Lesson 4 — Prediction by average

The simplest model: always guess the average

A baseline is a deliberately dumb model you compare everything against. The classic baseline for numbers ignores the input and always predicts the mean of the training data (the data the model learns from). We judge it with the mean absolute error (MAE): the average of how far each prediction is from the real value, ignoring the sign.

abs(20 - 25)   # 5 — abs() drops the minus sign

Why it matters: if your fancy model cannot beat "always guess the average", it has learned nothing useful. Always measure a baseline first.