Categorical Data

A color column holding red, green, and blue can't go into a model as text. But the obvious fix — red=1, green=2, blue=3 — is worse: it invents an ordering and a distance ('blue is bigger than green') that don't exist.

One-hot encoding gives each category its own dimension instead:

• red → [1, 0, 0]
• green → [0, 1, 0]
• blue → [0, 0, 1]

Now every category is equidistant from every other — no fake order, no fake distances.