Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

  > The probability of selecting the best applicant in the classical secretary problem converges toward 0.368
There are also some pretty big tradeoffs with this strategy. There seems to be a roughly 1/3 chance that you already rejected the best candidate. That's all good if not coming up with a result is acceptable and practically free of cost.

If the stakes are higher though, more conservative strategies might be in order. For example, you might want to sample a much smaller group than n/e first and make a decision based on their median, or even look at their standard deviation to get a reasonable sense about the value distribution - as in the real world values are seldomly distributed uniformly.

If there is a cost to coming up empty, you may also want to employ a hybrid solution, such as going with "1/e-law of best choice" until you hit the 70% mark and then accept the first candidate that is higher than a linear interpolation of the highest value and zero, lerped across the remaining percentage of samples.



view as:

Legal | privacy