I want multiple duplicate input layers from the first input layer. So that I don't have to send the input to the fit function twice.

Image

CodePudding user response：

You can reuse the instance of the input layer when creating your two models. I can see in the image that you want to concatenate the output of the two separate layers, so I also included that in my code snippet.

Firstly, I create the input layer. Then I create two sub-models that use the same instance of the input. I stack the output of both sub-models. You can also use tf.concat instead of tf.stack.

import tensorflow as tf
from tensorflow.python.keras import layers
from tensorflow.python.keras import Model


def get_model(input_layer):
    model = tf.keras.Sequential(
        [
            input_layer,
            layers.Dense(32, activation="relu"),
            layers.Dense(32, activation="relu"),
            layers.Dense(1),
        ]
    )
    return model


num_features = 3
input = tf.keras.Input(shape=(num_features,))

model1 = get_model(input)
model2 = get_model(input)

combined_output = tf.stack([model1.output, model2.output], axis=0)

model = Model(inputs=input, outputs=combined_output)

print(tf.shape(model(tf.ones([32, 3]))))

The batch size is 32, and the number of features is 3. The code snippet prints

tf.Tensor([ 2 32  1], shape=(3,), dtype=int32)