would be helpful if we had more standard python in sanbox..

eg a natural way to answer would be to loop :

train_data can be 'Train_small', 'Train_big', 'Train_tiny'

features should be a list of strings. Features to experiment with are listed below.

['Count_Person'],

['Percent_Person_WithHighBloodPressure'],

['Percent_NoHealthInsurance'],

['Median_Income_Person'],

['Commute_Time']

def run(): features_to_try = ['Count_Person', 'Percent_Person_WithHighBloodPressure', 'Percent_NoHealthInsurance', 'Median_Income_Person', 'Commute_Time'] best_feature, best_mse = None, float('inf') for f in features_to_try: mse = test_analytic_regress(train_data='Train_small', features=[f]) print(f, mse) if mse < best_mse: best_feature, best_mse = f, mse print('best:', best_feature, best_mse) return test_analytic_regress(train_data='Train_small', features=[best_feature])

call this "aggregate cost" or so ?

this

outputs