Deploy Space: update src

src/utils.py

@@ -123,40 +123,15 @@ def load_all_results(root_dir: str = "results") -> pd.DataFrame:
     df = df.sort_values(by=["model", "dataset"]).reset_index(drop=True)
     df["model"] = df["model"].replace(LEGACY_MODEL_NAMES)
     
-    # Dynamically detect evaluated models in root_dir.
-    # A model is valid if its subdirectory contains 'all_results.csv'.
-    detected_models = set()
-    root_path = Path(root_dir)
-    if root_path.exists():
-        for p in root_path.iterdir():
-            if p.is_dir() and (p / "all_results.csv").exists():
-                config_json = p / "config.json"
-                if config_json.exists():
-                    try:
-                        cfg = json.loads(config_json.read_text())
-                        if "model" in cfg:
-                            detected_models.add(cfg["model"])
-                    except Exception:
-                        pass
-                # Read model display names from the CSV file itself as a fallback
-                try:
-                    temp_df = pd.read_csv(p / "all_results.csv", nrows=5)
-                    if "model" in temp_df.columns:
-                        detected_models.update(temp_df["model"].dropna().unique())
-                except Exception:
-                    pass
-
-    # Fallback to defaults if no models are detected
-    if not detected_models:
-        detected_models = {
-            "Toto-2.0-1B",
-            "Moirai-1.1-Large",
-            "Chronos-2",
-            "Timer-S1",
-            "TiRex",
-            "TimesFM-2.5"
-        }
-    df = df[df["model"].isin(detected_models)]
+    ALLOWED_MODELS = {
+        "Toto-2.0-1B",
+        "Moirai-1.1-Large",
+        "Chronos-2",
+        "Timer-S1",
+        "TiRex",
+        "TimesFM-2.5"
+    }
+    df = df[df["model"].isin(ALLOWED_MODELS)]
 
     parts = df["dataset"].astype(str).str.split("/", expand=True)
     if parts.shape[1] < 3:
@@ -663,9 +638,10 @@ def _today_start_utc8() -> datetime:
 
 def get_dataset_last_eval_time(ds_base: str, root_dir: str = "results") -> datetime | None:
     latest_dt: datetime | None = None
+    ALLOWED_SLUGS = {"chronos_2", "moirai_1_1_large", "timer_s1", "timesfm_2_5", "tirex", "toto_2_0_1b"}
     # Scan model directories
     for model_dir in Path(root_dir).iterdir():
-        if not model_dir.is_dir() or not (model_dir / "all_results.csv").exists():
+        if not model_dir.is_dir() or model_dir.name not in ALLOWED_SLUGS:
             continue
         forecasts_dir = model_dir / "forecasts"
         if not forecasts_dir.is_dir():
@@ -717,32 +693,12 @@ def make_domain_pie_chart(root_dir: str = "results") -> object | None:
     today_start = _today_start_utc8()
     utc8 = today_start.tzinfo
 
-    # Read eval_history.jsonl
+    # Read eval_history.jsonl (Cumulative)
     history_path = Path(root_dir) / "eval_history.jsonl"
     timestamp_counts: dict[str, int] = {}  # domain → total timestamps
     domain_datasets: dict[str, set[str]] = {}  # domain → set of dataset short names
 
     if history_path.exists():
-        try:
-            with open(history_path) as f:
-                for line in f:
-                    if not line.strip():
-                        continue
-                    entry = json.loads(line)
-                    eval_at = entry.get("evaluated_at", "")
-                    if eval_at:
-                        eval_dt = datetime.fromisoformat(eval_at.replace("Z", "+00:00"))
-                        if eval_dt.astimezone(utc8) >= today_start:
-                            ds_base = entry.get("dataset", "")
-                            pred_len = entry.get("prediction_length", 0)
-                            domain = get_domain(ds_base)
-                            timestamp_counts[domain] = timestamp_counts.get(domain, 0) + pred_len
-                            domain_datasets.setdefault(domain, set()).add(dataset_short_name(ds_base))
-        except Exception:
-            pass
-
-    # Fallback to cumulative history (no today filter) if today is empty
-    if not timestamp_counts and history_path.exists():
         try:
             with open(history_path) as f:
                 for line in f:
@@ -757,46 +713,12 @@ def make_domain_pie_chart(root_dir: str = "results") -> object | None:
         except Exception:
             pass
 
-    # Fallback to current snapshots count (today-only first)
-    if not timestamp_counts:
-        seen_datasets: set[str] = set()
-        for model_dir in sorted(Path(root_dir).iterdir()):
-            if not model_dir.is_dir() or not (model_dir / "all_results.csv").exists():
-                continue
-            forecasts_dir = model_dir / "forecasts"
-            if not forecasts_dir.is_dir():
-                continue
-            for fp in forecasts_dir.glob("*.json"):
-                try:
-                    snap = json.loads(fp.read_text())
-                except Exception:
-                    continue
-
-                eval_at = snap.get("evaluated_at", "")
-                if eval_at:
-                    try:
-                        eval_dt = datetime.fromisoformat(eval_at.replace("Z", "+00:00"))
-                        if eval_dt.astimezone(utc8) < today_start:
-                            continue
-                    except Exception:
-                        continue
-
-                ds_full = snap.get("dataset", "")
-                ds_base = ds_full.split("/")[0]
-                if ds_base in seen_datasets:
-                    continue
-                seen_datasets.add(ds_base)
-
-                domain = get_domain(ds_base)
-                n_ts = len(snap.get("context", [])) + len(snap.get("actuals", []))
-                timestamp_counts[domain] = timestamp_counts.get(domain, 0) + n_ts
-                domain_datasets.setdefault(domain, set()).add(dataset_short_name(ds_base))
-
-    # Fallback to cumulative snapshots count (no today filter) if still empty
+    # Fallback to cumulative snapshots count (no today filter) if empty
     if not timestamp_counts:
         seen_datasets: set[str] = set()
+        ALLOWED_SLUGS = {"chronos_2", "moirai_1_1_large", "timer_s1", "timesfm_2_5", "tirex", "toto_2_0_1b"}
         for model_dir in sorted(Path(root_dir).iterdir()):
-            if not model_dir.is_dir() or not (model_dir / "all_results.csv").exists():
+            if not model_dir.is_dir() or model_dir.name not in ALLOWED_SLUGS:
                 continue
             forecasts_dir = model_dir / "forecasts"
             if not forecasts_dir.is_dir():
@@ -837,10 +759,9 @@ def make_domain_pie_chart(root_dir: str = "results") -> object | None:
         marker=dict(colors=colors),
         hole=0.25,
     )])
-    date_str = today_start.strftime("%Y-%m-%d")
     fig.update_layout(
         title=dict(
-            text=f"Evaluated timestamps by domain ({date_str}, UTC+8)",
+            text="Evaluated timestamps by domain",
             font=dict(size=16, weight="bold"),
         ),
         margin=dict(l=20, r=20, t=60, b=20),