Deploy Space: update src

src/utils.py

@@ -59,6 +59,7 @@ RANK_SCORE_BASE_METRICS = [
     "eval_metrics/MSE[mean]",
     "eval_metrics/mean_weighted_sum_quantile_loss",
 ]
+GEOMEAN_EPSILON = 1e-12
 
 VALUE_COLUMNS = ["MSE", "CRPS", RANK_SCORE_COLUMN] + [
     METRIC_LABELS[m] for m in DISPLAY_METRICS if METRIC_LABELS[m] not in {"MSE", "CRPS"}
@@ -163,16 +164,23 @@ LIVE_AGGREGATE_FILES = {
 
 
 def format_number(value):
-    """Format a displayed leaderboard metric with a fixed two-decimal precision."""
+    """Format a displayed leaderboard metric without hiding tiny non-zero values."""
     if pd.isna(value):
-        return np.nan
-    return f"{float(value):.2f}"
+        return "n/a"
+    number = float(value)
+    if not np.isfinite(number):
+        return "n/a"
+    if number != 0.0 and abs(number) < 0.005:
+        return f"{number:.2e}"
+    return f"{number:.2f}"
 
 
 def aggregate_gmean(series: pd.Series) -> float:
-    values = series.dropna()
+    values = pd.to_numeric(series, errors="coerce").dropna()
+    values = values[np.isfinite(values) & (values >= 0.0)]
     if values.empty:
         return np.nan
+    values = values.clip(lower=GEOMEAN_EPSILON)
     return float(stats.gmean(values))
 
 
@@ -262,6 +270,7 @@ def prepare_results_df(
 
     for metric in METRIC_COLUMNS:
         df[metric] = pd.to_numeric(df[metric], errors="coerce")
+        df.loc[~np.isfinite(df[metric]), metric] = np.nan
 
     return _add_per_dataset_ranks(df)
 
@@ -652,6 +661,46 @@ def group_datasets_by_domain(
     return ordered
 
 
+def _parse_timestamp(value: object) -> datetime | None:
+    if not value or value == "n/a":
+        return None
+    try:
+        return datetime.fromisoformat(str(value).replace("Z", "+00:00"))
+    except (TypeError, ValueError):
+        return None
+
+
+def _read_json(path: Path) -> dict:
+    if not path.exists():
+        return {}
+    try:
+        return json.loads(path.read_text())
+    except (OSError, json.JSONDecodeError):
+        return {}
+
+
+def leaderboard_refresh_timestamp(root_dir: str = "results") -> str:
+    """Return the latest timestamp that can change leaderboard-visible results."""
+    root = Path(root_dir)
+    candidates: list[datetime] = []
+
+    status = _read_json(root / "online_status.json")
+    for key in ("finished_at", "pushed_at"):
+        parsed = _parse_timestamp(status.get(key))
+        if parsed is not None:
+            candidates.append(parsed)
+
+    for metadata_name in ("metadata.json", "live_metadata.json"):
+        metadata = _read_json(root / "aggregates" / metadata_name)
+        parsed = _parse_timestamp(metadata.get("generated_at"))
+        if parsed is not None:
+            candidates.append(parsed)
+
+    if not candidates:
+        return "n/a"
+    return format_timestamp_utc8(max(candidates).isoformat())
+
+
 def build_leaderboard_summary_html(root_dir: str = "results") -> str:
     df = prepare_results_df(root_dir)
     if df.empty:
@@ -667,14 +716,7 @@ def build_leaderboard_summary_html(root_dir: str = "results") -> str:
     models = df["model"].nunique()
     datasets = df["dataset"].nunique()
     domains = df["domain"].nunique() if "domain" in df.columns else 0
-    last_refresh = "n/a"
-    status_path = Path(root_dir) / "online_status.json"
-    if status_path.exists():
-        try:
-            status = json.loads(status_path.read_text())
-            last_refresh = format_timestamp_utc8(status.get("finished_at", ""))
-        except Exception:
-            pass
+    last_refresh = leaderboard_refresh_timestamp(root_dir)
 
     return f"""
     <div class="summary-grid">