Spaces:
No application file
No application file
Download recognize.py from Erenuo/RAG_IR_API: direct link, hf CLI and curl.
- Browser
- Download file 3.53 kB
-
https://huggingface.co/spaces/Erenuo/RAG_IR_API/resolve/main/recognize.py
- Command line
-
hf download hf://spaces/Erenuo/RAG_IR_API/recognize.py
-
curl -L -o recognize.py https://huggingface.co/spaces/Erenuo/RAG_IR_API/resolve/main/recognize.py
3.53 kB
| import chromadb | |
| from sentence_transformers import SentenceTransformer | |
| from collections import defaultdict | |
| # --- Configuration --- | |
| COLLECTION_NAME = "intent_recognizer" | |
| DB_PATH = "db" | |
| EMBEDDING_MODEL = 'all-mpnet-base-v2' | |
| # --- Tunable Parameters --- | |
| N_RESULTS = 100 | |
| CONFIDENCE_THRESHOLD = 18.0 | |
| class IntentRecognizer: | |
| def __init__(self): | |
| print("--- Initializing Intent Recognizer ---") | |
| try: | |
| print(f"Loading embedding model: '{EMBEDDING_MODEL}'...") | |
| self.model = SentenceTransformer(EMBEDDING_MODEL) | |
| print(f"Connecting to persistent ChromaDB client at: '{DB_PATH}'...") | |
| client = chromadb.PersistentClient(path=DB_PATH) | |
| print(f"Getting collection: '{COLLECTION_NAME}'...") | |
| self.collection = client.get_collection(name=COLLECTION_NAME) | |
| print("--- Initialization Complete ---") | |
| except Exception as e: | |
| print(f"Error during initialization: {e}") | |
| self.model = None | |
| self.collection = None | |
| def recognize_intent(self, query_text): | |
| if not self.collection or not self.model: | |
| return [] # Return an empty list on error | |
| print(f"\n--- Recognizing Intent for Query: '{query_text}' ---") | |
| query_embedding = self.model.encode([query_text], show_progress_bar=False) | |
| results = self.collection.query( | |
| query_embeddings=query_embedding.tolist(), | |
| n_results=N_RESULTS, | |
| ) | |
| intent_scores = defaultdict(float) | |
| retrieved_metadatas = results.get('metadatas', [[]])[0] | |
| retrieved_distances = results.get('distances', [[]])[0] | |
| if not retrieved_metadatas: | |
| return [] | |
| print(f"Retrieved {len(retrieved_metadatas)} results. Calculating scores...") | |
| for metadata, distance in zip(retrieved_metadatas, retrieved_distances): | |
| intent = metadata.get('intent') | |
| if intent: | |
| similarity_score = 1 - distance | |
| intent_scores[intent] += similarity_score | |
| if not intent_scores: | |
| return [] | |
| # --- CHANGE: Sort all intents by score and get the top 2 --- | |
| sorted_intents = sorted(intent_scores.items(), key=lambda item: item[1], reverse=True) | |
| # Format as a list of dictionaries | |
| top_predictions = [{"intent": intent, "score": score} for intent, score in sorted_intents[:2]] | |
| print("\n--- Scoring Summary (Top 2) ---") | |
| for pred in top_predictions: | |
| print(f" - Intent: {pred['intent']:<25} | Total Score: {pred['score']:.4f}") | |
| # --- CHANGE: Apply threshold to the top prediction --- | |
| # If the best score is too low, return an 'unknown' intent. | |
| if not top_predictions or top_predictions[0]['score'] < CONFIDENCE_THRESHOLD: | |
| print(f"\nTop score is below threshold of {CONFIDENCE_THRESHOLD}.") | |
| return [{"intent": "unknown", "score": top_predictions[0]['score'] if top_predictions else 0.0}] | |
| print(f"\n=> Top Prediction: '{top_predictions[0]['intent']}' (Confidence: {top_predictions[0]['score']:.4f})") | |
| return top_predictions | |
| # This part is no longer needed for running the API, but can be kept for testing | |
| if __name__ == '__main__': | |
| recognizer = IntentRecognizer() | |
| if recognizer.collection: | |
| recognizer.recognize_intent("can you reduce the price on the house in islington") | |