Lesson 7 — Scaling new data with training stats

Scale new data with the training min and max

In Lesson 2 you scaled one list. In real ML you have training data (what the model learns from) and new data (what it sees later). The rule: compute min and max from training data only, then reuse those same numbers for new data. Peeking at new data while preparing the model is called data leakage and makes results look better than they really are.

lo, hi = 10, 50          # learned from training data
(60 - lo) / (hi - lo)    # 1.25 — new data can fall outside 0..1

Why it matters: the model was trained on one scale; new inputs must be translated to that exact same scale, or its predictions are nonsense.