Election post-mortem

[patio11]

I had to do a statistical analysis of the results of the election challenge in order to collect my winnings from my coworkers. My side bet, aside from playing, was that "Funding x 2, TV Blitz x 4" would beat other stratgies. It ended up winning in four out of five clubs.

Detailed move lists follow (yeah, parsing these out was a pain).

? = uncertain of move
( ) = tourbus
* = preceding move gaffed
f = fundraising
t = tv
w = no move
r = retail campaign (used once, and that was one time too many, this move was terrible)
m = media apperance
n = negative



shea_d mftmft
lwbeauclair mftt*mn
isamuel ff*tttn (won)
steinguitar fttftt
speaknsmile wwwwww

(The question marks there include, mathematically, at least one negative attack).



apda jonathan mf*(tt*)ft?
madgenius fmttr?*
jstbuch wwwft?
tinytall t*fm+(ft)mt (won)
travo f*fwwww

+ He actually got overcharged for it by one million dollars.


rlayne tf*tf(tt*)?
alan tauber mftmrw
ccbrat fft*t*tt (won)
vegetathalas w*w(ff)t*t?

brendo fft*tt?
patio11 fttftt (won)
uscglenn wwwwww
dog day prophet wwft*n*?

arabian knight (ff)tttt?
syphonhail (ff)tt*ttn
stanford fftt*t?* (won)
db8 missinglink fttft?*
pfuhrman w*ffttt


Patrick McKenzie

[Edit: Fixed 5 ?s on basis of posts later in thread.]

[syphos]

My final move was a negative attack, Seeing as I was in last place, I figured I would take out second to last and see what would happen.

[thedancingbear]

Several of us went negative on Darryl in the final move -- I think that is the _only_ use of that action, is combining it against a single person, because otherwise, your net change in a heads-up comparison is on average less by going negative than TV blitzing.

IS

[patio11]

I agree. Its only purpose is to allow kingmaking (I was very surprised we didn't see some triple GNs after move 3 was posted, because there were some people sitting on a lot of cash and not reasonably in the running). Otherwise the fact that the gaffe rate is *four* times that of TV Blitz (expected value = -2 points) moots most of the gain you can expect from it, especially since it gives free points to random people (the only real winners from the action).

Patrick McKenzie

[thedancingbear]

Can't be that hard. Finite set of moves, defined range of payoffs. You'd need to kinda play with genetic or learning stuff but it wouldn't be that hard.

Set up a program to generate an initial population of say, 5 random 6-move sequences. (Approximates five players.) Run the game a hundred times or so, that should take about a second. Evaluate fitness; combine their "DNA" (move sequence) to create the next generation. Repeat until your population looks kinda homogenous.

Ideally suited for situations where modelling is hard; also appeals to the soul of the engineer by being more practical and less about multiplying.

IS

[patio11]

There are only about two things which would make it hard to model:

a) Don't give a complete description of the rules of the game.
b) Vastly increase the capability of players to screw with each other.

Increasing the number of moves, possibilities per move, number of players, randomness, etc just makes me waste more of the department's vast computational resources on the problem.

I did two genetic simulations of this week's challenge. The first resulted in Ralph Nader genociding the rest of the population. I tweaked the payoff for tieing, and then the dominant strategy popped out pretty fast. Total CPU time used: fourty seconds. Thats approximately 100,000 generations of political deathmatches. The simulation of a set of moves program took me about fourty minutes to write, because I hate C. The genetic algorithm took 5, because I wrote it in gawk and gawk is the best programming language ever (I would have written the simulator in gawk, too, except I need more C practice and writing it in gawk would have clobbered performance, since the simulation is the bottleneck).

Patrick McKenzie

[Madgenius]

I was the one who used the retail campaigning. I had 9 million left with two moves left and a bus tour to burn. I realized the best thing to do would be to wait until the last move to do a double Media appearance. This means I has to spend 2 million or less on the next to last move. Might as well do something with it...

[thedancingbear]

Consider: fundraise instead and you'd probably make enough for a double TV blitz.

[patio11]

Expected value of a r(mm): +2 - 1 + 5.5 - 1 + 5.5 - 3.5 = 8.5
Expected value of a f(tt): +0 - 1 + 8.5 - .5 + 8.5 - 3.0 = 13.5

I'm quoting those numbers off the top of my head. Still, pretty clearly the advantage goes to the blitzkrieg.

Patrick McKenzie

[steinguitar]

just to clarify, my strat was steinguitar fttftt.

f(2) t(4) is definitely the best, but the ordering is key. the reason Ian beat me wasn't because of the moves used (aside from his GN at the end, we had the same strat). The key was to appear innocuous for as long as possible. by going fftttt, the payoffs are the same but the lead comes later. thus, people are less likely to target you for GN campaigns. Of course, I didn't think of this til after I'd lost - my pre game planning consisted of figuring out that gaffe deductions were not proportional to possible payoffs, so one had to "go big or go home." I perm'd.

[Brendo]

I did what Patrick told me would work. It kinda worked.

Then I ate pie.

[tinytall]

My strategy was down the tube at the first gaff. From then on, I went with my gut. Just fyi, my last move was a TV blitz. (I think)

You think I can claim that million in the next go-round of this challenge?

[Cait Ross]

I went fundraising 2x, then the rest TV blitzs. I risked the gaffe in the final move nad did a bus tour with 2 blitzs. I didn't really plan in advance, though.

[Vegetathalas]

That would had been my strategy had I not forgotten what day it was and missed out. Whoops.

[Vassar]

FYI: My strategy was FFTTT(TT). I planned out all six moves before the game even started, and only checked in once during the game.

My reasoning was simple: two fundraisers plus five television blitzes ought to, statistically speaking, do better than any other combination. I put the two fundraisers at the beginning to avoid my jumping out to a quick lead. I wanted others to be at the top at the beginning, in case people decided to go negative on the top dog.

-M

P.S. Your post said four out of five clubs followed the "Fx2, Tx4" strategy, but beneath it, you show only three people following the aforementioned strategy (ISam and Tinytall being the exceptions). Actually, technically speaking, only two out of five winners followed that strategy (I did Fx2, Tx5).

[Vassar]

Originally posted by patio11
I did two genetic simulations of this week's challenge. The first resulted in Ralph Nader genociding the rest of the population. I tweaked the payoff for tieing, and then the dominant strategy popped out pretty fast. Total CPU time used: fourty seconds. Thats approximately 100,000 generations of political deathmatches. The simulation of a set of moves program took me about fourty minutes to write, because I hate C. The genetic algorithm took 5, because I wrote it in gawk and gawk is the best programming language ever (I would have written the simulator in gawk, too, except I need more C practice and writing it in gawk would have clobbered performance, since the simulation is the bottleneck).

Jeez...You put far too much work into this. To figure out the dominant strategy, all it takes is paper, pencil, two minutes of your time, and a knowledge of very basic statistics.

-M

[thedancingbear]

Working prototypes are better.

IS

[Vassar]

Originally posted by ISamuel
[b]Working prototypes are better.

Not necessarily. Whereas there will certainly be instances where they are superior, it would be foolish to make the blanket statement that this is always the preferable means.

In this particular instance, with so few possibilities that can easily be worked by hand, it just makes more sense to do it by hand. With working prototypes, you need a large sample size to ensure the best outcome. And, even with a very large sample size, there's still a chance (albeit a small one) that you won't come out with the optimal strategy.

Why not just do it by hand, when it's both quicker and more effective? You're guaranteed to figure out the optimal method every time.

-M