Step three: the real price

The guess is 12 dollars.

Now the bottle gets turned around, and the price is read out.

The wine costs 22 dollars.

This is the first new information supplied after the model makes its guess. The tasting record also came from outside the model, but it did not say what the answer should be.

Twenty two dollars is a fact the model had no way of working out on its own. Somebody had to say it.

That is the arrangement this whole chapter runs on. Examples arrive with correct answers attached. The ordinary idea comes first: show the bottle, let the model guess, then reveal the target. The technical name is .

A bottle with no price attached cannot provide this particular correction. Other kinds of machine learning use different sources of feedback, but this chapter follows labelled bottles and prices.

Step three: the real price

The guess is 12 dollars.

Now the bottle gets turned around, and the price is read out.

The wine costs 22 dollars.

This is the first new information supplied after the model makes its guess. The tasting record also came from outside the model, but it did not say what the answer should be.

Twenty two dollars is a fact the model had no way of working out on its own. Somebody had to say it.

A bottle with no price attached cannot provide this particular correction. Other kinds of machine learning use different sources of feedback, but this chapter follows labelled bottles and prices.