输入和输出均为张量，它们都可以用来定义一个模型（Model），这样的模型同 Keras 的 Sequential 模型一样，都可以被训练。
1.建立Model
from keras.layers import Input,Dense，TimeDistributed,Embedding,LSTM,contatenate，Maxpooling2D，Flatten
from keras.models import Model
inputs=Input(shape=(784,))
x=Dense(64,activation='relu')(inputs)
x=Dense(64,activation='relu')(x)
out=Dense(10,activation='softmax')(x)
x=Embedding(output_dim=512,input_dim=1000,input_length=100)(x）
lstm_out=LSTM(32)(x)
x=keras.layers.concatenate([lstm,x],axis=-1)
x=MaxPooling2D((3,3),strides=(1,1),padding='same')(x)
z=keras.layers.add([x,y])#残差网络
x=Flatten()(x)
model = Model(inputs=[a1, a2], outputs=[b1, b3, b3])
2.编译
model=Model(inputs=inputs,outputs=out)
processed_sequences=TimeDistributed(model)(input_sequences)#将图像分类模型转换成为视频分类模型，input_sequences=Input(shape=(时间序列，向量维度))
model.compile(optimizer='rmsprop',loss='categorical_crossentropy',metrics=['accuracy']）
compile(optimizer, loss=None, metrics=None, loss_weights=None, sample_weight_mode=None, weighted_metrics=None, target_tensors=None)
3.训练
model.fit(data,labels)
fit(x=None, y=None, batch_size=None, epochs=1, verbose=1, callbacks=None, validation_split=0.0, validation_data=None, shuffle=True, class_weight=None, sample_weight=None, initial_epoch=0, steps_per_epoch=None, validation_steps=None)
4.评估
evaluate(x=None, y=None, batch_size=None, verbose=1, sample_weight=None, steps=None)
5.预测、
predict(x, batch_size=None, verbose=0, steps=None)
a.数据并行(串)
from keras.utils import mulit_gpu_model
parallel_model=multi_gpu_model(model,gpu=8)#8个gpu并运的model。
b.设备并行（并）
with tf.device_scope('/gpu:0'):
encode_a=Lstm(x1)#在一个GPU上处理第一个序列
with tf.device_scope('/gpu:1'):
encode_a=Lstm(x2)#在另一个GPU上处理另一个序列
with tf.device_scope('/cpu:0'):
merged_vector=keras.layers.concatenate([encode_a,encode_b],axis=-1)#在cpu上连接结果
c.保存并重载模型(结构权重，优化器状态)
from keras.models import load_model
model.save('my.h5')
del model
model= load_model('my.h5')
d。只保存加载模型结构
json_string=model.to_json()#保存为JSON结构
yaml_string=model.to_yaml()#保存为YAML结构
from keras.models import model_from_json，model_from_yaml
重建模型：model=model_from_json(json_string)
model=model_from_yaml(yaml_json)
e.只保存加载模型权重
model.save_weights('weights.h5')
model.load_weights('weights.h5',by_name=True)#true时，就是将名字一样的层的权重加载，不一样的层不加载
f.批量训练预测数据
model.train_on_batch(x,y)
model.test_on_batch(x,y)
g.验证集误差不再下降的早停;
from keras.callbacks import EarlyStopping
early_stopping = EarlyStopping(monitor='val_loss', patience=2)
model.fit(x, y, validation_split=0.2, callbacks=[early_stopping])
h.冻结释冻模型参数
layer.trainable=True/False
I.图形模型：
from keras.applications.xception import Xception
from keras.applications.vgg16 import VGG16
from keras.applications.vgg19 import VGG19
from keras.applications.resnet50 import ResNet50
from keras.applications.inception_v3 import InceptionV3
from keras.applications.inception_resnet_v2 import InceptionResNetV2
from keras.applications.mobilenet import MobileNet
from keras.applications.densenet import DenseNet121
from keras.applications.densenet import DenseNet169
from keras.applications.densenet import DenseNet201
from keras.applications.nasnet import NASNetLarge
from keras.applications.nasnet import NASNetMobile
from keras.applications.mobilenet_v2 import MobileNetV2
model = VGG16(weights='imagenet', include_top=True)
j.model.summary() 打印出模型概述信息
model.get_config() 返回包含模型配置信息的字典
k.置换输入维度
keras.layers.Permute(dims)
L.将任意表达式封装成layer对象
keras.layers.Lambda(function, output_shape=None, mask=None, arguments=None)
m.keras.layers.UpSampling2D(size=(2, 2), data_format=None, interpolation='nearest')
keras.layers.ZeroPadding2D(padding=(1, 1), data_format=None)
keras.layers.MaxPooling2D(pool_size=(2, 2), strides=None, padding='valid', data_format=None)
keras.layers.AveragePooling2D(pool_size=(2, 2), strides=None, padding='valid', data_format=None)
全局最大池化keras.layers.GlobalMaxPooling2D(data_format=None)
全局平均池化keras.layers.GlobalAveragePooling2D(data_format=None)