services/sentence_transformer_service.py

import pickle
import warnings

# Filter NumPy array implementation warnings specifically
warnings.filterwarnings(
    "ignore", 
    message=".*array.*implementation doesn't accept a copy keyword.*", 
    category=DeprecationWarning
)

# Or alternatively, target the exact warning message:
warnings.filterwarnings(
    "ignore", 
    message=".*NumPy will pass `copy` to the `__array__` special method.*", 
    category=DeprecationWarning
)

from config import (
    MODEL_NAME, MODEL_TYPE, DEVICE_TYPE,
    SENTENCE_EMBEDDING_FILE,  
    STANDARD_NAME_MAP_DATA_FILE, SUBJECT_DATA_FILE, DATA_DIR
)
from sentence_transformer_lib.sentence_transformer_helper import SentenceTransformerHelper
from data_lib.subject_data import SubjectData
from data_lib.standard_name_map_data import StandardNameMapData
import os


class SentenceTransformerService:
    def __init__(self):
        self.sentenceTransformerHelper = None
        self.dic_standard_subject = None
        self.anchor_name_sentence_embeddings = None
        self.sampleData = None

    def load_model_data(self):
        """Load model and data only once at startup"""
        if self.sentenceTransformerHelper is not None:
            print("Model already loaded. Skipping reload.")
            return  # Không load lại nếu đã có model

        print("Loading models and data...")
        # Load sentence transformer model
        self.sentenceTransformerHelper = SentenceTransformerHelper(model_name=MODEL_NAME, model_type=MODEL_TYPE)
        print(f"Loading model {MODEL_NAME} with type {MODEL_TYPE}") 

        # Load standard subject dictionary
        self.dic_standard_subject = SubjectData.create_standard_subject_dic_from_file(SUBJECT_DATA_FILE)

        # Load and process sample data
        self.standardNameMapData = StandardNameMapData(self.anchor_name_sentence_embeddings)
        self.standardNameMapData.load_data_from_csv(STANDARD_NAME_MAP_DATA_FILE)
        self.standardNameMapData.process_data()

        # Load pre-computed embeddings and similarities
        if SENTENCE_EMBEDDING_FILE is not None:
            with open(SENTENCE_EMBEDDING_FILE, "rb") as f:
                self.anchor_name_sentence_embeddings = pickle.load(f)
            print(f"Loaded anchor name sentence embeddings shape: {self.anchor_name_sentence_embeddings.shape}")
        else:
            list_anchor_name_sentence = self.standardNameMapData.processed_data["anchor_name_sentences"]
            anchor_name_sentence_embeddings = self.sentenceTransformerHelper.create_embeddings(list_anchor_name_sentence)
            self.anchor_name_sentence_embeddings = anchor_name_sentence_embeddings
            sentence_embedding_file_path = os.path.join(DATA_DIR, "anchor_name_sentence_embeddings_" + MODEL_NAME.rsplit("/", maxsplit=1)[-1] + ".pkl")
            with open(sentence_embedding_file_path, "wb") as f:
                pickle.dump(self.anchor_name_sentence_embeddings, f)
            print(f"Saved anchor name sentence embeddings to {sentence_embedding_file_path}")

        print("Models and data loaded successfully")

# Global instance (singleton)
sentence_transformer_service = SentenceTransformerService()