Fix Classification Performance

api/tests/functional-tests/backend/metrics/test_classification.py

@@ -219,12 +219,12 @@ def test_compute_confusion_matrix_at_grouper_key(
         models.Annotation.datum_id.label("datum_id"),
         filters=gFilter,
         label_source=models.GroundTruth,
-    ).alias()
+    ).cte()
     predictions = generate_select(
         models.Prediction,
         filters=pFilter,
         label_source=models.Prediction,
-    ).alias()
+    ).cte()
 
     cm = _compute_confusion_matrix_at_grouper_key(
         db=db,
@@ -294,12 +294,12 @@ def test_compute_confusion_matrix_at_grouper_key(
         models.Annotation.datum_id.label("datum_id"),
         filters=gFilter,
         label_source=models.GroundTruth,
-    ).alias()
+    ).cte()
     predictions = generate_select(
         models.Prediction,
         filters=pFilter,
         label_source=models.Prediction,
-    ).alias()
+    ).cte()
 
     cm = _compute_confusion_matrix_at_grouper_key(
         db=db,
@@ -445,12 +445,12 @@ def test_compute_confusion_matrix_at_grouper_key_and_filter(
         models.Annotation.datum_id.label("datum_id"),
         filters=gFilter,
         label_source=models.GroundTruth,
-    ).alias()
+    ).cte()
     predictions = generate_select(
         models.Prediction,
         filters=pFilter,
         label_source=models.Prediction,
-    ).alias()
+    ).cte()
 
     cm = _compute_confusion_matrix_at_grouper_key(
         db,
@@ -595,12 +595,12 @@ def test_compute_confusion_matrix_at_grouper_key_using_label_map(
         models.Annotation.datum_id.label("datum_id"),
         filters=gFilter,
         label_source=models.GroundTruth,
-    ).alias()
+    ).cte()
     predictions = generate_select(
         models.Prediction,
         filters=pFilter,
         label_source=models.Prediction,
-    ).alias()
+    ).cte()
 
     cm = _compute_confusion_matrix_at_grouper_key(
         db,
@@ -1256,32 +1256,44 @@ def test__compute_curves(
         models.Dataset.name.label("dataset_name"),
         filters=gFilter,
         label_source=models.GroundTruth,
-    ).alias()
+    ).cte()
     predictions = generate_select(
         models.Prediction,
+        models.Annotation.datum_id.label("datum_id"),
         models.Dataset.name.label("dataset_name"),
         filters=pFilter,
         label_source=models.Prediction,
-    ).alias()
+    ).cte()
 
     # calculate the number of unique datums
     # used to determine the number of true negatives
 
     gt_datums = generate_query(
+        models.Datum.id,
         models.Dataset.name,
         models.Datum.uid,
         db=db,
         filters=groundtruth_filter,
         label_source=models.GroundTruth,
     ).all()
     pd_datums = generate_query(
+        models.Datum.id,
         models.Dataset.name,
         models.Datum.uid,
         db=db,
         filters=prediction_filter,
         label_source=models.Prediction,
     ).all()
-    unique_datums = set(pd_datums + gt_datums)
+    unique_datums = {
+        datum_id: (dataset_name, datum_uid)
+        for datum_id, dataset_name, datum_uid in gt_datums
+    }
+    unique_datums.update(
+        {
+            datum_id: (dataset_name, datum_uid)
+            for datum_id, dataset_name, datum_uid in pd_datums
+        }
+    )
 
     curves = _compute_curves(
         db=db,

api/valor_api/backend/core/prediction.py

@@ -9,20 +9,30 @@
 def _check_if_datum_has_prediction(
     db: Session, datum: schemas.Datum, model_name: str, dataset_name: str
 ) -> None:
+    """Checks to see if datum has existing annotations."""
     if db.query(
         select(models.Annotation.id)
-        .join(models.Model)
-        .join(models.Datum)
-        .join(models.Dataset)
-        .where(
+        .select_from(models.Annotation)
+        .join(
+            models.Model,
             and_(
-                models.Dataset.name == dataset_name,
-                models.Datum.dataset_id == models.Dataset.id,
-                models.Datum.uid == datum.uid,
+                models.Model.id == models.Annotation.model_id,
                 models.Model.name == model_name,
-                models.Annotation.datum_id == models.Datum.id,
-                models.Annotation.model_id == models.Model.id,
-            )
+            ),
+        )
+        .join(
+            models.Datum,
+            and_(
+                models.Datum.id == models.Annotation.datum_id,
+                models.Datum.uid == datum.uid,
+            ),
+        )
+        .join(
+            models.Dataset,
+            and_(
+                models.Dataset.id == models.Datum.dataset_id,
+                models.Dataset.name == dataset_name,
+            ),
         )
         .subquery()
     ).all():
@@ -119,15 +129,14 @@ def create_predictions(
         for i, annotation in enumerate(prediction.annotations):
             for label in annotation.labels:
                 prediction_mappings.append(
-                    {
-                        "annotation_id": annotation_ids_per_prediction[i],
-                        "label_id": label_dict[(label.key, label.value)],
-                        "score": label.score,
-                    }
+                    models.Prediction(
+                        annotation_id=annotation_ids_per_prediction[i],
+                        label_id=label_dict[(label.key, label.value)],
+                        score=label.score,
+                    )
                 )
-
     try:
-        db.bulk_insert_mappings(models.Prediction, prediction_mappings)
+        db.add_all(prediction_mappings)
         db.commit()
     except IntegrityError as e:
         db.rollback()