Computer Science I / The Definite Integral and Area
Practice question · Multiple choice

A model fits its training data almost perfectly and performs poorly on new data. Which fitting concept describes this, and what is the geometric picture?

Hints
  1. Recall what a degree-19 polynomial through 20 noisy points looks like between the points.
  2. Ask what the model learned that it should not have.
Show the answer

A. Overfitting, the model passes through the noise as well as the signal

Why

The model fitted the noise as faithfully as the signal, and noise does not repeat in new data. It is the interpolation-versus-fitting distinction under another name, which is why held-out validation exists, and why a model flexible enough to fit anything has told you nothing by fitting your data.

Read the lesson: The Definite Integral and Area →

Practise The Definite Integral and Area

The app has 5 more questions on this lesson, and keeps your place in the course. Computer Science I is free to start.

More questions on The Definite Integral and Area