The Sorting Machine

You have collected labeled examples. You have built a dataset. Now comes the exciting part: you hand all of that to a machine, and the machine studies it — and after enough studying, it can sort things it has never seen before! How does that work? Let's find out.

Learning From Examples

Think about how you learned to read. At first, your teacher showed you the letter A over and over — big A, small a, A in different fonts. She told you 'this is A' every time. After many examples, your brain learned the pattern. Now you can recognize A anywhere, even in a handwriting style you have never seen. A sorting machine learns the same way. You show it hundreds of labeled examples. The machine looks for patterns — clues that tell one group apart from another. For apples and oranges, the machine might notice: apples are often red or green, with a smooth skin. Oranges are usually orange-colored and have a bumpy skin. It learns these patterns from the labels.

Once the machine has learned, you can show it a brand-new photo — one that was NOT in the training dataset. The machine looks at the photo, checks its patterns, and makes a guess: 'apple' or 'orange.' This guess is called a prediction. The machine predicts which group the new thing belongs to. If the machine learned well from good labeled examples, its predictions will usually be right. If the labels were messy or there were not enough examples, the predictions might be wrong.

Here is an important thing to remember: the machine is not magic. It does not think like a person or feel anything. It is just very good at finding patterns in numbers and using those patterns to make fast guesses. A human teacher first has to label the examples. Without those labels, the machine has nothing to learn from.