Section 1 — What you will learn
✓ By the end you will understand
- Degree = how “bendy” the curve brain is.
- Training vs Testing = the truth moment.
- Underfit (too stiff) & Overfit (too wiggly).
- Smoothness Shield (λ) = regularization that calms the curve.
- Error beams show MSE visually.
- Noise is visible (beams from “true curve” to points).
Section 2 — Controls
Degree 1 is a straight stick. Degree 12 can become spaghetti.
Noise = real-life randomness. You will see it as vertical “noise beams”.
More points → curve becomes more reliable.
Higher λ forces the curve to stay calm (less wiggle).
Drag yellow points on the graph.
Cyan points are test points (hidden until you reveal).
Tip: low Train MSE + high Test MSE = overfitting.
Cyan points are test points (hidden until you reveal).
Tip: low Train MSE + high Test MSE = overfitting.
Section 3 — Live stats
Train MSE
—
Test MSE
Hidden
Complexity
—
Hint
—
Section 4 — Math (simple)
What the model is doing
Drag a point → watch the curve “re-learn”
Loading…
🧠 ML Buddy
Loading…
Tip: reveal test to catch overfitting.
Section 5 — Curve Playground
true curve (ghost)
training points
test points
learned curve
error beams
noise beams
Live explanation
Drag a yellow point. The model will change its curve to reduce error.
Game HUD
Goal: make Train MSE low AND Test MSE low (generalization).
Model Health
Reveal test points to judge honestly.
Unknown
Overfit Meter
Hidden until you reveal test.
Quick actions
Center X helps you “see” spread. Calm increases λ. Spice increases degree.