ValueError：模型的输出张量必须是带有tf.keras Lambda层的TensorFlow层的输出

Question

I'm trying to use the output of a tf.keras.layers.Lambda function as the last layer in a tf.keras model, but tf is interpreting the lambda layer's output as a Tensor(as opposed to a Layer) object.

The error is: "ValueError: Output tensors to a Model must be the output of a TensorFlow Layer (thus holding past layer metadata). Found: Tensor("Discriminator/mullayer/mul:0", shape=(2, 2), dtype=float32)"

and the code is attached below

from tensorflow.contrib.keras import layers, models

#lots of stuff up here, all working fine...

logits = layers.Dense(1, name=name+'fc')(x)# x works fine

mullayer = layers.Lambda(lambda x: x * self.alphaVal,name = "mullayer")
test = tf.constant([1.0],shape = (2,2))
testOut = mullayer(test)

outputs = [logits, testOut]
self.disc = models.Model(inputs=inp, outputs=outputs)

'self.alphaVal' is not a keras variable, just a float, which I suspect may be part of the problem. If so, what is the equivalent of keras' backend K in tf.keras?

Thanks

Answer 1

test is not coming from anywhere considered a Keras layer.

If test intended to be a model's input, it must be:

test = Input(tensor=tf.constant([1.0], shape=(2,2))
#there may be some implications with shape, batch size and other stuff....

As a model with two inputs, you should remember to add it when defining the Model .

If you want a constant value to be used without it being an input, you must not pass it as the "input" of a layer. You just refer to it from inside the layer, or you create it inside the layer.

---

If you just want to test your Lambda layer:

inp = Input((2,2))
out = mullayer(inp)
testModel = Model(inp,out)

testModel.predict(np.ones((1,2,2)))