For the probabilitists in the audience

[memnus]

I'd like your thoughts on a problem. (Here, all "randomly chosen" numbers are taken from the same continuous real distribution.) Suppose you have a set S of n randomly-chosen numbers, and a set D, initially empty. Begin by removing the largest number from S and placing it in D, and adding another random number to S. Now take the largest number from S that is smaller than the last number you chose, place it in D, and replace it with a random number. Repeat until you've taken the smallest number from S.

- What is the distribution of numbers in D?
- What is the expected value for the size of D?
- Bonus question (I know the answer to this one!): Why is Brian thinking about this problem?

I'm about to start in on a Monte Carlo estimator of the problem, but I'd like some theoretical input.

click

Comments

[zane314.livejournal.com]

With the problem as defined, you're right (immediately done). With the problem as "stylistically feels right" - you stop when there are no elements in S that are smaller than the item you just took (so this check happens after replacement), the other approach is appropriate.