ML Math Lab — KNN: Neighborhood Scanner (1-File)

Section 1 — What you will learn

✓ By the end you will understand

K: how many neighbors vote for the answer.
Majority vote: whichever team is bigger among the K neighbors wins.
Distance: Euclidean (circle) vs Manhattan (diamond).
Noise: infiltrators + jitter that tries to fool the scanner.
Why K matters: small K can overfit; big K can underfit.

Section 2 — Controls

Mode

Challenge adds scoring + auto-stop when accuracy becomes “smart”.

K Neighbors: 7

Tip: K=1 can be too jumpy. Higher K becomes smoother.

Distance Metric

Watch the radar shape change.

Team Balance (Green share): 0.50

Try 0.20 vs 0.80 and watch the territory change.

Noise (Infiltrators): 0.10

Noise is visual: more mismatched dots + wobble + weaker confidence.

Scan Rate (animation speed): 2.0x

This controls how fast probes and stats update.

Break it (Chaos slider): 0.00

Break it adds: extra label flips + drifting clusters + outliers.

Real-world toggle

The arena is a “feature space” map. Each dot is one real case.

How to play:
• Move mouse on the map → radar scans neighbors.
• Click → drop a probe (it will be scored).
• Yellow probes = training probes. Cyan probes = test probes (hidden until Reveal).

Section 3 — Live stats

Predicted class (cursor)

—

Confidence (vote %)

—

Global Accuracy

—

Test Accuracy

Hidden

Noise Confusion

—

Overfit Risk (Train−Test)

Hidden

Live Chart — Accuracy over time

This updates continuously while probes run.

Section 4 — Math (simple)

How KNN decides Hover = see distance + votes

Loading…

🧠 Kid Coach

Loading…

Tip: Try K=1 vs K=15 and compare the territory.

Section 5 — Arena legend (kept LEFT)

Green dots = class 1 (YES)
Red dots = class 0 (NO)
Blue glow = probability field (how likely class 1 is)
Radar = your scanner ring (or diamond)
Lasers = links to K nearest neighbors
Yellow probe = train probe, Cyan probe = test probe (truth)