Keras custom loss with dynamic global variable

Question

So, I am trying to write a custom loss function for my keras model. The loss function needs a global variable which changes after every epoch to calculate the loss, But I am not able to get the dynamic loss. with tf.print() it prints the one static value.. so, can anyone point to some resource/solution so that I can use a global variable in the loss function which changes after every epoch. Thank you.

Answer 1

I'm also having the same inquiry. I found multiple references that can address the mentioned issue. However, I'm still struggling in implementing it with my own code.

Solution One (recommended) : Adding a custom loss function by using model.add_loss. For references, you can check these sources:

• Use additional *trainable* variables in Keras/Tensorflow custom loss function (read the last notes as it describes the difference between callbacks and API loss functions
• Adding loss functions as layers: https://tensorflow.google.cn/guide/keras/train_and_evaluate#handling_losses_and_metrics_that_dont_fit_the_standard_signature
• Keras Loss Function with Additional Dynamic Parameter
• https://www.py4u.net/discuss/212977
• https://tensorflow.google.cn/guide/keras/train_and_evaluate#handling_losses_and_metrics_that_dont_fit_the_standard_signature

Solution Two: subclassing the loss function (might not be a good solution for your case)

How to custom losses by subclass tf.keras.losses.Loss class in Tensorflow2.x

• Another Sub-classing example: What is the purpose of the add_loss function in Keras?
• see the below note from this source: What is the purpose of the add_loss function in Keras?

    def loss_carrier(extra_param1, extra_param2):
        def loss(y_true, y_pred):
            #x = complicated math involving extra_param1, extraparam2, y_true, y_pred
            #remember to use tensor objects, so for example keras.sum, keras.square, keras.mean
            #also remember that if extra_param1, extra_maram2 are variable tensors instead of simple floats,
            #you need to have them defined as inputs=(main,extra_param1, extraparam2) in your keras.model instantiation.
            #and have them defind as keras.Input or tf.placeholder with the right shape.
            return x
        return loss

    model.compile(optimizer='adam', loss=loss_carrier