 'softmax'))
    model.compile(loss = 'categorical_crossentropy',
              optimizer = 'rmsprop',
              metrics = ['accuracy'])
    return model

from keras.wrappers.scikit_learn import KerasClassifier

#将所有定义的模型整合入分类器API中
estimator = KerasClassifier(build_fn = M_model,
                            epochs = 50,verbose = 0)
estimator
#fit函数仍然会返回history对象
hist = estimator.fit(X_train,y_train)
hist.history

拟合结果及预测

estimator.score(X_test,y_test)
estimator.predict(X_test)

拟合结果

预测结果

进行交叉验证

from sklearn.model_selection import KFold

kfold = KFold(n_splits = 3,shuffle = True)

from sklearn.model_selection import cross_val_score

results = cross_val_score(estimator,irisZX,y,cv = kfold)

print('模型准确率：%.2f%% (%.2f%%)'%(results.mean()*100,results.std()*100))

来源：CSDN

作者：晓菜成长记

链接：https://blog.csdn.net/qq_42871249/article/details/104465431