apply_sigmoid

app.py

@@ -100,6 +100,10 @@ def get_embedding(text: str):
 # ABLATION STUDY
 # ==========================================
 
+# Helper Function
+def apply_sigmoid(logit):
+    return 1 / (1 + math.exp(-logit))
+
 def bm25_only(query: str, match_count: int = 50):
     """Hanya Lexical / Full-Text Search (Tanpa Vector, Tanpa Gemini, Tanpa Reranker)"""
     # Catatan: Pastikan Anda sudah membuat RPC 'search_kbli_lexical' di Supabase
@@ -191,20 +195,12 @@ def hybrid_search_no_gemini(query: str, match_count: int = 50):
         print("Reranker error:", e)
         return {"results": sorted(candidates, key=lambda x: x.get("similarity", 0), reverse=True)[:10]}
 
-    rerank_vals = [float(s) for s in scores]
-    rmin, rmax = min(rerank_vals), max(rerank_vals)
-    
-    for c, s in zip(candidates, rerank_vals):
-        c["rerank_score"] = s
-        if rmax - rmin > 1e-9:
-            c["rerank_norm"] = (s - rmin) / (rmax - rmin)
-        else:
-            c["rerank_norm"] = 0.0
-
-        db_hybrid_sim = c.get("similarity", 0.0)
-        c["final_score"] = (0.5 * db_hybrid_sim) + (0.5 * c["rerank_norm"])
+    for c, s in zip(candidates, scores):
+        c["rerank_score"] = float(s)
+        # Ubah logit jadi probabilitas absolut (0 sampai 1)
+        c["rerank_prob"] = apply_sigmoid(float(s))
 
-    candidates = sorted(candidates, key=lambda x: x["final_score"], reverse=True)
+    candidates = sorted(candidates, key=lambda x: x["rerank_prob"], reverse=True)
     return {"results": candidates[:10]}
 
 # ==========================================