Bias-Variance Playground
Published:
Built an interactive Plotly Dash playground that lets learners dial up sample size, feature count, correlation, and noise to watch bias-variance tradeoffs unfold. I built it as a learning aid so the user could see why regularization matters when high-dimensional regression teeters on the edge of instability.
Highlights
- Scenario presets that recreate common scenarios: comfortable OLS, edge-case OLS vs Ridge, underdetermined LASSO, and high-correlation comparisons.
- Live controls for
n,p, regularization strength, model choice, feature correlation, noise, and random seeds. - Real-time charts for error curves, coefficient shrinkage, and predicted vs actual fits, plus warnings when OLS breaks (p ≥ n).
- Snapshot panel and CSV export so users can capture configurations and bring them into follow-up analysis.