random.sample with large weighted sample-sets?

[Ben Finney]

Tim Chase <python.list at tim.thechases.com> writes:

> I'm not coming up with the right keywords to find what I'm hunting.
> I'd like to randomly sample a modestly compact list with weighted
> distributions, so I might have
>
>   data = (
>     ("apple", 20),
>     ("orange", 50),
>     ("grape", 30),
>     )

That's not a list, it's a tuple. I think you want a list.

When you want a sequence where each position has a semantic meaning, use
a tuple (such as ‘("apple", 20)’). Each item has a meaning *because of*
the position it's in; if the items were in a different order, they'd
mean different things.

When you want a sequence where the positions don't have a special
meaning – each item means exactly the same no matter if you change the
order – that's sometimes called a “homogeneous” sequence, and you want a
list.

So a “record” should be represented as a tuple, and a “table” of records
should be represented as a list of tuples:

    records = [
            ("apple", 20),
            ("orange", 50),
            ("grape", 30),
            ]

> and I'd like to random.sample() it as if it was a 100-element list.

The implication being, I suppose, that you'd like the number in each
tuple to be a weighting for the probability of choosing that item.

For probability weightings, you should arrange for the weightings to sum
to 1 (instead of 100 in your example). Then each weighting is simply the
desired probability of that item, and those values will work with
various libraries that deal with probability.

> What am I missing? (links to relevant keywords/searches/algorithms
> welcome in lieu of actually answering in-line)

You're looking for a “probability distribution” and “weighted choice”.

Hope that helps!

-- 
 \     “This world in arms is not spending money alone. It is spending |
  `\      the sweat of its laborers, the genius of its scientists, the |
_o__)           hopes of its children.” —Dwight Eisenhower, 1953-04-16 |
Ben Finney