Way to utilize `loss_weights` in custom `fit` method!

I’m going through this documents (customizing_what_happens_in_fit by @fchollet), and eventually feel like getting stuck if I want to weighted the loss value. For example, observe the following code

class CustomModel(keras.Model):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.loss_tracker = keras.metrics.Mean(name="loss")
        self.mae_metric = keras.metrics.MeanAbsoluteError(name="mae")

    def train_step(self, data):
        x, y = data

        with tf.GradientTape() as tape:
            y_pred = self(x, training=True)  # Forward pass
            # Compute our own loss
            loss = keras.losses.mean_squared_error(y, y_pred)

        # Compute gradients
        trainable_vars = self.trainable_variables
        gradients = tape.gradient(loss, trainable_vars)

        # Update weights
        self.optimizer.apply_gradients(zip(gradients, trainable_vars))

        # Compute our own metrics
        self.loss_tracker.update_state(loss)
        self.mae_metric.update_state(y, y_pred)
        return {"loss": self.loss_tracker.result(), "mae": self.mae_metric.result()}

    @property
    def metrics(self):
        # We list our `Metric` objects here so that `reset_states()` can be
        # called automatically at the start of each epoch
        # or at the start of `evaluate()`.
        # If you don't implement this property, you have to call
        # `reset_states()` yourself at the time of your choosing.
        return [self.loss_tracker, self.mae_metric]

# Construct an instance of CustomModel
inputs = keras.Input(shape=(32,))
outputs = keras.layers.Dense(1)(inputs)
model = CustomModel(inputs, outputs)

# We don't passs a loss or metrics here.
model.compile(optimizer="adam")

# Just use `fit` as usual -- you can use callbacks, etc.
x = np.random.random((1000, 32))
y = np.random.random((1000, 1))
model.fit(x, y, epochs=5)

In the above, no loss is passed in model.compile method, because, we’re manually computing loss value inside the train_step method. That’s great. Now, let’s assume I’ve more than one output from model and want to weighted them towards the final loss. In such case, what would be the approach to do so?

In model.compile, it has loss_weights arguments that can be used but for above cases, how to achieve this?

Model.compile(
    optimizer="rmsprop",
    loss=None,
    metrics=None,
    loss_weights=None,
    weighted_metrics=None,
    run_eagerly=None,
    steps_per_execution=None,
    jit_compile=None,
    **kwargs
)

In the above mention document, it somewhat discuss about class_weights and sample_weights, but not about loss_weights.

[rchao]

Eventually, I’ve just realized that…

Yes, this approach looks reasonable to me. Only suggestion is to name the attribute .loss_weights or something that incorporates loss in it to not be confused with layers’ weights. Thanks!

[rchao]

Thanks Mohammed!