In tensorflow.keras, what is the difference between the loss and the metric parameter? code example

Example: keras compile loss

tf.keras.losses.MeanAbsolutePercentageError(
    reduction="auto", name="mean_absolute_percentage_error"
)