Build API using FastAPI for Classification Model produced using pycaret

Question

I'm using pycaret as my ML workflow, I tried to create an API using FastAPI. This is my first time playing into production level, so I'm bit confused about API

I have 10 features; age: float, live_province: str, live_city: str, live_area_big: str, live_area_small: str, sex: float, marital: float, bank: str, salary: float, amount: float and a label which it contains the binary value (0 and 1).

This is what my script for building the API

from pydantic import BaseModel
import numpy as np
from pycaret.classification import *

import uvicorn
from fastapi import FastAPI

app = FastAPI()

model = load_model('catboost_cm_creditable')

class Data(BaseModel):
    age: float
    live_province: str
    live_city: str
    live_area_big: str
    live_area_small: str
    sex: float
    marital: float
    bank: str
    salary: float
    amount: float

input_dict = Data

@app.post("/predict")
def predict(model, input_dict):
    predictions_df = predict_model(estimator=model, data=input_dict)
    predictions = predictions_df['Score'][0]
    return predictions

When I tried to run uvicorn script:app and went to the documentation I can't find the parameter for my features, the parameters only show model and input_dict

How to take my Features onto Parameters in the API?

Yagiz Degirmenci · Accepted Answer · 2020-08-31T15:28:02.913

3

You need to Type-hint your Pydantic model to make it work with your FastAPI

Imagine like you are really working with Standard Python, when you need to documentate that function,

def some_function(price: int) ->int:
    return price

With Pydantic there is nothing different than the example above

Your class Data is actually a python @dataclass with super-powers(comes from Pydantic)

from fastapi import Depends

class Data(BaseModel):
    age: float
    live_province: str
    live_city: str
    live_area_big: str
    live_area_small: str
    sex: float
    marital: float
    bank: str
    salary: float
    amount: float


@app.post("/predict")
def predict(data: Data = Depends()):
    predictions_df = predict_model(estimator=model, data=data)
    predictions = predictions_df["Score"][0]
    return predictions

There is a one little trick, with Depends, you 'll get a single queries like when you are defining each field seperately.

With Depends

Without Depends

edited Aug 31 '20 at 15:28

answered Aug 31 '20 at 15:20

Yagiz Degirmenci

16,595
7
65
85

Hi, I tried this and my API works perfectly but when I tried to run the API it returns ```500 Internal Server Error ERROR: Exception in ASGI application``` I believe because I did not call the model correctly, but I can't figured out the correct ways to load the pickle file. Can you help me? – ebuzz168 Sep 13 '20 at 09:09
@ebuzz168 can you provide a traceback ? – Yagiz Degirmenci Sep 13 '20 at 09:14
it throws `AttributeError: 'Data' object has no attribute 'columns'` error, this is not related to the question, you should ask a new question, but as far as i understand you are doing `data.columns = [str(i) for i in data.columns]` , which is seems weird for me because, but you are just reiterating it and the error says Data object has no attr columns, it does hasattr(Data "columns") underneath. – Yagiz Degirmenci Sep 13 '20 at 11:57

score 1 · Answer 2 · answered Aug 31 '20 at 15:08

Your problem is with the definition of the API's function. You added an argument for you data input but you didn't tell FastAPI it's type. Also I assume that you mean't to use the model that you've loaded globally instead of received it as a parameter. Also you don't need to create a global instance for your input data, as you want to get it from the user.

Therefore, simply change the signature of your function to:

def predict(input_dict: Data):

and remove the line:

input_dict = Data

(Which just creates an Alias to your class Data, named input_dict)

You'll end up with:

app = FastAPI()

model = load_model('catboost_cm_creditable')

class Data(BaseModel):
    age: float
    live_province: str
    live_city: str
    live_area_big: str
    live_area_small: str
    sex: float
    marital: float
    bank: str
    salary: float
    amount: float

@app.post("/predict")
def predict(input_dict: Data):
    predictions_df = predict_model(estimator=model, data=input_dict)
    predictions = predictions_df['Score'][0]
    return predictions

Also, I would recommend changing the name of the class Data to something more clear and easier to understand, even DataUnit would be better in my opinion as Data is too general.

Build API using FastAPI for Classification Model produced using pycaret

2 Answers2

With Depends

Without Depends