reckless intuitions of an epistemic hygienist ([info]gustavolacerda) wrote,
@ 2006-08-02 15:55:00
Previous Entry  Add to memories!  Tell a Friend!  Next Entry
Entry tags:machine_learning

Rock-Paper-Scissors
People are bad at being random, i.e. unpredictable. Randomness is a depletable resource. I'd like to look at Machine Learning algorithms that learn to exploit such patterns.

This one is intelligent. Still, I've been managing to maintain a small lead, which is close to the best I can expect if I beat their expectations about patterns the average person would exhibit. ("You have won 56, lost 48, and tied 46 games". What is the probability of such a favourable outcome for if this were random?). One known bias is that people think they are being random when they avoid repeating the same choice. The machine will try to take advantage of this.

This one did not learn to exploit some predictable players.


(Post a new comment)


[info]rdore
2006-08-02 09:24 pm UTC (link)
On playing the "intelligent" one 60 times:

"You have won 28, lost 14, and tied 18 games."

Of course I played by trying to think about what sorts of statistical things I might be doing and break them. Doing the stats fairly halfassed: getting that many wins on 42 nonties, sigma is sqrt(42*.5*.5) or something like 3.2, which would put me just over 2 sigmas up, which is a few percent. So not impossible but a bit unlikely. Got bored of playing to get more data.

(Reply to this) (Thread)

[info]gustavolacerda
2006-08-02 10:16 pm UTC (link)
s.d. = sqrt (sample size), right?

(Reply to this) (Parent)(Thread)

[info]rdore
2006-08-02 10:51 pm UTC (link)
for a Bernoulli process (i.e. counting the number of success when you keep repeating the same test, where each test is independent), the standard deviation is:

sqrt(n * p * (1-p))

where p is the probability of success.

(Reply to this) (Parent)

[info]williamallthing
2006-08-02 10:06 pm UTC (link)
roughly 20%. so, sadly, not below the traditional "this wasn't random" 5% cutoff.

(Reply to this) (Thread)

[info]williamallthing
2006-08-02 10:10 pm UTC (link)
sorry, that should be 17%.

you'll need to win 61 or more times (out of 150) to qualify for the "not sheer luck" award.

(Reply to this) (Parent)

[info]ronebofh
2006-08-02 11:03 pm UTC (link)
What is the result with the intelligent one if you play 150 games where you make the same play each time?

(Reply to this) (Thread)

[info]gustavolacerda
2006-08-02 11:11 pm UTC (link)
I did that about 30 times, and won about 1, lost about 15, drew about 15.

(Reply to this) (Parent)

[info]marknau
2006-08-03 12:20 am UTC (link)
Good ol' reliable rock...

(Reply to this) (Parent)(Thread)

[info]ronebofh
2006-08-03 12:26 am UTC (link)
Nuthin' beats that!

(Reply to this) (Parent)

Create an Account
Forgot your login or password?
Login w/ OpenID
English • Español • Deutsch • Русский…