'Tensor'對象沒有屬性'_keras_history'Keras沒有Tensorflow張量

Question

這段代碼：

a = Input(ish)
for i in range(a.shape[1]):
  x=Conv2D(filters=50, kernel_size=3, padding='same', activation=rl)(a[:,i])
  x=MaxPooling2D(pool_size=2)(x)
  x=Dropout(0.5)(x)
  x=Conv2D(filters=100, kernel_size=5, padding='same', activation=rl)(x)
  x=MaxPooling2D(pool_size=2)(x)
  x=Dropout(0.5)(x)
  x=Conv2D(filters=200, kernel_size=7, padding='same', activation=rl)(x)
  x=MaxPooling2D(pool_size=2)(x)
t=Flatten()(x)
t=Dropout(0.7)(t)
b=Dense(num_classes, activation='softmax')(t)

model = Model(inputs=a, outputs=b)

在最后一行中，給出此錯誤：

AttributeError: 'Tensor' object has no attribute '_keras_history'

您知道導致問題的原因是什么，如何解決？

Answer 1

為了保留Keras元數據，您必須在Lambda層中進行索引編制：

a = Input(ish)
for i in range(a.shape[1]):
  x=Lambda(lambda x: x[:, i])(a)
  x=Conv2D(filters=50, kernel_size=3, padding='same', activation=rl)(x)
  x=MaxPooling2D(pool_size=2)(x)
  x=Dropout(0.5)(x)
  x=Conv2D(filters=100, kernel_size=5, padding='same', activation=rl)(x)
  x=MaxPooling2D(pool_size=2)(x)
  x=Dropout(0.5)(x)
  x=Conv2D(filters=200, kernel_size=7, padding='same', activation=rl)(x)
  x=MaxPooling2D(pool_size=2)(x)
t=Flatten()(x)
t=Dropout(0.7)(t)
b=Dense(num_classes, activation='softmax')(t)

model = Model(inputs=a, outputs=b)