Section 1 — What you will learn
✓ By the end you will understand
- Decision boundary: the laser wall that separates two teams.
- Margin: the safe corridor — SVM tries to make it as wide as possible.
- Support vectors: the closest fighters that physically hold the wall in place.
- Soft margin (C): allow some mistakes when data is noisy (real world).
- Kernel Trick: when a straight wall can’t win, we warp the world so it can.
How to play (step-by-step):
1) Click Start Training → watch the wall push and the margin stretch.
2) Increase C → wall becomes stricter (less forgiving). Decrease C → more forgiving.
3) Switch dataset to Rings/XOR → Linear fails → switch to RBF Kernel to win.
4) Press Break it 💥 → noise + outliers → see why soft margin matters.
5) In Challenge mode, your goal is: high truth accuracy + stable margin + low gap.
1) Click Start Training → watch the wall push and the margin stretch.
2) Increase C → wall becomes stricter (less forgiving). Decrease C → more forgiving.
3) Switch dataset to Rings/XOR → Linear fails → switch to RBF Kernel to win.
4) Press Break it 💥 → noise + outliers → see why soft margin matters.
5) In Challenge mode, your goal is: high truth accuracy + stable margin + low gap.
Section 2 — Controls
Challenge rewards: good boundary + wide margin + generalization (truth) not just training.
Story helps kids remember what numbers mean.
Pro tip: Rings + Linear kernel = fail. Switch to RBF = win.
Kernel Trick: we don’t literally draw higher dimensions — but the boundary becomes curved.
High C = strict soldier. Low C = chill soldier.
Noise makes some points lie.
High γ = very bendy boundary.
Degree ↑ → more curves.
If wall shakes: reduce α.
More steps = faster training animation.
Section 3 — Live stats
Train Accuracy
—
Truth Accuracy
—
Hinge Loss
—
Margin Width
—
Support Vectors
—
Generalization Gap
—
Live Chart
Loss + accuracy trend (updates live)
Section 4 — Live Math (kid-friendly)
Loading…
🧠 Kid Coach
Loading…
Tip: Linear fails on Rings/XOR → switch kernel.