[FAQ] Homework 2 Q3 - why do fill-with-0 and fill-with-mean look identical?

[yazdanparasthesam]

Course

machine-learning-zoomcamp

Question

Homework 2 Q3: filling the horsepower NAs with 0 and with the mean gives me the same RMSE - how do I tell which option is better?

Answer

They are not actually identical, the difference just hides behind rounding. In the pinned 2026
release the validation RMSEs are:

round(rmse_zero, 3)   # 2.205
round(rmse_mean, 3)   # 2.202   <- better

Round to 3 decimals as the homework asks (the 2026 wording says this explicitly because 2 decimals
turned it into a tie). So the answer is "With mean".

Two things that silently break this question:

1. The mean must be computed on the training split only, never on the full dataset - otherwise
   you leak information from validation/test:

   hp_mean = df_train.horsepower.mean()   # correct
   hp_mean = df.horsepower.mean()         # leakage

2. Use the same fill value on train, validation and test. Don't recompute the mean per split.

def prepare_X(d, fill_value):
    return d[['engine_displacement', 'horsepower', 'vehicle_weight', 'model_year']].fillna(fill_value).values

Checklist

• I have searched existing FAQs and this question is not already answered
• The answer provides accurate, helpful information
• I have included any relevant code examples or links