example.py

import ...

def hpo(train_data, test_data, feature_transformation):

 best_accuracy = 0.0
 best_regularizer = None

 for regularizer in [None, 'l1', 'l2']:

   pipeline = Pipeline([
     ('features', feature_transformation),
     ('learner', SGDClassifier(loss='log', penalty=regularizer))])

   model = pipeline.fit(train_data, train_data[target_column])
   accuracy = model.score(test_data, test_data[target_column])

   if accuracy > best_accuracy:
     best_accuracy = accuracy
     best_regularizer = regularizer

 return best_regularizer

data_file = ...
zip_codes_for_training = ...

numerical_columns = ['age_in_years', 'monthly_income', 'total_savings']
categorical_columns = ['job_level', 'education_level', 'zip_code']
target_column = 'is_eligible_for_loan'

data = pd.read_csv(data_file)
data = data.drop(columns=['race', 'gender'])
data[numerical_columns] = data[numerical_columns].apply(lambda col: col - col.mean())

train_data, test_data = train_test_split(data, test_size=0.2)
train_data = train_data[train_data.zip_code.isin(zip_codes_for_training))

feature_transformation = ColumnTransformer(transformers=[
 ('num_features', StandardScaler(with_mean=False), numerical_columns),
 ('cat_features', OneHotEncoder(handle_unknown='ignore'), categorical_columns)
])

best_regularizer = hpo(train_data, test_data, feature_transformation)

final_pipeline = Pipeline([
 ('features', feature_transformation),
 ('learner', SGDClassifier(loss='log', penalty=best_regularizer))
])

model = final_pipeline.fit(train_data, train[target_column])
accuracy = model.score(test_data, test[target_column])

print(f'Final accuracy on test set is {accuracy}')