add average_on_enroll argument to Distance

bob/bio/base/pipelines/vanilla_biometrics/biometric_algorithms.py

@@ -10,11 +10,12 @@ import functools
 
 class Distance(BioAlgorithm):
     def __init__(
-        self, distance_function=scipy.spatial.distance.cosine, factor=-1, **kwargs
+        self, distance_function=scipy.spatial.distance.cosine, factor=-1, average_on_enroll=True, **kwargs
     ):
         super().__init__(**kwargs)
         self.distance_function = distance_function
         self.factor = factor
+        self.average_on_enroll = average_on_enroll # if True average of features is calculated, if False average of scores is calculated
 
     def _make_2d(self, X):
         """
@@ -58,7 +59,7 @@ class Distance(BioAlgorithm):
         # That dumps vectors in the format `Nx1xd`
         assert enroll_features.ndim == 2
 
-        return np.mean(enroll_features, axis=0)
+        return np.mean(enroll_features, axis=0) if self.average_on_enroll else enroll_features
 
     def score(self, biometric_reference, data):
         """score(model, probe) -> float
@@ -91,7 +92,9 @@ class Distance(BioAlgorithm):
         assert data.ndim == 2
 
         # return the negative distance (as a similarity measure)
-        return self.factor * self.distance_function(biometric_reference, data)
+        scores = self.factor * self.distance_function(biometric_reference, data)
+        
+        return scores if self.average_on_enroll else np.mean(scores)
 
     def score_multiple_biometric_references(self, biometric_references, data):
 
@@ -106,39 +109,4 @@ class Distance(BioAlgorithm):
         references_stacked = np.vstack(biometric_references)
         scores = self.factor * cdist(references_stacked, data, self.distance_function)
 
-        return scores
-
-
-class Distance2(Distance):
-    def __init__(
-        self, distance_function=scipy.spatial.distance.cosine, factor=-1, **kwargs
-    ):
-        super().__init__(distance_function=distance_function, factor=factor, **kwargs)
-    
-    def enroll(self, enroll_features):
-        """enroll(enroll_features) -> model
-
-        Enrolls the model by storing all given input vectors.
-
-        Parameters
-        ----------
-
-        ``enroll_features`` : [:py:class:`numpy.ndarray`]
-          The list of projected features to enroll the model from.
-
-        Returns
-        -------
-
-        ``model`` : 2D :py:class:`numpy.ndarray`
-          The enrolled model.
-        """
-
-        enroll_features = check_array(enroll_features, allow_nd=True, ensure_2d=True)
-
-        enroll_features = self._make_2d(enroll_features)
-
-        # This avoids some possible mistakes in the feature extraction
-        # That dumps vectors in the format `Nx1xd`
-        assert enroll_features.ndim == 2
-
-        return enroll_features
\ No newline at end of file
+        return scores
\ No newline at end of file