Loading slide

The second bottle

A viognier arrives, drenched in apricot. It costs 46 dollars.

The weights have not been touched since the last bottle. They are still the ones that priced the plummy cabernet perfectly: cabernet and plummy at 14.7, viognier and apricotty untouched at 8.

Run the new wine through them:

  • cabernet0x14.7=$0
  • viognier1x8=$8
  • plummy0x14.7=$0
  • apricotty1x8=$8

guess $16

16 dollars.

The wine costs 46 dollars. The guess is 30 dollars short.

That large miss does not mean the model has become worse. Before it met the first bottle, it would also have guessed 16 dollars for this wine.

The picture shows why. This is a viognier, so cabernet sits at 0 and the carefully tuned cabernet weight counts for nothing. Plum is also at 0. The two inputs that matter here, viognier and apricotty, still have weights of 8 because the first bottle had nothing to say about either of them.

Fitting the first bottle did not make this prediction worse. It simply changed nothing that this bottle uses.

# citations

Loading contents...

[███████░░░░░░░░░░░][██████████░░░░░░░░░░░░░░░░░░]15 / 41
<back>next