What will it do?

Before running anything, it is worth committing to a hunch.

Here is the machine again, and nothing about it has changed. The same four units on the left, the two grapes and the two tastes. One weight on each. Multiply every unit by its weight, add the four results, and that sum is the model's guess at the price.

The rule for correcting it has not changed either. When a guess turns out to be wrong, each weight moves by error x unit x learning rate, exactly as it did all through the last chapter.

Every weight starts at 20, for the same reason they all started equal last time. Equal weights are the machine saying it has no idea yet which of the four matters. The particular number is not important. They could start at 8, as they did last chapter, or at nothing at all, and everything that follows would happen just the same. The machine would end up equally wrong, by exactly the same amount. Only the number of passes it took to get there would change.

The only thing that is different is the wine.

You have the rule for these four wines already, and it fits in one sentence: matched pairs are dear, mismatched ones are cheap. So:

Will the machine get there too, given a few hundred bottles?