python 3.x - Tensorflow: Adding losses of Two models and then apply_gradient method

Question

I am trying to train two models in tensorflow/keras i.e.:

model_r

model_t

on different datasets, then adding theirs losses in such a way

total_loss = loss_r + 0.02 * loss_t

and i want to apply backpropagation by considering total_loss on both models, but unfortunately I am getting Error message

ValueError: No gradients provided for any variable

my code is: '''

    # loss calculate for model_r
    with tf.GradientTape() as tape:

        logits_ref = model_r(x_ref_batch, training=True)
        loss_value_ref = loss_fun_ref(y_ref_batch, logits_ref)
        print(f'loss_value_ref {loss_value_ref}')

    # loss calculate for model_t
    with tf.GradientTape() as tape1:
        logits_t = model_t(x_target_batch, training=True)
        loss_value_t = loss_fun_t(y_target_batch, logits_t)
        print(f'loss_value_t {loss_value_t}')

    loss_total = loss_value_ref + lambdaa * loss_value_t
    print(f'loss_total {loss_total}')

    # gradient for model_a
    grads_r = tape.gradient(loss_total, model_r.trainable_weights)
    # weights update for model_a
    optimizer.apply_gradients(zip(grads_r, model_r.trainable_weights))

    # gradient for model_a
    grads_t = tape1.gradient(loss_total, model_t.trainable_variables)
    # weights update for model_b
    optimizer.apply_gradients(zip(grads_t, model_t.trainable_variables))

'''

and i am getting error message :

''' ValueError: No gradients provided for any variable: ['dense_2/kernel:0', 'dense_2/bias:0', 'dense_3/kernel:0', 'dense_3/bias:0', 'dense_4/kernel:0', 'dense_4/bias:0', 'predictions/kernel:0', 'predictions/bias:0']. '''

Anyone have the idea how to add two losses from two different models, add them and than backpropagate through total loss in tensorflow 2x?

question from:https://stackoverflow.com/questions/65889976/tensorflow-adding-losses-of-two-models-and-then-apply-gradient-method

Answer 1

Waitting for answers