Update app.py

app.py

@@ -80,12 +80,32 @@ def load_yolo_from_hub():
     return YOLO(w)
 
 def load_autogluon_tabular_from_hub():
+    """Download and load AutoGluon predictor, removing any FastAI submodels."""
     z = hf_hub_download(repo_id=AG_REPO_ID, filename=AG_ZIP_NAME, cache_dir=CACHE_DIR)
     extract_dir = CACHE_DIR / "ag_predictor_native"
-    if extract_dir.exists(): shutil.rmtree(extract_dir)
-    with zipfile.ZipFile(z, "r") as zip_ref: zip_ref.extractall(extract_dir)
+    if extract_dir.exists():
+        shutil.rmtree(extract_dir)
+    with zipfile.ZipFile(z, "r") as zip_ref:
+        zip_ref.extractall(extract_dir)
+
+    # --- delete fastai models before loading to avoid deserialization errors ---
+    fastai_dirs = list(extract_dir.rglob("*fastai*"))
+    for p in fastai_dirs:
+        try:
+            if p.is_dir():
+                shutil.rmtree(p)
+            else:
+                p.unlink()
+        except Exception as e:
+            print(f"[WARN] Could not remove {p}: {e}")
+    print(f"[CLEANUP] Removed {len(fastai_dirs)} FastAI model files.")
+
+    # Now load normally (no version check)
+    from autogluon.tabular import TabularPredictor
+    predictor = TabularPredictor.load(str(extract_dir), require_py_version_match=False)
     print(f"[INFO] Loaded AutoGluon predictor from {extract_dir}")
-    return TabularPredictor.load(str(extract_dir), require_py_version_match=False)
+    return predictor
+
 
 
 _YOLO = None
@@ -191,39 +211,16 @@ def predict_scores(df):
     feats = ["red_ratio", "green_ratio", "red_diff", "green_diff", "z_red", "z_green"]
     X = df[feats].copy()
     ag = ag_predictor()
-
-    # List all models and exclude any FastAI learners (which can cause tabular DL issues)
-    try:
-        available_models = ag.model_names()
-    except Exception:
-        available_models = ag.model_names
-    safe_models = [m for m in available_models if "fastai" not in m.lower()]
-    print(f"[INFO] Using safe models: {safe_models}")
-
-    # Access AutoGluon's internal learner, which supports the `models` argument
-    learner = ag._learner
-
     try:
-        # Get probabilities from the safe models only
-        proba = learner.predict_proba(X, models=safe_models)
+        proba = ag.predict_proba(X)
         if isinstance(proba, pd.DataFrame) and (1 in proba.columns):
             return proba[1]
     except Exception as e:
-        print("[WARN] FastAI submodel failed — retrying with safe models only:", e)
-        proba = learner.predict_proba(X, models=safe_models)
-        if isinstance(proba, pd.DataFrame) and (1 in proba.columns):
-            return proba[1]
-
-    # Fallback if probabilities aren’t available
-    try:
-        preds = learner.predict(X, models=safe_models)
+        print("[WARN] Predictor .predict_proba() failed, falling back:", e)
+        preds = ag.predict(X)
         s = pd.Series(preds).astype(float)
         rng = (s.quantile(0.95) - s.quantile(0.05)) or 1.0
         return ((s - s.quantile(0.05)) / rng).clip(0, 1)
-    except Exception as e:
-        print("[ERROR] Predictor fallback failed:", e)
-        return pd.Series(np.zeros(len(df)))
-
 
 def pick_events(df,score,fps):
     z=rolling_z(score,45); strong=(z>4.0); keep=strong.rolling(3,min_periods=1).sum()>=2