add_data metadata change

Author: czaloom

src/valor_lite/classification/evaluator.py

@@ -174,13 +174,13 @@ def _create_rocauc_intermediate(
             sink=self._intermediate_writer,
             batch_size=batch_size,
             sorting=[
-                ("score", "descending"),
+                ("pd_score", "descending"),
                 # ("pd_label_id", "ascending"),
                 ("match", "ascending"),
             ],
             columns=[
                 "pd_label_id",
-                "score",
+                "pd_score",
                 "match",
             ],
         )
@@ -194,7 +194,7 @@ def _create_rocauc_intermediate(
         for tbl in intermediate.iterate_tables():
             pd_label_ids = tbl["pd_label_id"].to_numpy()
             tps = tbl["match"].to_numpy()
-            scores = tbl["score"].to_numpy()
+            scores = tbl["pd_score"].to_numpy()
             fps = ~tps
 
             for idx in index_to_label.keys():
@@ -286,7 +286,7 @@ def finalize(
         # sort in-place and locally
         self._writer.sort_by(
             [
-                ("score", "descending"),
+                ("pd_score", "descending"),
                 ("datum_id", "ascending"),
                 ("gt_label_id", "ascending"),
                 ("pd_label_id", "ascending"),
@@ -497,8 +497,8 @@ def iterate_values(self):
             "datum_id",
             "gt_label_id",
             "pd_label_id",
-            "score",
-            "winner",
+            "pd_score",
+            "pd_winner",
             "match",
         ]
         for tbl in self._reader.iterate_tables(columns=columns):
@@ -512,8 +512,8 @@ def iterate_values(self):
                     ]
                 ]
             )
-            scores = tbl["score"].to_numpy()
-            winners = tbl["winner"].to_numpy()
+            scores = tbl["pd_score"].to_numpy()
+            winners = tbl["pd_winner"].to_numpy()
             matches = tbl["match"].to_numpy()
             yield ids, scores, winners, matches
 
@@ -529,8 +529,8 @@ def iterate_values_with_tables(self):
                     ]
                 ]
             )
-            scores = tbl["score"].to_numpy()
-            winners = tbl["winner"].to_numpy()
+            scores = tbl["pd_score"].to_numpy()
+            winners = tbl["pd_winner"].to_numpy()
             matches = tbl["match"].to_numpy()
             yield ids, scores, winners, matches, tbl
 

src/valor_lite/classification/loader.py

@@ -74,19 +74,20 @@ def add_data(
                 pidx = self._add_label(plabel)
                 rows.append(
                     {
+                        # metadata
+                        **datum_metadata,
                         # datum
                         "datum_uid": classification.uid,
                         "datum_id": self._datum_count,
-                        **datum_metadata,
                         # groundtruth
                         "gt_label": classification.groundtruth,
                         "gt_label_id": gidx,
                         # prediction
                         "pd_label": plabel,
                         "pd_label_id": pidx,
+                        "pd_score": float(score),
+                        "pd_winner": max_score_idx == idx,
                         # pair
-                        "score": float(score),
-                        "winner": max_score_idx == idx,
                         "match": (gidx == pidx) and pidx >= 0,
                     }
                 )

src/valor_lite/classification/shared.py

@@ -45,9 +45,9 @@ def generate_schema(
         # prediction
         ("pd_label", pa.string()),
         ("pd_label_id", pa.int64()),
+        ("pd_score", pa.float64()),
+        ("pd_winner", pa.bool_()),
         # pair
-        ("score", pa.float64()),
-        ("winner", pa.bool_()),
         ("match", pa.bool_()),
     ]
 
@@ -65,7 +65,7 @@ def generate_intermediate_schema() -> pa.Schema:
     return pa.schema(
         [
             ("pd_label_id", pa.int64()),
-            ("score", pa.float64()),
+            ("pd_score", pa.float64()),
             ("match", pa.bool_()),
         ]
     )