Re: Real tough problem.....pls try...
- From: David Kinny <dnk@xxxxxxxxxxxxxxxx>
- Date: Thu, 06 Jul 2006 01:32:18 GMT
In <44ac481b$0$4015$9a6e19ea@xxxxxxxxxxxxxxxxxxxx> Chris <spam_me_not@xxxxxxxxxx> writes:
arun.madras@xxxxxxxxx wrote:
You have 100 to start with and 10 bets to make. Each bet turns on theWhat is the significance of the fact that the Oracle may lie after
result of a coin flip. The oracle will tell you which way the coin will
fall, but may lie on just one occasion and may do so after seeing your
bet for that flip. Placing an x dollar bet will return 2x dollars if
the oracle is correct, 0 dollars if wrong. Come up with the strategy
which ensures you the largest possible amount (that is, the strategy
which whose worst-case scenario is as good as possible).
seeing your bet? Does the Oracle want to hurt you or help you? Or is the
Oracle's choice about when to lie entirely random?
And what do you mean "may lie"? Is it possible that the Oracle won't lie
at all?
I agree, the problem spec is unclear about "after seeing your bet".
If one's bets are to be "assisted" by the oracle, then the
oracle must make its pronouncements prior to the bet, not after.
Perhaps he means that one must declare the bet amount, wait for
the oracle to speak, and then decide between heads or tails?
Also, are fractional bet amounts permitted? Zero bets ok?
In any case, it's clear that after the oracle has lied once, one
can "bet the house" on each subsequent flip, doubling the stake.
Prior to a lie, one must withhold some amount in order to survive
the loss incurred on a lie. If that fraction is X in (0,1), then on
a lie one is left with fraction X of one's bank, otherwise with 2-X.
So the best result for the player consists in choosing suitable X,
given the worst-case future return available at each step.
Consider the situation where there are N rounds left, and the oracle
has not yet lied. If N = 1, one should bet nothing, since in the
worst case one will then lose and have no further chance to recover.
For N = 2, one should bet 1/3 of one's bank, either getting a return
of 4/3, or losing and then doubling on the last bet to achieve 4/3.
For N = 3, bet 1/2, winning for an ultimate return 3/2 * 4/3 = 2, or
losing and doubling twice to achieve exactly the same (worst) return.
For N = 4, bet 3/5, for an ultimate return of 16/5.
....
For N = 10, bet 9/11, for an ultimate return of 1024/11.
Generalising, bet the fraction 1 - 2/(N+1) when N rounds are left,
for an ultimate return (multiplier) of 2^N/(N+1) over the entire
sequence of bets. Of course, this must turn into some clumsy
approximation if only integer bet values are permitted.
David
.
- Follow-Ups:
- Re: Real tough problem.....pls try...
- From: harry389@xxxxxxxxx
- Re: Real tough problem.....pls try...
- References:
- Real tough problem.....pls try...
- From: arun . madras
- Re: Real tough problem.....pls try...
- From: Chris
- Real tough problem.....pls try...
- Prev by Date: Re: Real tough problem.....pls try...
- Next by Date: Re: Real tough problem.....pls try...
- Previous by thread: Re: Real tough problem.....pls try...
- Next by thread: Re: Real tough problem.....pls try...
- Index(es):
Relevant Pages
|