[FIXED] Using a `tf.Tensor` as a Python `bool` is not allowed in Graph execution - when I do not use it as a bool

Issue

That is the issue I am getting, but I don't use it as a bool anywhere. Or at least not that I can see. I have read similar questions, they were resolved because there was in fact an instance where it is being treated as a bool. However, I have not used it as a bool.

Functions:

def compile_cnn(model, loss = None, optimizer = None):

    # Compile the CNN using the specified loss function
    model.compile(loss=loss, optimizer=optimizer)

    # return the compiled model
    return model

Custom Keras Loss functions:

def contrastive_loss(label, embedding, margin = 0.4):

    # Assign the label
    y = label

    # Assign the embeddings
    p1 = embedding[0]
    p2 = embedding[1]

    # Get the euclean distance    
    d = tf.norm(p1 - p2, axis=-1)

    if y == 0:
        return (1/2) * math.sqrt(d)
    else:
        return (1/2) * math.sqrt(max(0, (margin-d)))

Calling Code:

# We use Adam as optimizer
optimizer = keras.optimizers.Adam() 

# Compile the model with the contrastive loss function
cont_loss_model = compile_cnn(model, contrastive_loss, optimizer)

Here is the full error message as requested.

 C:\Users\User\anaconda3\lib\site-packages\tensorflow\python\keras\engine\training.py:806 train_function  *
        return step_function(self, iterator)
    C:\Users\User\Documents\Uni\2020\IFN680\Assignment 2\SiameseNetwork.py:88 contrastive_loss  *
        return (1/2) * tf.math.sqrt(max(0, (margin-d)))
    C:\Users\User\anaconda3\lib\site-packages\tensorflow\python\framework\ops.py:877 __bool__  **
        self._disallow_bool_casting()
    C:\Users\User\anaconda3\lib\site-packages\tensorflow\python\framework\ops.py:486 _disallow_bool_casting
        self._disallow_when_autograph_enabled(
    C:\Users\User\anaconda3\lib\site-packages\tensorflow\python\framework\ops.py:472 _disallow_when_autograph_enabled
        raise errors.OperatorNotAllowedInGraphError(

    OperatorNotAllowedInGraphError: using a `tf.Tensor` as a Python `bool` is not allowed: AutoGraph did convert this function. This might indicate you are trying to use an unsupported feature.

Solution

This was solved by adding @tf.function to the custom loss function:

@tf.function
def contrastive_loss(label, embedding, margin = 0.4):

Answered By - Cornelis