Support multi-sample promotion evidence

2026-07-09 14:52:06 +02:00
parent f4af3b9eb0
commit 006819ca51
5 changed files with 142 additions and 17 deletions
@@ -109,24 +109,35 @@ def collect_positive_runs(
 ) -> dict[CandidateKey, list[dict[str, Any]]]:
    best_by_candidate_sample: dict[tuple[CandidateKey, str], dict[str, Any]] = {}
    portfolio_model_asset_id = portfolio.get("model_asset_id")
+
+    positive_runs: list[tuple[str, dict[str, Any]]] = []
    for sample in portfolio.get("samples") or []:
        sample_slug = str(sample.get("sample_slug") or "unknown")
        for run in sample.get("runs") or []:
-            key = as_candidate_key(
-                run,
-                fallback_model_asset_id=str(portfolio_model_asset_id) if portfolio_model_asset_id else None,
-                fallback_tile_size=default_tile_size,
-                fallback_tile_overlap=default_tile_overlap,
-            )
-            f1 = numeric(run, "f1_score")
-            if key is None or f1 is None:
-                continue
-            enriched = dict(run)
-            enriched["sample_slug"] = sample_slug
-            existing = best_by_candidate_sample.get((key, sample_slug))
-            existing_f1 = numeric(existing or {}, "f1_score")
-            if existing is None or existing_f1 is None or f1 > existing_f1:
-                best_by_candidate_sample[(key, sample_slug)] = enriched
+            positive_runs.append((sample_slug, run))
+    for item in portfolio.get("items") or []:
+        sample_slug = str(item.get("sample_slug") or "unknown")
+        positive_runs.append((sample_slug, item))
+
+    for sample_slug, run in positive_runs:
+        key = as_candidate_key(
+            run,
+            fallback_model_asset_id=str(portfolio_model_asset_id) if portfolio_model_asset_id else None,
+            fallback_tile_size=default_tile_size,
+            fallback_tile_overlap=default_tile_overlap,
+        )
+        f1 = numeric(run, "f1_score")
+        if f1 is None:
+            f1 = numeric(run, "f1")
+        if key is None or f1 is None:
+            continue
+        enriched = dict(run)
+        enriched["f1_score"] = f1
+        enriched["sample_slug"] = sample_slug
+        existing = best_by_candidate_sample.get((key, sample_slug))
+        existing_f1 = numeric(existing or {}, "f1_score")
+        if existing is None or existing_f1 is None or f1 > existing_f1:
+            best_by_candidate_sample[(key, sample_slug)] = enriched

    grouped: dict[CandidateKey, list[dict[str, Any]]] = defaultdict(list)
    for (key, _sample_slug), run in best_by_candidate_sample.items():