Underfitting and overfitting
Ten noisy training points and three held-out test points are fit with polynomials of degree 1, 3, and 9. The animation tracks training and test error side by side, showing how degree 1 underfits with high error on both sets, degree 9 memorizes the training points but fails on test data, and degree 3 balances fit and generalization. Useful for students and teachers introducing model selection, bias-variance tradeoff, and why test error matters more than training error.
16:9 · every frame verified for overlaps, spacing and edges before rendering
The prompt that made it
Ten noisy training points and three held-out points; polynomial fits of degree 1, 3 and 9 with training and test error: degree 1 underfits, 9 overfits, 3 generalises best. Judge a model on unseen data.
Make the next one in this series
- Show how cross-validation picks the best polynomial degree automatically
- Visualize the bias-variance tradeoff curve as model complexity increases
- Demonstrate regularization shrinking the degree-9 polynomial toward a simpler fit
