Dice Game

[ChromePony]

This could probably go in the probability forum, but theres way too much philosophy around here and not enough math so lets give it a shot.

A friend of mine posed this to me after he was asked it on a job interview, heres the game, its pretty simple.

You roll a fair 6 sided die and are given an amount of money equal to the number you roll, 1=$1 2=$2...etc. The caveat is that if you wish to reroll you may, but would then be stuck with whatever you get the second time.

The question is, how much would you be willing to pay to play this game? read: what is the EV
(this is what he was asked to figure out on the spot)

A slightly less trival extension, what if you get up to 6 rolls, meaning you can choose to reroll 5 times before you are stuck with the 6th.
(he was given this as 'homework')

Finally and most interestingly, how about generalizing it for n rolls...just for kicks

[durron597]

Clearly if you get a 1-3 you should reroll and if you get a 4-6 you should keep. If the game was played with only one roll the $EV is 3.5. Thus the EV of this game is:

1/6 * (4) + 1/6 * (5) + 1/6 * (6) + 1/2 * (3.5) = $4.25

[ChromePony]

[ QUOTE ]
Clearly if you get a 1-3 you should reroll and if you get a 4-6 you should keep. If the game was played with only one roll the $EV is 3.5. Thus the EV of this game is:

1/6 * (4) + 1/6 * (5) + 1/6 * (6) + 1/2 * (3.5) = $4.25


[/ QUOTE ]

Very nice, but thats the easy part. I'm guessing most people who hang out around here could get that pretty quick, I'm more interested in the n-generalization since I have an answer but its not all that elegant, wondering if someone can come up with a better approach.

[durron597]

Well, I thought about it for awhile. The answer is definitely 5 - 3/(2^n), though I'm having trouble proving it. I'll think about it some more.

[Moozh]

If you had infinite rolls, wouldn't you expect to end up with a 6?

[durron597]

[ QUOTE ]
If you had infinite rolls, wouldn't you expect to end up with a 6?

[/ QUOTE ]

But you don't have infinite rolls. You have a certain number of rolls and then you stop.

Hm. Maybe my optimal strategy is not, in fact, optimal. If you have 20 rolls left and you roll a 5, my strategy says stop. That is clearly wrong, but it's where my answer comes from.

[Jordan Olsommer]

Can the general solution even be expressed mathematically? It seems like it requires a conditional ("if current roll is < EV of replacing, replace. Lather, rinse, repeat."). It could be easily done as a recursive function, though.

FWIW, According to my calculations, for n=20 rolls you should replace anything lower than a six for the first 15 rolls, then take a 5 or 6 for rolls 16-18, then 4, 5, or 6 on roll 19, and of course whatever you can get on roll 20.

[Jordan Olsommer]

I still don't know how to collapse it down to a simple mathematical statement*, but this will do it if you have a python interpreter handy (n = max number of rolls you're allowed)

<font class="small">Code:</font><hr /><pre>
import math

def EV(n):
if n==1:
return 3.5
else:
ev_replace = EV(n-1)
threshold = math.floor(ev_replace)
ev_keep = (21 - (((threshold)*(threshold+1))/2))/6
ev = (threshold / 6)*(ev_replace) + ev_keep
return ev
</pre><hr />

Here are the results for the first 20 values of n:

3.5
4.25
4.66666666667
4.94444444444
5.12962962963
5.27469135802
5.39557613169
5.49631344307
5.58026120256
5.6502176688
5.708514724
5.75709560333
5.79757966944
5.8313163912
5.859430326
5.882858605
5.90238217084
5.91865180903
5.93220984086

and here's what your EV is if you are given 100 rolls

5.99999997385

*edit: Duh. It was staring me right in the face, but it's really ugly, and I don't like being stared at by things that are ugly, so that's probably why I didn't include it before. It's just a recursive function with a couple floors thrown in, same as in the code snippet

F(1) = 3.5
F(n) = ( floor( F(n-1) ) * F(n-1) ) / 6 + ( 21 - ( floor( F(n-1) )* (floor ( F(n-1) + 1)) / 2 ) ) / 6

I'm quite sure I have the wrong number of parens in there, but oh well.

[Jordan Olsommer]

And since the max edit time is up on my last post, and because the recursive function I included at the end looks like a keyboard barf, I made it look perty:

http://tinypic.com/b3lgg6.jpg

[BruceZ]

[ QUOTE ]
You roll a fair 6 sided die and are given an amount of money equal to the number you roll, 1=$1 2=$2...etc. The caveat is that if you wish to reroll you may, but would then be stuck with whatever you get the second time.

The question is, how much would you be willing to pay to play this game? read: what is the EV

[/ QUOTE ]

The average for 1 die is 3.5, so keep 4,5,6 (avg = 5) and roll 1,2,3 again. The EV is

(1/2)*5 + (1/2)*3.5 = $4.25 for 2 dice.


[ QUOTE ]
A slightly less trival extension, what if you get up to 6 rolls, meaning you can choose to reroll 5 times before you are stuck with the 6th.

[/ QUOTE ]


EV(n) = EV for n rolls.
Work backward from 2 remaining rolls.
Since EV(2) = 4.25, keep 5+ (avg = 5.5) and roll 1,2,3,4:

2 rolls left, keep 5+, EV(3) = (1/3)*5.5 + (2/3)*4.25 = 4 + 2/3

3 rolls left, keep 5+, EV(4) = (1/3)*5.5 + (2/3)*(4 2/3) = 4 + 17/18

4 rolls left, keep 5+, EV(5) = (1/3)*5.5 + (2/3)*(4 + 17/18) = 5 + 7/54

5 rolls left, keep 6, EV(6) = (1/6)*6 + (5/6)*(5 + 7/54) = 5 + 84/325 =~ $5.27 for 6 dice.


[ QUOTE ]
Finally and most interestingly, how about generalizing it for n rolls...just for kicks

[/ QUOTE ]

For n &gt; 6, EV(n) = (1/6)*6 + (5/6)*EV(n-1)

recursion: EV(n) = 1 + (5/6)*EV(n-1)

EV(6) = 5.27
EV(7) = 1 + (5/6)*5.27
EV(8) = 1 + 5/6 + (5/6)^2 * 5.27
EV(9) = 1 + 5/6 + (5/6)^2 + (5/6)^3 * 5.27
...
EV(n) = sum [i = 0 to n-7](5/6)^i + (5/6)^(n-6) * 5.27

from finite sum geometric series:

EV(n) = [1 - (5/6)^(n-6)]/(1 - 5/6) + (5/6)^(n-6) * 5.27

EV(n) = 6 - 6*(5/6)^(n-6) + 5.27 * (5/6)^(n-6)

EV(n) = 6 - 0.73*(5/6)^(n-6), for n &gt; 6 dice.

[ChromePony]

Looks like you worked through this a lot like I did, but I like your presentation here, makes it nice and easy to follow. I had it down as a messy recursion type deal, but that last line there

[ QUOTE ]
EV(n) = 6 - 0.73*(5/6)^(n-6), for n &gt; 6.

[/ QUOTE ]

seems to sum it up nicely in the closed form I was looking for.

[BruceZ]

[ QUOTE ]
Looks like you worked through this a lot like I did, but I like your presentation here, makes it nice and easy to follow. I had it down as a messy recursion type deal, but that last line there

[ QUOTE ]
EV(n) = 6 - 0.73*(5/6)^(n-6), for n &gt; 6.

[/ QUOTE ]

seems to sum it up nicely in the closed form I was looking for.

[/ QUOTE ]

So I get the job?

[Jordan Olsommer]

[ QUOTE ]
EV(n) = 6 - 0.73*(5/6)^(n-6), for n &gt; 6 dice.

[/ QUOTE ]

One-upped again!

I swear Bruce, if I wasn't such a nice guy, you'd totally be my arch-nemesis by now. /images/graemlins/tongue.gif

[hmkpoker]

For the first situation:

If I roll 1-3, I will re-roll, as my expectation on a random roll (3.5) is higher than 1-3.

If I roll 4-6, I will stand, as my expectation on a random roll (3.5) is lower than 4-6.

So 50% of the time I will re-roll and stand with that number (on average 3.5), and the other 50% of the time I will stand with 4-6 (on average 5).

50%(3.5) + 50%(5) = 4.25

If I pay $4.25, that's break-even odds. So the gamble is profitable as long as I am paying less than that.

(took me a minute and a half, I so could have done that on the spot ^_^)

[hmkpoker]

What job does your friend have anyway?

[Thythe]

[ QUOTE ]
What job does your friend have anyway?

[/ QUOTE ]

Dice roller.

[ChromePony]

[ QUOTE ]
What job does your friend have anyway?

[/ QUOTE ]

Its some sort of investment finance stuff but he didnt get too specific, perhaps a back alley dice game manager or something.

I havent heard from him since, but it seems that he got called back if they gave him 'homework'...I wonder if answering the first one right was enough to impress them.

[ChromePony]

[ QUOTE ]
So I get the job?

[/ QUOTE ]

Hired!