Merge pull request #2890 from voxel51/reload

Revalidate stages upon view reload

fiftyone/core/clips.py

@@ -284,7 +284,7 @@ def keep_fields(self):
         super().keep_fields()
 
     def reload(self):
-        """Reloads this view from the source collection in the database.
+        """Reloads the view.
 
         Note that :class:`ClipView` instances are not singletons, so any
         in-memory clips extracted from this view will not be updated by calling
@@ -298,11 +298,14 @@ def reload(self):
         # This assumes that calling `load_view()` when the current clips
         # dataset has been deleted will cause a new one to be generated
         #
-
         self._clips_dataset.delete()
         _view = self._clips_stage.load_view(self._source_collection)
         self._clips_dataset = _view._clips_dataset
 
+        _view = self._base_view
+        for stage in self._stages:
+            _view = _view.add_stage(stage)
+
     def _sync_source_sample(self, sample):
         if not self._classification_field:
             return

fiftyone/core/patches.py

@@ -297,7 +297,7 @@ def keep_fields(self):
         super().keep_fields()
 
     def reload(self):
-        """Reloads this view from the source collection in the database.
+        """Reloads the view.
 
         Note that :class:`PatchView` instances are not singletons, so any
         in-memory patches extracted from this view will not be updated by
@@ -311,11 +311,14 @@ def reload(self):
         # This assumes that calling `load_view()` when the current patches
         # dataset has been deleted will cause a new one to be generated
         #
-
         self._patches_dataset.delete()
         _view = self._patches_stage.load_view(self._source_collection)
         self._patches_dataset = _view._patches_dataset
 
+        _view = self._base_view
+        for stage in self._stages:
+            _view = _view.add_stage(stage)
+
     def _sync_source_sample(self, sample):
         for field in self._label_fields:
             self._sync_source_sample_field(sample, field)

fiftyone/core/video.py

@@ -258,7 +258,7 @@ def keep_fields(self):
         super().keep_fields()
 
     def reload(self):
-        """Reloads this view from the source collection in the database.
+        """Reloads the view.
 
         Note that :class:`FrameView` instances are not singletons, so any
         in-memory frames extracted from this view will not be updated by
@@ -272,11 +272,14 @@ def reload(self):
         # This assumes that calling `load_view()` when the current patches
         # dataset has been deleted will cause a new one to be generated
         #
-
         self._frames_dataset.delete()
         _view = self._frames_stage.load_view(self._source_collection)
         self._frames_dataset = _view._frames_dataset
 
+        _view = self._base_view
+        for stage in self._stages:
+            _view = _view.add_stage(stage)
+
     def _set_labels(self, field_name, sample_ids, label_docs):
         super()._set_labels(field_name, sample_ids, label_docs)
 

fiftyone/core/view.py

@@ -1100,14 +1100,18 @@ def clone(self, name=None, persistent=False):
         )
 
     def reload(self):
-        """Reloads the underlying dataset from the database.
+        """Reloads the view.
 
         Note that :class:`fiftyone.core.sample.SampleView` instances are not
         singletons, so any in-memory samples extracted from this view will not
         be updated by calling this method.
         """
         self._dataset.reload()
 
+        _view = self._base_view
+        for stage in self._stages:
+            _view = _view.add_stage(stage)
+
     def to_dict(
         self,
         rel_dir=None,

tests/unittests/similarity_tests.py

@@ -99,19 +99,32 @@ def test_image_similarity(self):
         query_id = dataset.first().id
 
         view1 = dataset.sort_by_similarity(query_id)
+        values1 = view1.values("id")
+
         view2 = dataset.sort_by_similarity(query_id, reverse=True)
+        values2 = view2.values("id")
 
-        self.assertEqual(
-            view1.values("id"), list(reversed(view2.values("id")))
-        )
+        self.assertListEqual(values2, values1[::-1])
 
         view3 = dataset.sort_by_similarity(query_id, k=4)
+        values3 = view3.values("id")
 
-        self.assertEqual(len(view3), 4)
+        self.assertListEqual(values3, values1[:4])
 
         view4 = dataset.sort_by_similarity(query_id, brain_key="img_sim")
+        values4 = view4.values("id")
+
+        self.assertListEqual(values4, values1)
+
+        view5 = view4.limit(2)
+        values5 = view5.values("id")
+
+        self.assertListEqual(values5, values1[:2])
+
+        view5.reload()
+        values5 = view5.values("id")
 
-        self.assertEqual(view1.values("id"), view4.values("id"))
+        self.assertListEqual(values5, values1[:2])
 
     @drop_datasets
     def test_object_similarity(self):
@@ -163,19 +176,32 @@ def test_object_similarity(self):
         patches = dataset.to_patches("ground_truth")
 
         view1 = patches.sort_by_similarity(query_id)
+        values1 = view1.values("id")
+
         view2 = patches.sort_by_similarity(query_id, reverse=True)
+        values2 = view2.values("id")
 
-        self.assertEqual(
-            view1.values("id"), list(reversed(view2.values("id")))
-        )
+        self.assertListEqual(values2, values1[::-1])
 
         view3 = patches.sort_by_similarity(query_id, k=4)
+        values3 = view3.values("id")
 
-        self.assertEqual(len(view3), 4)
+        self.assertListEqual(values3, values1[:4])
 
         view4 = patches.sort_by_similarity(query_id, brain_key="obj_sim")
+        values4 = view4.values("id")
+
+        self.assertEqual(values4, values1)
+
+        view5 = view4.limit(2)
+        values5 = view5.values("id")
+
+        self.assertListEqual(values5, values1[:2])
+
+        view5.reload()
+        values5 = view5.values("id")
 
-        self.assertEqual(view1.values("id"), view4.values("id"))
+        self.assertListEqual(values5, values1[:2])
 
 
 if __name__ == "__main__":

tests/unittests/view_tests.py

@@ -281,6 +281,35 @@ def test_view_name_readonly(self):
         with self.assertRaises(AttributeError):
             view.name = "new_name"
 
+    @drop_datasets
+    def test_reload(self):
+        dataset = fo.Dataset()
+        dataset.add_samples(
+            [
+                fo.Sample(filepath="image1.jpg", foo="bar"),
+                fo.Sample(filepath="image2.jpg", spam="eggs"),
+                fo.Sample(filepath="image3.jpg"),
+                fo.Sample(filepath="image4.jpg"),
+                fo.Sample(filepath="image5.jpg"),
+            ]
+        )
+
+        view = dataset.take(3).sort_by("filepath").select_fields("foo")
+        sample_ids = view.values("id")
+
+        # Reloading should not cause dataset-independent view stage parameters
+        # like Take's internal random seed to be changed
+        view.reload()
+        same_sample_ids = view.values("id")
+
+        self.assertListEqual(sample_ids, same_sample_ids)
+
+        dataset.delete_sample_field("foo")
+
+        # Field `foo` no longer exists, so validation should fail on reload
+        with self.assertRaises(ValueError):
+            view.reload()
+
 
 class ViewFieldTests(unittest.TestCase):
     @skip_windows  # TODO: don't skip on Windows