Score0
RankScaling Beginner
Epoch0 / 80
Best accuracy—
Streak0
Challenge mission
Choose a scaler that improves convergence without leaking test information.
Scaling arena
p5.js renders the animated HTML canvas.Experiment controls
Educational simulation. Metrics illustrate common scaling effects.Dataset
Scaling
Model
Evaluation
Simulation
Selected features
Use a header row. The final column is treated as the target.
Results dashboard
Experiment ready
Generate data and start training to compare scaled and unscaled behaviour.
| Metric | Unscaled | Scaled | Difference |
|---|
Feature statistics
Statistics are fitted on the training portion unless leakage is enabled.| Feature | Original min | Original max | Mean | Standard deviation | Scaled min | Scaled max |
|---|
Experiment history
Challenge progress
Timer is disabled.
Space start or pause
R reset
H hint
Learning concepts
Normalisation
Maps values into a chosen range. It is sensitive to extreme minimums and maximums.
Standardisation
Centres values using the mean and standard deviation. It often supports gradient-based learning.
Robust scaling
Uses the median and interquartile range. It reduces the influence of outliers.
Distance models
KNN, SVM, and K-means can be dominated by features with larger numerical ranges.
Gradient descent
Comparable feature scales can produce smoother optimisation and faster convergence.
Data leakage
Fit preprocessing only on training data. Reuse those fitted parameters for validation and test data.