Bias–Variance Tradeoff Explorer

How to use

1. Click Generate to sample a new train/validation set from a noisy sine.
2. Move the Degree slider to change model complexity; Lambda adds smoothing.
3. Click “Compute Curve” to plot Train vs Val error across degrees; the “Best degree” marks the sweet spot.

What to observe

• Low degree: high bias → both Train and Val errors are large (underfit).
• High degree: low bias but high variance → Train error drops, Val error rises (overfit).
• Middle range: Val error is lowest → best generalization.

Tips

• Increase Noise to make the task harder; the sweet spot shifts to smaller degrees.
• Increase Lambda to smooth the fit and reduce variance at high degrees.
• Change Train/Val sizes to see how more data stabilizes validation error.