Think Together · Start with a Short Story
When telling fruits apart, the girl looks at shape, color, and size. A robot also needs clues that can be used in calculations.
Lay Out the Process
- Real-world objectFruit
- Useful cluesColor, shape, weight
- Model calculationCombine the clues
- PredictionA result such as a category
A feature is information a model can use in its calculations.
People may prepare some features, such as weight. A model may learn others from the original pictures. A useful feature should relate to the task, not just happen to appear in the training examples.
For example, the color of the tablecloth in a photo usually should not be a main clue for deciding what kind of fruit is shown.
Now You Try
If every apple is placed on a red plate, which feature might the model mistakenly rely on?
Ready? Open this to see the explanation
The plate’s color. The model might fail when the plate is blue, so check whether it learned clues that truly relate to the target.
One Idea to Take Away
A feature is information a model can use in its calculations.
Try covering the diagram above and explaining each step to someone in your family. If one step is hard to explain, return to that card and look again.