PART A · BUILD THE MODEL data → train (by remembering) → test
Zone 1 · The Data
24 songs, each with two features (tempo, length) and one label. The designer split them up front: 16 for training, 8 for testing — the test songs never enter the machine’s memory.
| # | Song | Tempo | Length | Label | Role | Prediction |
|---|
Zone 2 · The Model
Training = remembering the 16 train songs — it finished the moment the page opened; nothing else was computed. That is why there is no train button. The painted background shows the model’s decision boundary. The 8 test songs are NOT on this map — they stay locked in the table until the test runs.
DANCE
CHILL
the decision boundary
Zone 3 · Model Testing
Training error: —
Test error: —
The 8 test songs are scored by the 16-song memory only — a test song is never among its own voters. A new song, not one it memorized.
PART B · USE THE MODEL apply it to songs it has never met
Zone 4 · Model Application
The same model, drawn again right here — so the feature space is at your fingertips when you drag.
Drag me onto THIS map. Where you drop me is my tempo and my length — my two features. A drop is a prediction, not data: nothing gets recorded.