Campaign
Intermediate
Mission 1: Polynomial Panic
Reduce the generalisation gap without making the model underfit.
Mission progress0%
Score: 0
Streak: 0
Train score
0.96
Unregularised
Validation score
0.68
Needs defence
Generalisation gap
0.28
Lower is safer
Complexity
91
0–100 scale
Stability
43
Across folds
Time
02:30
Remaining
Defence console
Model and regularisation controls
0.25
0.50
0.20
12
2
8
80
0.010
18
0.00
0%
Live simulation
Decision boundary and model complexity
Class A
Class B
Boundary
The model memorises training noise. Increase regularisation until validation performance improves without collapsing useful structure.
Learning curves
Training and validation loss
Trade-off
Complexity versus validation score
Coefficient control
Regularisation path
Reliability
Cross-validation stability
Mission board
Challenges and mastery
Performance history
Defence results
| Mission | Technique | Gap | Complexity | Score | Result |
|---|---|---|---|---|---|
| Complete a defence to build your history. | |||||
Best score
0
Mastery
0%
Defences
0
Game settings
Accessibility and presentation
Keyboard: T train, P pause, R reset, H hint, N next.
How to play
Build a safer model
- Inspect the initial training and validation gap.
- Choose a technique suitable for the current model.
- Tune penalties, architecture limits, or stopping rules.
- Train and compare curves, stability, and complexity.
- Submit when the model generalises without severe underfitting.