fix

source/triplet_selection.py

@@ -115,7 +115,7 @@ def select_triplets_with_threshold(label_distances, positive_threshold, negative
 #         create_positive_mask(positive_indices, batch_size),
 #         create_negative_mask(negative_indices, batch_size))
 
-def select_random_triplets(label_distances, num_elements, positive_threshold=0.8, negative_threshold=0.2):
+def select_random_triplets(label_distances, num_elements, positive_threshold=0.7, negative_threshold=0.3):
     batch_size = len(label_distances)
 
     base_mask = np.ndarray((batch_size, batch_size, batch_size), dtype=bool)
@@ -150,9 +150,9 @@ def select_random_triplets(label_distances, num_elements, positive_threshold=0.8
             # negative_mask_temp[a, :, n] = label_distances[a, n] >= negative_threshold
             if label_distances[a, n] > negative_threshold:
                 neg_indices.append(n)
-        import ipdb; ipdb.set_trace()
-        positive_mask[a, np.random.choice(pos_indices, size=num_elements, replace=False), :] = True
-        negative_mask[a, :, np.random.choice(neg_indices, size=num_elements, replace=False)] = True
+        # import ipdb; ipdb.set_trace()
+        positive_mask[a, np.random.choice(pos_indices, size=num_elements, replace=True), :] = True
+        negative_mask[a, :, np.random.choice(neg_indices, size=num_elements, replace=True)] = True
 
     return tf.cast(tf.constant(np.logical_and(positive_mask, negative_mask)), tf.bool)