训练模型:keras非Sequential模型的保存加载再训练和预测

import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import *
from tensorflow.keras.layers import *

1、保存
用model.save_weights(’/d￥￥￥￥t888.h5’)

2、加载再次训练模型
加载和预测都需要先有原来模型的结构

from tensorflow.keras import backend as K 

def create_model():
    i1 = Input(shape=(200,), dtype='int32')
    i2 = Input(shape=(200,), dtype='int32')

    x1 = Embedding(len(embeddings_matrix_creative_id),200,weights=[embeddings_matrix_creative_id],trainable=False)(i1)
    x2 = Embedding(len(embeddings_matrix_advertiser_id),150,weights=[embeddings_matrix_advertiser_id],trainable=False)(i2)

    x1 = Bidirectional(LSTM(250, return_sequences=True))(x1)
    # x1 = Attention()([x1,x1])
    x2 = Bidirectional(LSTM(250, return_sequences=True))(x2)
    # x2 = Attention()([x2,x2])

    # att = Attention()([gru,gru])

    e = Dot(axes=2)([x1, x2])
    e1 = Softmax(axis=2)(e)
    e2 = Softmax(axis=1)(e)
    e1 = Lambda(K.expand_dims, arguments={'axis' : 3})(e1)
    e2 = Lambda(K.expand_dims, arguments={'axis' : 3})(e2)

    _x1 = Lambda(K.expand_dims, arguments={'axis' : 1})(x2)
    _x1 = Multiply()([e1, _x1])
    _x1 = Lambda(K.sum, arguments={'axis' : 2})(_x1)
    _x2 = Lambda(K.expand_dims, arguments={'axis' : 2})(x1)
    _x2 = Multiply()([