Add MetricsContainer.weighted_metrics and MetricsContainer.unweighted_metrics

properties to help distinguish between metrics that should and shouldn't be
passed sample_weight argument.
Note these properties are set to None before Model.fit is called, since metrics
are potentially broadcast to match the structure of data seen in Model.fit.

PiperOrigin-RevId: 339892649
Change-Id: I0abffae08efde2b8adc58014ef205d318d66a9ab

tensorflow/python/keras/engine/compile_utils.py

@@ -292,11 +292,25 @@ class MetricsContainer(Container):
 
   @property
   def metrics(self):
-    """Metrics created by this container."""
+    """All metrics in this container."""
     if not self._built:
       return []
     return self._metrics_in_order
 
+  @property
+  def unweighted_metrics(self):
+    """Metrics in this container that should not be passed `sample_weight`."""
+    if not self._built:
+      return None
+    return nest.flatten(self._metrics)
+
+  @property
+  def weighted_metrics(self):
+    """Metrics in this container that should be passed `sample_weight`."""
+    if not self._built:
+      return None
+    return nest.flatten(self._weighted_metrics)
+
   def build(self, y_pred, y_true):
     """One-time setup of metric objects."""
     super(MetricsContainer, self).build(y_pred)

tensorflow/python/keras/engine/compile_utils_test.py

@@ -420,6 +420,18 @@ class MetricsContainerTest(keras_parameterized.TestCase):
     self.assertEqual(acc_metric_2.result().numpy(), 0.)
     self.assertEqual(acc_metric_2._fn, metrics_mod.binary_accuracy)
 
+    weighted_metrics = metric_container.weighted_metrics
+    self.assertLen(weighted_metrics, 2)
+    self.assertEqual(weighted_metrics[0].name, 'output_1_accuracy')
+    self.assertEqual(weighted_metrics[1].name, 'output_2_accuracy')
+
+    unweighted_metrics = metric_container.unweighted_metrics
+    self.assertLen(unweighted_metrics, 4)
+    self.assertEqual(unweighted_metrics[0].name, 'output_1_mse')
+    self.assertEqual(unweighted_metrics[1].name, 'output_1_mae')
+    self.assertEqual(unweighted_metrics[2].name, 'output_2_mse')
+    self.assertEqual(unweighted_metrics[3].name, 'output_2_mae')
+
   def test_metric_dict(self):
     metric_container = compile_utils.MetricsContainer(
         metrics={