Because we are drawing machine learning models we are robust against certain side-effects that we might have in production. Let's see how we're robust against missing values.
old_example = pd.DataFrame([{
'island': 'Torgersen',
'bill_length_mm': 39.1,
'bill_depth_mm': 18.7,
'flipper_length_mm': 220.0,
'body_mass_g': 5750.0,
'sex': 'male'}
])
new_example = pd.DataFrame([{
'island': 'Torgersen',
'bill_length_mm': np.nan,
'bill_depth_mm': 18.7,
'flipper_length_mm': 220.0,
'body_mass_g': 5750.0,
'sex': 'male'}
])
model.predict_proba(old_example)
model.predict_proba(new_example)