Update agent.py

agent.py

@@ -1,4 +1,4 @@
-# agent.py — AGENTE SEMÁNTICO CON PRESERVACIÓN DE ENTIDADES v2.1
+# agent.py — AGENTE SEMÁNTICO CON PRESERVACIÓN DE ENTIDADES
 import os
 import time
 import logging
@@ -10,35 +10,24 @@ import faiss
 import spacy
 from spacy.lang.en import English
 
-# Configurar logging
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
-# Cargar modelo de spaCy (con descarga automática si falta)
 try:
     NLP = spacy.load("en_core_web_sm")
     logger.info("✅ spaCy 'en_core_web_sm' cargado.")
 except OSError:
-    logger.info("📥 Descargando 'en_core_web_sm' (primera ejecución)...")
+    logger.info("📥 Descargando 'en_core_web_sm'...")
     from spacy.cli import download
     download("en_core_web_sm")
     NLP = spacy.load("en_core_web_sm")
     logger.info("✅ spaCy 'en_core_web_sm' descargado y cargado.")
 except Exception as e:
-    logger.warning(f"⚠️ Error inesperado con spaCy: {e}. Usando tokenizer básico.")
+    logger.warning(f"⚠️ Error con spaCy: {e}. Usando tokenizer básico.")
     NLP = English()
     NLP.add_pipe("sentencizer")
 
 class ImprovedSemanticAgent:
-    """
-    🧠 AGENTE SEMÁNTICO CON PRESERVACIÓN DE ENTIDADES v2.1
-    
-    ✅ Extrae entidades clave con spaCy (descarga automática si es necesario).
-    ✅ Filtra ejemplos que no comparten entidades con el usuario.
-    ✅ Sintetiza prompts nuevos (no copia).
-    ✅ Usa índice FAISS desde disco.
-    """
-
     def __init__(self):
         logger.info("🚀 Cargando modelo de embeddings (bge-small-en-v1.5)...")
         self.embedding_model = SentenceTransformer('BAAI/bge-small-en-v1.5')
@@ -59,7 +48,7 @@ class ImprovedSemanticAgent:
                 try:
                     return future.result(timeout=60)
                 except FutureTimeoutError:
-                    return "❌ Timeout inicializando agente (más de 60s)"
+                    return "❌ Timeout inicializando agente"
         except Exception as e:
             return f"❌ Error: {str(e)}"
 
@@ -86,7 +75,7 @@ class ImprovedSemanticAgent:
             if len(chunk.text) > 2 and not all(t.is_stop for t in chunk):
                 entities.add(chunk.lemma_.replace(" ", "_"))
         text_lower = text.lower()
-        if "fire" in text_lower or "flame" in text_lower or "burning" in text_lower:
+        if "fire" in text_lower or "flame" in text_lower:
             entities.add("on_fire")
         if "ice" in text_lower or "frozen" in text_lower:
             entities.add("frozen")
@@ -112,43 +101,30 @@ class ImprovedSemanticAgent:
             query_embedding = query_embedding.astype('float32').reshape(1, -1)
             distances, indices = self.index.search(query_embedding, 5)
 
-            user_entities = self._extract_core_entities(user_prompt)
-            logger.info(f"🔑 Entidades clave del usuario: {user_entities}")
-
             candidates = []
-            filtered_count = 0
             for idx in indices[0]:
-                if idx >= len(self.indexed_examples):
-                    continue
-                caption = self.indexed_examples[idx]['caption']
-                if not user_entities:
-                    candidates.append(caption)
-                    continue
-                caption_entities = self._extract_core_entities(caption)
-                if user_entities & caption_entities:
-                    candidates.append(caption)
-                else:
-                    caption_lower = caption.lower()
-                    literal_match = any(
-                        ent.replace("_", " ") in caption_lower or ent in caption_lower
-                        for ent in user_entities
-                    )
-                    if literal_match:
-                        candidates.append(caption)
-                    else:
-                        filtered_count += 1
-
-            logger.info(f"🗂️ Recuperados: {len(candidates)} ejemplos útiles ({filtered_count} filtrados)")
-
+                if idx < len(self.indexed_examples):
+                    candidates.append(self.indexed_examples[idx]['caption'])
             if not candidates:
                 return self._structural_fallback(user_prompt, category), "🔧 Fallback estructural"
 
+            # 🔑 EXTRAER ENTIDADES DEL USUARIO
+            user_entities = self._extract_core_entities(user_prompt)
+            user_has_clothing = any("swimsuit" in e or "dress" in e or "suit" in e or "armor" in e for e in user_entities)
             user_words = set(user_prompt.lower().split())
             all_parts = []
+
             for caption in candidates:
                 parts = [p.strip() for p in caption.split(',') if 8 <= len(p) <= 120]
                 for part in parts:
                     part_lower = part.lower()
+                    part_entities = self._extract_core_entities(part)
+                    part_has_clothing = any("coat" in e or "jacket" in e or "scarf" in e or "hood" in e or "sweater" in e or "parka" in e for e in part_entities)
+
+                    # ❌ Saltar si hay conflicto de ropa
+                    if user_has_clothing and part_has_clothing:
+                        continue
+
                     if len(set(part_lower.split()) - user_words) >= 2:
                         all_parts.append(part)