[Python] FastAPI: Internal Server Error when returning ML model prediction

I would like to test my pipeline in FastAPI, but I can't find the mistake in my code. When I test it using the Visual Studio Code (using a print() statement), it works. However, when I try accessing the endpoint through the browser, I get Internal Server Error. It works when I return something else as the prediction results (e.g., some string), instead of the actual prediction results.

Here is my code:

class FraudDetection333(BaseModel):
"""
Input features validation for the ML model
"""
user_id: int
signup_day: int
signup_month: int
signup_year: int
purchase_day: int
purchase_month: int
purchase_year: int
purchase_value: float
source: str
browser: str
sex: str
age: int

@api.post("/predictions_test",tags=['DecisionTreeClassifier'])
def predictions_test(fraud:FraudDetection333):
"""
aram:input data from the post request
:return predicted type
"""
features = [[
fraud.user_id,
fraud.signup_day,
fraud.signup_month,
fraud.signup_year,
fraud.purchase_day,
fraud.purchase_month,
fraud.purchase_year,
fraud.purchase_value,
fraud.source,
fraud.browser,
fraud.sex,
fraud.age
]]
rf_model = joblib.load('./rf_model.pkl')
new = (pd.DataFrame(features, index = ['0'], columns = ['user_id','signup_day',
'signup_month', 'signup_year',
'purchase_day', 'purchase_month', 'purchase_year','purchase_value',
'source','browser','sex','age']))
new_prediction = rf_model.predict(new)
return {
"Predicted transaction(1 - fraud, 0 - not fraud)": new_prediction
}


If I try the below (using a print() statement), it prints out the expected outcome:

featuress={
"user_id": 22058,
"signup_day": 24,
"signup_month": 2,
"signup_year": 2015,
"purchase_day": 18,
"purchase_month": 4,
"purchase_year": 2015,
"purchase_value": 34,
"source": "SEO",
"browser": "Chrome",
"sex": "M",
"age": 39
}
rf_model = joblib.load('./rf_model.pkl')
new = (pd.DataFrame(featuress, index = ['0'], columns = ['user_id','signup_day',
'signup_month', 'signup_year',
'purchase_day', 'purchase_month', 'purchase_year','purchase_value',
'source','browser','sex','age']))
new_prediction = rf_model.predict(new)
print(new)
print(new_prediction)


If I type return {"Predicted transaction(1 - fraud, 0 - not fraud)": 'Hi'}, it also works. Image here.

Continue reading...