Hacker News new | past | comments | ask | show | jobs | submit login
The German tank problem (wikipedia.org)
64 points by RiderOfGiraffes on Dec 3, 2016 | hide | past | favorite | 6 comments

The trick to understand frequentist approach here is, it doesn't matter what number the tanks have other than the one with max number. 1 2 3 60 is same as 57 58 59 60 ( like it doesn't change the probability to win lottery with ticket 000001 or 639465 )

So the knowledge is only you have 4 tanks with max ID=60.

That's a large part of it, the other half is picking N numbers from a range 0-100% what is the reasonable minimum and maximum percentile for the largest number. Next, what happens when you keep adding new numbers to chose from over time.

Calculate the mean and double it seems like a decent initial estimate. In the example, that value is 80.5.

That could easily result in a number smaller than the maximum value observed, which would be impossible.

I think with more sample size, "median" can be better choice.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact