MangoJ
MangoJ
  • Threads: 10
  • Posts: 905
Joined: Mar 12, 2011
August 27th, 2012 at 9:32:51 AM permalink
I'm currently doing some research calculating strategies of two-player games with imperfect information distribution (for example poker).

Reading about Nash equilbria (in short summary: no player can improve his strategy knowing all other players strategies), I remember a coorporation game from my childhood (and could not solve by means of logic).

The game is as follows: Two players each pick a natural number from 1 to 10, and reveal it at the same time. If both numbers are off by at most 1, both players win.
What is the best strategy for the one player, under the conditions that the other player will play the best strategy (and vice versa) ?

In my naive days my logic was like this:
Obviously you should never pick 1, since 2 is a dominant strategy over 1 (2 will win on others pick 1-3, while 1 will win only on 1-2).
So you can assume that you (and likewise the other player) should never pick 1.
But then, if the other player will never pick 1, 3 is a dominant strategy over 2, since 3 will win on 2-4, while 2 will win on 1-3 (but 1 is never played).
With the same logic going further, n+1 will be a dominant strategy over n, since n-1 is never played.
By this conclusion the best strategy should be 10, which is a contradiction since 9 is always a dominant strategy over 10 - so the initial statement "Obviously you should never pick 1" must be wrong.

Now, what is the best strategy ? I.e. what is the Nash equilibrium for this game, and is it a unique equilibrium ? Should you pick 1 and 10 with non-zero probability although you know that 2 and 9 are a dominant choice ? Maybe the best strategy would be to pick either 5 or 6, but by which logic ? - obvisously you win each time if played by both player, but this is also true for the strategy "always pick 1".
Wizard
Administrator
Wizard
  • Threads: 1493
  • Posts: 26483
Joined: Oct 14, 2009
August 27th, 2012 at 10:17:02 AM permalink
This reminds me a little of a this puzzle:

A drill seargent says to his platoon that one day of the week (Mon-Fri) he will do a surprise inspection. He will choose the day so that the platoon will never know for sure when it happens. What is the probability of the inspection for each day?

To get back to the question at hand, my inclination would be to pick 5 or 6 for the reasons stated above, and the logic of the inspection puzzle at least seems to push away from the extreme ends.
"For with much wisdom comes much sorrow." -- Ecclesiastes 1:18 (NIV)
AZDuffman
AZDuffman
  • Threads: 240
  • Posts: 13952
Joined: Nov 2, 2009
August 27th, 2012 at 10:29:35 AM permalink
Quote: Wizard

This reminds me a little of a this puzzle:

A drill seargent says to his platoon that one day of the week (Mon-Fri) he will do a surprise inspection. He will choose the day so that the platoon will never know for sure when it happens. What is the probability of the inspection for each day?



At the beigining of the week it should be 1/7, after sunday 1/6, and so on. Or so I think, surely I am missing some smarter-than-me-math.
All animals are equal, but some are more equal than others
Wizard
Administrator
Wizard
  • Threads: 1493
  • Posts: 26483
Joined: Oct 14, 2009
August 27th, 2012 at 11:07:08 AM permalink
Quote: AZDuffman

At the beigining of the week it should be 1/7, after sunday 1/6, and so on. Or so I think, surely I am missing some smarter-than-me-math.



First, it was supposed to be a five-day week, for simplicity. You might check this hint, but you won't get full credit for the right answer if you do.

Let's say it is the midnight between Thursday and Friday and the inspection hasn't happened yet. What would be the probability the inspection is Friday?
"For with much wisdom comes much sorrow." -- Ecclesiastes 1:18 (NIV)
Paigowdan
Paigowdan
  • Threads: 115
  • Posts: 5692
Joined: Apr 28, 2010
August 27th, 2012 at 11:19:56 AM permalink
I suppose the boss could add in a factor to the equation: his awareness of the employees/platoon's general sloth or maticulousness could weigh the desicision to accelerate or delay the hardship of the look-see. Lazy underlings get a full-treatment Monday white-glove inspection, and goody-two-shoes workers get a Thursday pat-on-the-head, and a pizza & beer party.
Beware of all enterprises that require new clothes - Henry David Thoreau. Like Dealers' uniforms - Dan.
jc2286
jc2286
  • Threads: 2
  • Posts: 145
Joined: Apr 15, 2011
August 27th, 2012 at 12:04:02 PM permalink
Quote: Wizard

First, it was supposed to be a five-day week, for simplicity. You might check this hint, but you won't get full credit for the right answer if you do.

Let's say it is the midnight between Thursday and Friday and the inspection hasn't happened yet. What would be the probability the inspection is Friday?



The probabilities change as you advance through the week. Your original problem never stated from what point in time we are calculating this.

In general, the probability of inspection occurring on Day X is (time remaining in Day X)/(time remaining in week).

So on Sunday, the probability is the same for each day - 1/5.
At midnight between Mon/Tue, the probabilities are 0 for Mon and 1/4 for each day Tue-Fri.
At noon on Wed, the probabilities are 0 for Mon-Tue, 1/5 for Wed, and 2/5 for each day Thur-Fri.
And so on.

Or am I over-thinking this? (And I looked at the hint, for the record)
MangoJ
MangoJ
  • Threads: 10
  • Posts: 905
Joined: Mar 12, 2011
August 27th, 2012 at 1:56:33 PM permalink
Quote: AZDuffman

Or so I think, surely I am missing some smarter-than-me-math.



the keyword is "you will never know for sure when it happens". If the drill seargent would pick friday, the platoon would know at friday morning that the inspection is scheduled for today. Then it would violate the rule that he will never know. Since it's never friday (by the same reasoning) it's never thursday, and never wednesday, ... and never tuesday... so it must be monday for sure. But then it can't be monday for sure by the premise of the puzzle.


( I heard the puzzle in a version of a prisoner awaiting his execution day somewhere within this week, but the judge promised him he will never know which day it is).
Wizard
Administrator
Wizard
  • Threads: 1493
  • Posts: 26483
Joined: Oct 14, 2009
August 27th, 2012 at 2:59:34 PM permalink
Mango got it right.

For those who still don't give up, the perspective is from early Monday. Yes, I think I've heard the prisoner on death row version as well.
"For with much wisdom comes much sorrow." -- Ecclesiastes 1:18 (NIV)
dwheatley
dwheatley
  • Threads: 25
  • Posts: 1246
Joined: Nov 16, 2009
August 27th, 2012 at 3:07:22 PM permalink
There is no unique Nash equilibrium to the original puzzle. Many mixed & pure strategies can't be improved upon:

example: Player A chooses 3. Player B's best response is any mix of 2,3,4. Now Player A cannot improve by switching. Neither can B. Nash equilibrium found.

It's an interesting paradox, but ultimately, not that interesting a math puzzle.
Wisdom is the quality that keeps you out of situations where you would otherwise need it
ThatDonGuy
ThatDonGuy
  • Threads: 117
  • Posts: 6266
Joined: Jun 22, 2011
August 27th, 2012 at 3:22:23 PM permalink
Quote: MangoJ

I'm currently doing some research calculating strategies of two-player games with imperfect information distribution (for example poker).

Reading about Nash equilbria (in short summary: no player can improve his strategy knowing all other players strategies), I remember a coorporation game from my childhood (and could not solve by means of logic).

The game is as follows: Two players each pick a natural number from 1 to 10, and reveal it at the same time. If both numbers are off by at most 1, both players win.
What is the best strategy for the one player, under the conditions that the other player will play the best strategy (and vice versa) ?


Here's what I get, although my logic is probably suspect:
Let P(1), P(2), ..., P(10) be the probability of you choosing 1, 2, ..., 10 respectively; the sum of the P values = 1.
The probability of winning is the same regardless of what the other player chooses if all ten of these sums are equal:
1: P(1) + P(2)
2: P(1) + P(2) + P(3)
3: P(2) + P(3) + P(4)
4: P(3) + P(4) + P(5)
5: P(4) + P(5) + P(6)
6: P(5) + P(6) + P(7)
7: P(6) + P(7) + P(8)
8: P(7) + P(8) + P(9)
9: P(8) + P(9) + P(10)
10: P(9) + P(10)
#1 = #2 only if P(3) = 0; #9 = #10 only if P(8) = 0
#6 = #7 only if P(5) = P(8), so P(5) = 0
#3 = #4 only if P(2) = P(5), so P(2) = 0
#4 = #5 only if P(6) = P(3), so P(6) = 0
#7 = #8 only if P(9) = P(6), so P(9) = 0
This reduces the sums to #s 1 and 2 = P(1), #s 3-5 = P(4), #s 6-8 = P(7), and #s 9-10 = P(10).
#2 = #3 only if P(1) = P(4)
#5 = #6 only if P(4) = P(7)
#8 = #9 only if P(7) = P(10)
This is possible only if P(1) = P(4) = P(7) = P(10) = 1/4.
The strategy appears to be choose 1, 4, 7, or 10 equally.
Assuming the other player does the same thing, both players must choose the same number to win; this happens 1/4 of the time.
MangoJ
MangoJ
  • Threads: 10
  • Posts: 905
Joined: Mar 12, 2011
August 27th, 2012 at 3:28:52 PM permalink
Well, all those puzzles are all fun to talk about, but what is the solution ?

I'm not a mathematician, but I heard a Nash equilibrium are guaranteed under very general premises.

Using the approach that a expected result of a good strategy should be identical to all of the other players action,
one should pick the numbers 1, 4, 7 and 10 each with same probability 1/4. With this strategy, regardless of the other players strategy, the expected win is always 1/4.

If this is indeed the Nash equilibrium, it suggests some "funny" moves (picking 1 and 10 instead of 2 and 9).
This reminds me of the Cold War, where risking the global atomic war suddenly became an "option".
  • Jump to: