Keras - Custom Loss Function - ValueError: No gradients provided for any variable

Question

I'm trying to create a custom loss function for my model. I have remove all numpy, and used only keras function to create my custom loss.

def kweighted_accuracy(y_true, y_pred):
    y_true = tf.cast(y_true, tf.float32)
    y_pred = tf.cast(y_pred, tf.float32)
    
    y_abs = K.abs(y_true)
    norm = K.sum(y_abs)
    v = tf.equal(K.sign(y_pred),K.sign(y_true))
    s = tf.boolean_mask(y_abs,v)
    score = K.sum(s) / norm
    return 1-score

Despite that I encounter the error : ValueError: No gradients provided for any variable.

I really don't understand why. Is it the tf.boolean_mask, or tf.equal which I should not use ?

Thanks for the help,

regards,

Alex