Two Baselines to Beat

Before you build a forecasting model, you need a number to beat. A baseline is the simplest guess there is. If a model can't do better than it, the model isn't worth building.

Two baselines for weekly revenue: guess each week as last week's revenue, with .shift(1), or as the average of the 4 weeks before, with .rolling(4).mean().shift(1). The .shift(1) at the end matters: without it, a week's average would include that week's own revenue, the very number you're guessing.

mean_absolute_error(real, guesses) from sklearn.metrics gives back how far off the guesses are on average, in the same units as the data. On a small series:

temps = pd.Series([20, 22, 21, 25])
guess = temps.shift(1)   # NaN, 20, 22, 21
print(mean_absolute_error(temps.iloc[1:], guess.iloc[1:]))   # (2 + 1 + 4) / 3 = 2.333…

The starting code builds table, one row per week with its revenue. Your script must print the last 8 weeks (the same test weeks as in the last lesson) with both guesses, then each baseline's error:

            revenue  last_week    avg_4
date
2026-10-11     2122     3084.0  2428.50
2026-10-18     2451     2122.0  2631.75
2026-10-25     2692     2451.0  2616.00
2026-11-01     3110     2692.0  2587.25
2026-11-08     3549     3110.0  2593.75
2026-11-15     3200     3549.0  2950.50
2026-11-22     2671     3200.0  3137.75
2026-11-29     3090     2671.0  3132.50
Last week MAE: 460.8
Average of 4 MAE: 350.0

Your task: add the columns last_week and avg_4, keep the last 8 rows of table in test, print it with .to_string(), and print the mean absolute error of each guess on those weeks, rounded to 1 decimal.

Assignment
Follow the Coding Tutorial to add the two baseline columns and print their errors on the test weeks.

Hint
Each guess for a week may use only the weeks before it, so both columns end with .shift(1). iloc[-8:] gives back the last 8 rows; they have no NaN, so mean_absolute_error works on them.