fix

app.py

@@ -16,8 +16,14 @@ from chronos import ChronosPipeline
 AGTS_AVAILABLE = False
 try:
     from autogluon.timeseries import TimeSeriesPredictor, TimeSeriesDataFrame
+    try:
+        # optional: AutoGluon global seeding helper (nicht in allen Versionen vorhanden)
+        from autogluon.common.utils.seed import set_seed as ag_set_seed
+    except Exception:
+        ag_set_seed = None
     AGTS_AVAILABLE = True
 except Exception:
+    ag_set_seed = None
     pass
 
 # unsere bestehende Daten-Pipeline
@@ -51,6 +57,24 @@ AUTO_SEED = int(os.getenv("AUTO_SEED", "0"))
 device = "cuda" if torch.cuda.is_available() else "cpu"
 dtype = torch.bfloat16 if device == "cuda" else torch.float32
 
+# ---- global seeding (kompatibel über Versionen hinweg) ----
+def set_global_seed(seed: int):
+    random.seed(seed)
+    np.random.seed(seed)
+    try:
+        torch.manual_seed(seed)
+        if torch.cuda.is_available():
+            torch.cuda.manual_seed_all(seed)
+    except Exception:
+        pass
+    if ag_set_seed is not None:
+        try:
+            ag_set_seed(seed)
+        except Exception:
+            pass
+
+set_global_seed(AUTO_SEED)
+
 def _extract_close(df: pd.DataFrame) -> pd.Series:
     if isinstance(df.columns, pd.MultiIndex):
         for name in ["Adj Close", "Adj_Close", "adj close", "adj_close"]:
@@ -169,7 +193,6 @@ def ensure_finetuned_predictor(log_cb=print):
     predictor = TimeSeriesPredictor(
         prediction_length=PREDICTION_LENGTH,
         target="target",
-        random_seed=AUTO_SEED,
         eval_metric="WQL",
     )
 
@@ -179,11 +202,10 @@ def ensure_finetuned_predictor(log_cb=print):
             "fine_tune": True,
             "fine_tune_steps": AUTO_STEPS,
             "fine_tune_lr": AUTO_LR,
-            # "device": "gpu"  # AutoGluon verwendet automatisch CUDA, wenn verfügbar
+            # "device": "gpu"  # AutoGluon nutzt automatisch CUDA, wenn verfügbar
         }
     }
 
-    # Training
     predictor.fit(train_data=tsdf, hyperparameters=hyperparams, time_limit=None, presets=None)
     predictor.save(FINETUNED_DIR)
     log_cb(f"Saved finetuned predictor to: {FINETUNED_DIR}")