← All tools
EDTOOLKITS GUIDES

Choose complexity using held-out error

By Team Edtoolkits · AI learning

Compare training and test curves before calling a detailed fit an improvement.

Use the guide alongside the working tool.

Open Overfitting and Underfitting →

Explore the lesson

Watch three curves try to fit the same dots.

Slide the complexity and compare training error with test error.

Flip the memory card and answer the quick check.

A comparison to try

Increase polynomial degree and find a setting where training error falls while test error rises. Inspect the curve between the dots as well as at them.

Read the result

A flexible model can fit sample noise. The smallest training residuals therefore need not produce the best unseen predictions.

A mistake to avoid

Do not repeatedly tune against a final test set in a real project. Use validation for selection and reserve independent evaluation for the finished choice.

Teaching model and privacy

A teaching version: it fits polynomials to a small made-up dataset. Real models are bigger, but the pattern of training error falling while test error rises again is the same. Everything runs in your browser.

Ready to apply this workflow?

Try Overfitting and Underfitting →

Explore a related workflow

All practical guides →

Found unclear instructions or an unexpected result? Contact Team Edtoolkits.

YOUR BROWSER, YOUR CHOICE