from sentence_transformers import SentenceTransformer
from sklearn.cluster import KMeans
from memory_utils import get_history, decrypt_data
import numpy as np

class QuantumLearner:
    def __init__(self):
        self.model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
        self.model.max_seq_length = 256  # Optimizacija za performanse

    def analyze_conversations(self, history_limit=1000):
        """Analizira povijest razgovora i identificira ključne teme"""
        try:
            history = get_history(limit=history_limit)
            if not history:
                return {"info": "Nema dostupne povijesti razgovora"}
            
            # Priprema i dekriptiranje poruka
            texts = [
                f"{decrypt_data(row[2])} → {decrypt_data(row[3])}" 
                for row in history
                if len(row) >= 4  # Zaštita od nepotpunih podataka
            ]
            
            # Vektorizacija teksta
            embeddings = self.model.encode(texts, show_progress_bar=False)
            
            # Dinamičko grupiranje (2-5 tema)
            optimal_clusters = min(5, max(2, len(texts)//3))
            kmeans = KMeans(n_clusters=optimal_clusters).fit(embeddings)
            
            # Organizacija tema
            topics = {}
            for label, text in zip(kmeans.labels_, texts):
                if label not in topics:
                    topics[label] = []
                topics[label].append(text[:200])  # Skraćivanje za prikaz
                
            return {
                "topics": topics,
                "cluster_centers": kmeans.cluster_centers_.tolist()
            }
            
        except Exception as e:
            return {"error": str(e)}