Add BinaryAUC metric and Best Checkpoint callback to Text Classifier

PiperOrigin-RevId: 581276382

mediapipe/model_maker/python/core/utils/metrics.py

@@ -94,6 +94,23 @@ def _get_sparse_metric(metric: tf.metrics.Metric):
   return SparseMetric
 
 
+class BinaryAUC(tf.keras.metrics.AUC):
+  """A Binary AUC metric for binary classification tasks.
+
+  For update state, the shapes of y_true and y_pred are expected to be:
+    - y_true: [batch_size x 1] array of 0 for negatives and 1 for positives
+    - y_pred: [batch_size x 2] array of probabilities where y_pred[:,0] are the
+      probabilities of the 0th(negative) class and y_pred[:,1] are the
+      probabilities of the 1st(positive) class
+
+  See https://www.tensorflow.org/api_docs/python/tf/keras/metrics/AUC for
+    details.
+  """
+
+  def update_state(self, y_true, y_pred, sample_weight=None):
+    super().update_state(y_true, y_pred[:, 1], sample_weight)
+
+
 SparseRecall = _get_sparse_metric(tf.metrics.Recall)
 SparsePrecision = _get_sparse_metric(tf.metrics.Precision)
 BinarySparseRecallAtPrecision = _get_binary_sparse_metric(

mediapipe/model_maker/python/core/utils/metrics_test.py

@@ -14,6 +14,7 @@
 
 
 from absl.testing import parameterized
+import numpy as np
 import tensorflow as tf
 
 from mediapipe.model_maker.python.core.utils import metrics
@@ -23,16 +24,15 @@ class SparseMetricTest(tf.test.TestCase, parameterized.TestCase):
 
   def setUp(self):
     super().setUp()
-    self.y_true = [0, 0, 1, 1, 0, 1]
-    self.y_pred = [
+    self.y_true = np.array([0, 0, 1, 1, 0, 1])
+    self.y_pred = np.array([
         [0.9, 0.1],  # 0, 0 y
         [0.8, 0.2],  # 0, 0 y
         [0.7, 0.3],  # 0, 1 n
         [0.6, 0.4],  # 0, 1 n
         [0.3, 0.7],  # 1, 0 y
         [0.3, 0.7],  # 1, 1 y
-    ]
-    self.num_classes = 3
+    ])
 
   def _assert_metric_equals(self, metric, value):
     metric.update_state(self.y_true, self.y_pred)
@@ -69,6 +69,10 @@ class SparseMetricTest(tf.test.TestCase, parameterized.TestCase):
     ):
       _ = metrics.BinarySparsePrecisionAtRecall(1.0, class_id=2)
 
+  def test_binary_auc(self):
+    metric = metrics.BinaryAUC(num_thresholds=1000)
+    self._assert_metric_equals(metric, 0.7222222)
+
 
 if __name__ == '__main__':
   tf.test.main()

mediapipe/model_maker/python/text/text_classifier/text_classifier.py

@@ -372,9 +372,19 @@ class _BertClassifier(TextClassifier):
   ):
     super().__init__(model_spec, label_names, hparams.shuffle)
     self._hparams = hparams
-    self._callbacks = model_util.get_default_callbacks(
-        self._hparams.export_dir, self._hparams.checkpoint_frequency
-    )
+    self._callbacks = list(
+        model_util.get_default_callbacks(
+            self._hparams.export_dir, self._hparams.checkpoint_frequency
+        )
+    ) + [
+        tf.keras.callbacks.ModelCheckpoint(
+            os.path.join(self._hparams.export_dir, "best_model"),
+            monitor="val_auc",
+            mode="max",
+            save_best_only=True,
+            save_weights_only=False,
+        )
+    ]
     self._model_options = model_options
     self._text_preprocessor: preprocessor.BertClassifierPreprocessor = None
     with self._hparams.get_strategy().scope():
@@ -465,6 +475,7 @@ class _BertClassifier(TextClassifier):
         ),
         metrics.SparsePrecision(name="precision", dtype=tf.float32),
         metrics.SparseRecall(name="recall", dtype=tf.float32),
+        metrics.BinaryAUC(name="auc", num_thresholds=1000),
     ]
     if self._num_classes == 2:
       if self._hparams.desired_precisions:

mediapipe/model_maker/python/text/text_classifier/text_classifier_test.py

@@ -230,6 +230,7 @@ class TextClassifierTest(tf.test.TestCase, parameterized.TestCase):
         'accuracy',
         'recall',
         'precision',
+        'auc',
         'precision_at_recall_0.2',
         'recall_at_precision_0.9',
     ]