How to use keras Input layer in eagerly execution?

  eager-execution, keras, numpy, python, tensorflow

I’m trying have a keras functional API model and I want to use EagerTensor.numpy() method in it, so it’s important to use the model in eager mode. Lets say that I want to run the code below:

input = tensorflow.keras.layers.Input((3,))
x = input.numpy()
y = 2 * x
model = keras.Model(input, y, name = "model")

But it gives me an error saying:

‘Tensor’ object has no attribute ‘numpy’

And thats because Input layer returns a Tensor not an EagerTensor.

keras lets me to build a model in graph mode and then change it to eager mode (in models compile method) but it’s weird that I cant build a model in eager mode and use its benefits (or actually there is a way that I don’t know).

My current solution is to use GradientTape and forget the model. But I’m waiting for a better solution.

Source: Python Questions

LEAVE A COMMENT