I am building an API where I need to load a model through pickle and make a prediction

The code is below:

from fastapi import FastAPI
import uvicorn
import joblib
from utilities import *  # upload the functions
from schema import Customer  # load the schema
from sklearn.linear_model import LogisticRegression
from sklearn.base import BaseEstimator, TransformerMixin
import os
import numpy as np
import asyncio

app = FastAPI()

path = 'data/'

X_test = import_test_data(path + 'X_test.xlsx')

MODEL_FILE = "./model/model.pkl"  # path to the model   
   
async def predict_client(data:Customer):

    model = joblib.load(MODEL_FILE) # loading model parameters
    y_hat_test = model.predict(X_test)

    # get the predicted probabilities
    y_hat_test_proba = model.predict_proba(X_test)[:][: , 1]

    default = y_hat_test_proba[0] >= 0.5    

    return {"probability_of_defaulting": float(y_hat_test_proba[0]), "is_defaulting": bool(default)}        

if __name__ == "__main__":     
    uvicorn.run("main:app", host="0.0.0.0", port=8000, reload=True)

The code above works fine as it is. When I load the model through async def predict_client(data:Customer): The API works and I get an output with a prediction. However, my problem is that the variable model is a local one as it is described inside the function.

When i start loading the model (without using async def predict_client) I get as error as shown below:

model = joblib.load(MODEL_FILE)           
print(model['model'].coef_) 

I get an error as follows:

             ^^^^^^^^^^^^^^^^^^^^^^^
  File "xxxxx-packagesjoblibnumpy_pickle.py", line 658, in load     
    obj = _unpickle(fobj, filename, mmap_mode)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "xxxxx/venvLibsite-packagesjoblibnumpy_pickle.py", line 577, in _unpickle
    obj = unpickler.load()
          ^^^^^^^^^^^^^^^^
  File "C:Python311Libpickle.py", line 1213, in load
    dispatch[key[0]](self)
  File "C:Python311Libpickle.py", line 1538, in load_stack_global
    self.append(self.find_class(module, name))
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:Python311Libpickle.py", line 1582, in find_class
    return _getattribute(sys.modules[module], name)[0]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:Python311Libpickle.py", line 331, in _getattribute
    raise AttributeError("Can't get attribute {!r} on {!r}"
AttributeError: Can't get attribute 'WoE_Binning' on <module '__main__' (built-in)>

It seems that the script does not have time to load the model and it runs into an error when it tried to read print(model['model'].coef_).

I tried to fix that issue by doing the following changes.

async def read_model(MODEL_FILE):

    model = joblib.load(MODEL_FILE)

    return model

model=asyncio.run(read_model(MODEL_FILE))
print(model['model'].coef_)

but again i get similar error:

AttributeError: Can't get attribute 'WoE_Binning' on <module '__main__' (built-in)>

How can I successfully load the model once? When I do it through a function the model is a local variable that I cannot re-use.

Thank you