|
||
Title: Can you help this guy keep the maximal money Post by wonderful on Aug 12th, 2008, 3:08pm A gives B $ 50 and challanges B: - A will think of a number from 1 to 100. Let's call this number x. - B's responsibility is to find out what that number is. - To do so, B will go throught a process of investigating x. In particular, B can speak out a number as long as he gives A $1. If this number is smaller than or equal to x , A will say "It is not correct, give me some money !". If this number is greater than x, A will reply:"You're almost there !" and B has to give A additional $15. The above process goes on until B has enough confidence to tell A what is x. If B is correct he can keep all the reamaining $. Otherwise, B has to invite A to a concert. Can you help B find out the optimal strategy? Have A Great Day! |
||
Title: Re: Can you help this guy keep the maximal money Post by Eigenray on Aug 13th, 2008, 12:37am Does the concert cost less than [hide]$43[/hide]? Actually, I'm not sure what model to use, because [hide]either A isn't very bright, or he must think B isn't very bright[/hide]. |
||
Title: Re: Can you help this guy keep the maximal money Post by towr on Aug 13th, 2008, 1:32am B can break even at least, but in the worst case I don't see how he could end up with more than a few dollars. That $15 penalty really cuts into the profits. |
||
Title: Re: Can you help this guy keep the maximal money Post by Eigenray on Aug 13th, 2008, 3:30am [hideb]C(1) = 0 C(n) = min { max[1+C(n-k+1), 16+C(k-1)], k=2...n }[/hideb] gives C(100) = [hide]43[/hide] as the worst-case cost. Edit: Whoops! |
||
Title: Re: Can you help this guy keep the maximal money Post by wonderful on Aug 13th, 2008, 10:08pm Eigenray, Your approach is very interesting. Can you elaborate on it a little bit? Have A Great Day! |
||
Title: Re: Can you help this guy keep the maximal money Post by Eigenray on Aug 14th, 2008, 1:03am C(n) is the worst-case cost to find x when it is known to lie in an interval of size n, WLOG [1,n]. If you guess k, then either: you pay $1, and you've narrowed it down to [k,n]; or you pay $16, and you've narrowed it down to [1,k-1]. Therefore, if you guess k, you won't have to pay more than max[ 1+C(n-k+1), 16+C(k-1) ] so you pick k to minimize this. Finding the Nash equilibrium is more tricky. A is going to try to pick x to maximize the amount B pays; since B knows this, this strategy might not be optimal. But B can guarantee a profit of at least $7 in any case. Edit: Whoops! |
||
Title: Re: Can you help this guy keep the maximal money Post by towr on Aug 14th, 2008, 2:31am on 08/14/08 at 01:03:49, Eigenray wrote:
If you pay $16, then your number was larger, so x must lie in [1..k-1]; and if you pay $1, your number was smaller, so x lies in [k..n]. |
||
Title: Re: Can you help this guy keep the maximal money Post by Eigenray on Aug 14th, 2008, 2:46am You're right of course. And I had double checked it too, because I thought I had it wrong when I posted it... |
||
Title: Re: Can you help this guy keep the maximal money Post by towr on Aug 14th, 2008, 3:03am c(n) = n+14, for 1 < n <= 18 c(n) = ceil((sqrt(1+8*(n-16))-1)/2) + 30, for n >= 16 (Granted, I haven't proved it yet, but the pattern seems to be there) [edit]Nevermind, it seems to break at 188[/edit] |
||
Title: Re: Can you help this guy keep the maximal money Post by Hippo on Aug 14th, 2008, 6:31am I didn't actually think about it, I just want to note a variant I have sloved 20 year ago ... with costs 1,2 this leads to Fibonacci / golden rate solution. |
||
Title: Re: Can you help this guy keep the maximal money Post by Eigenray on Aug 14th, 2008, 1:19pm Certainly C(n) < 16 lg n... Heuristically, suppose C(n) ~ a*log n, and that the optimal choice of k is k(n) ~ b*n. Then we should have a log n = 1 + a log((1-b)n) = 16 + a log(bn), which gives a log (1/(1-b)) = 1, a log (1/b) = 16 b ~ 0.123 is a root of b = (1-b)16, and a = -16/log b ~ 7.63. So C(n) ~ 7.63 log n ~ 5.29 lg n. Heuristically. |
||
Title: Re: Can you help this guy keep the maximal money Post by Hippo on Aug 15th, 2008, 8:23am Yes it is An=An-1+An-16 recursion with A0=...=A15=1. So http://www.ocf.berkeley.edu/~wwu/YaBBImages/symbols/lambda.gif16- http://www.ocf.berkeley.edu/~wwu/YaBBImages/symbols/lambda.gif15-1=0 ... |
||
Powered by YaBB 1 Gold - SP 1.4! Forum software copyright © 2000-2004 Yet another Bulletin Board |