keras模型保存和加载

踩坑

不同h5文件来保存，不管是用save和load，虽然写法比较简单，但是经常遇到各种各样的毛病。还有save_weights和load_weights，尽管能用，但是仍然存在问题

最后发现一种最好用的方法

保存

     Model1 =  Net1()
    scce = tf.keras.losses.SparseCategoricalCrossentropy()
    adam = Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.99, epsilon=10e-8, amsgrad=False, name="Aadm")
    Model1.compile(optimizer=adam,loss=scce, metrics=['accuracy'])
    checkpoint_save_path = './/model1//Model1.ckpt'
    cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path, save_weights_only=True)
    Model1.fit(k_x_train,k_y_train,batch_size=64,epochs=30,callbacks=cp_callback)
    predictions_nn1 = Model1.predict(X_test)

加载

	model1 = net.Net1()
	scce = tf.keras.losses.SparseCategoricalCrossentropy()
	adam = Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.99, epsilon=10e-8, amsgrad=False, name="Aadm")
	model1.compile(optimizer=adam,loss=scce, metrics=['accuracy'])
	checkpoint_save_path = './/Model123.ckpt'
	model1.load_weights(checkpoint_save_path)
	predictions_nn = model1.predict(X_test)