FastAPI/app/utils/embedding.py

import os
from fastapi import HTTPException
import gensim.downloader as api

# Ensure the environment variable is set correctly
gensim_data_dir = os.getenv("GENSIM_DATA_DIR", "/home/user/gensim-data")

# Load the GloVe model
model = api.load("glove-wiki-gigaword-50")


def get_embedding(word: str) -> list:
    min_val = -5.4593
    max_val = 5.3101
    global model
    try:
        embediing = model[word.lower()]
        # normalize vector min max to -1 to 1
        embediing = (embediing - min_val) / (max_val - min_val) * 2 - 1
        return embediing
    except KeyError:
        print("Word not in vocabulary")
        raise HTTPException(status_code=404, detail="Word not in vocabulary")