inference.py

import pickle
import json
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity

class BuildingMaterialsInference:
    def __init__(self):
        # Load model data
        with open('building_materials_model.pkl', 'rb') as f:
            model_data = pickle.load(f)
        
        self.knowledge_base = model_data['knowledge_base']
        self.vectorizer = model_data['vectorizer']
        self.query_vectors = model_data['query_vectors']
        self.queries = model_data['queries']
        self.responses = model_data['responses']
    
    def __call__(self, inputs):
        """Hugging Face inference API compatible method"""
        if isinstance(inputs, dict):
            query = inputs.get('query', inputs.get('inputs', ''))
        else:
            query = str(inputs)
        
        return self.search_materials(query)
    
    def search_materials(self, user_query, top_k=3):
        """Search for building materials based on user query"""
        user_query_lower = user_query.lower()
        
        # Direct match first
        if user_query_lower in self.knowledge_base:
            response_data = json.loads(self.knowledge_base[user_query_lower])
            return {
                "results": response_data.get('results', []),
                "confidence": 1.0,
                "query_matched": user_query
            }
        
        # Semantic similarity search
        user_vector = self.vectorizer.transform([user_query_lower])
        similarities = cosine_similarity(user_vector, self.query_vectors)[0]
        
        # Get top matches
        top_indices = np.argsort(similarities)[::-1][:top_k]
        
        # Combine results from top matches
        all_results = []
        seen_suppliers = set()
        
        for idx in top_indices:
            if similarities[idx] > 0.1:  # Minimum similarity threshold
                response_data = json.loads(self.responses[idx])
                for result in response_data.get('results', []):
                    supplier = result.get('supplier', '')
                    if supplier not in seen_suppliers:
                        all_results.append(result)
                        seen_suppliers.add(supplier)
        
        # Sort by price if available
        def extract_price(result):
            price_str = result.get('price', '£0')
            try:
                return float(price_str.replace('£', '').replace(',', ''))
            except:
                return 999999
        
        all_results.sort(key=extract_price)
        
        return {
            "results": all_results[:5],  # Return top 5
            "confidence": float(max(similarities)) if len(similarities) > 0 else 0.0,
            "query_matched": user_query
        }

# Create model instance for Hugging Face
model = BuildingMaterialsInference()