One Knob, and What It Is Really Selling You
Last timeFinding Without Looking at Everything
Every approximate index has a single dial trading accuracy against time. Where to set it is an arithmetic question, and the answer is usually lower than instinct suggests.
The previous lesson ended with a structure that sometimes fails to find the right
piece. The obvious question is how often, and the useful question is how much it
would cost to make that number better.
Both are answerable, and the first one is answerable without any labelled data at
all, which makes it unusual in this field.
What is being measured
Exhaustive comparison gives the correct answer by construction. So you already
own a reference: run a sample of questions both ways, and compare.
- the k pieces the index returned for question j
- the k pieces exhaustive comparison returns, which are the true nearest
- questions in the sample, a thousand being plenty
- the share of the true nearest that the index actually found, averaged over questions
One point about this measurement is worth dwelling on, because it is a rare
luxury. Almost everything else in retrieval needs somebody to decide what the
right answer was, and such decisions are slow, expensive and contested. Here the
right answer is defined by a program you already have. That makes this the one
number in the pipeline you can know exactly, which is probably why it receives
more attention than its importance deserves.
The lesson stops here
4 more paragraphs to go
You have read the opening. The rest of the argument, the problems that check whether it landed, and the lines worth keeping at the end all come with a plan.
The first lesson of every course in the library reads the whole way through, free, so you can see exactly what the rest of them are.
See the planThe contentsThis is the reading half
Starting the course gives you your own copy of it. Every idea on every page has problems standing under it, marked with a reason rather than a tick, and any sentence you do not believe can be opened and argued with. None of that can happen on a page nobody owns.
The contents