Update app/src/predict_one_image_with_metrics.py

app/src/predict_one_image_with_metrics.py

@@ -23,9 +23,8 @@ from tag_pin_builder import build_candidate_pins, build_candidate_tags
 # =========================
 
 PROJECT_ROOT = Path(
-    os.getenv("POSTER_PROJECT_ROOT", str(Path(__file__).resolve().parents[1]))
+    os.getenv("POSTER_PROJECT_ROOT", "/Practise/app")
 ).resolve()
-
 AI_DETECTOR_PATH = PROJECT_ROOT / "artifacts" / "ai_detector" / "ai_detector_logreg.joblib"
 
 CLIP_MODEL_NAME = "openai/clip-vit-base-patch32"
@@ -600,13 +599,7 @@ def build_neutral_hybrid_row(image_path: Path, embedding: np.ndarray) -> pd.Data
     return pd.DataFrame([row])
 
 
-def predict_with_model(model, X: Any) -> Tuple[str, float]:
-    """
-    Поддерживает два формата:
-    1) обычный sklearn Pipeline / estimator
-    2) dict {"preprocessor": ..., "model": ...} для calibrated RF
-    """
-    # Новый формат: вручную сохранённый preprocessor + calibrated model
+def predict_with_model(model, X):
     if isinstance(model, dict) and "preprocessor" in model and "model" in model:
         X_proc = model["preprocessor"].transform(X)
         pred = model["model"].predict(X_proc)[0]
@@ -614,23 +607,20 @@ def predict_with_model(model, X: Any) -> Tuple[str, float]:
         if hasattr(model["model"], "predict_proba"):
             prob_good = float(model["model"].predict_proba(X_proc)[0, 1])
         else:
-            prob_good = float("nan")
-
-    # Старый формат: обычный sklearn Pipeline / estimator
+            prob_good = None
     else:
         pred = model.predict(X)[0]
 
         if hasattr(model, "predict_proba"):
             prob_good = float(model.predict_proba(X)[0, 1])
         else:
-            prob_good = float("nan")
+            prob_good = None
 
-    if hasattr(model, "predict_proba"):
-        prob_good = float(model.predict_proba(X)[0, 1])
+    if prob_good is not None:
+        label = "good" if prob_good >= 0.5 else "bad"
     else:
-        prob_good = None
-    
-    label = "good" if (prob_good is not None and prob_good >= 0.5) else "bad"
+        label = "good" if int(pred) == 1 else "bad"
+
     return label, prob_good
 
 
@@ -1036,6 +1026,8 @@ def main() -> None:
 
     hybrid_logreg_label, hybrid_logreg_prob = predict_with_model(hybrid_logreg, X_hybrid)
     hybrid_rf_label, hybrid_rf_prob = predict_with_model(hybrid_rf, X_hybrid)
+    hybrid_rf_prob_display = hybrid_rf_prob * (2 / 3)
+    hybrid_rf_label_display = "good" if hybrid_rf_prob_display >= 0.5 else "bad"
 
     ensemble_prob_good = float(
         0.50 * clip_prob_good +
@@ -1188,9 +1180,9 @@ def main() -> None:
                 "prob_good": round(hybrid_logreg_prob, 4),
             },
             "clean_hybrid_random_forest": {
-                "label": hybrid_rf_label,
-                "prob_good": round(hybrid_rf_prob*(2/3), 4),
-            },
+            "label": hybrid_rf_label_display,
+            "prob_good": round(hybrid_rf_prob_display, 4),
+        },
         },
         "ai_detection": ai_detection_final,
         "ai_detection_ml": ai_detection_ml,