Internal change

PiperOrigin-RevId: 465127497

official/projects/yt8m/eval_utils/eval_util.py

@@ -103,8 +103,8 @@ def top_k_by_class(predictions, labels, k=20):
   Args:
     predictions: A numpy matrix containing the outputs of the model. Dimensions
       are 'batch' x 'num_classes'.
-    labels: A numpy matrix containing the ground truth labels.
-        Dimensions are 'batch' x 'num_classes'.
+    labels: A numpy matrix containing the ground truth labels. Dimensions are
+      'batch' x 'num_classes'.
     k: the top k non-zero entries to preserve in each prediction.
 
   Returns:
@@ -143,9 +143,10 @@ def top_k_triplets(predictions, labels, k=20):
   Args:
     predictions: A numpy matrix containing the outputs of the model. Dimensions
       are 'batch' x 'num_classes'.
-    labels: A numpy matrix containing the ground truth labels.
-        Dimensions are 'batch' x 'num_classes'.
+    labels: A numpy matrix containing the ground truth labels. Dimensions are
+      'batch' x 'num_classes'.
     k: The number top predictions to pick.
+
   Returns:
     a sparse list of tuples in (prediction, class) format.
   """
@@ -175,7 +176,7 @@ class EvaluationMetrics(object):
     self.sum_hit_at_one = 0.0
     self.sum_perr = 0.0
     self.map_calculator = map_calculator.MeanAveragePrecisionCalculator(
-        num_class, top_n=top_n)
+        num_class, filter_empty_classes=False, top_n=top_n)
     self.global_ap_calculator = ap_calculator.AveragePrecisionCalculator()
     self.top_k = top_k
     self.num_examples = 0
@@ -217,9 +218,13 @@ class EvaluationMetrics(object):
 
     return {"hit_at_one": mean_hit_at_one, "perr": mean_perr}
 
-  def get(self):
+  def get(self, return_per_class_ap=False):
     """Calculate the evaluation metrics for the whole epoch.
 
+    Args:
+      return_per_class_ap: a bool variable to determine whether return the
+        detailed class-wise ap for more detailed analysis. Default is `False`.
+
     Raises:
       ValueError: If no examples were accumulated.
 
@@ -243,6 +248,10 @@ class EvaluationMetrics(object):
         "map": mean_ap,
         "gap": gap
     }
+
+    if return_per_class_ap:
+      epoch_info_dict["per_class_ap"] = aps
+
     return epoch_info_dict
 
   def clear(self):