Comments on: Not as Bad as the Best Expert http://dabacon.org/pontiff/?p=1365 Theoretical Musings Thu, 09 Sep 2010 14:06:56 +0000 hourly 1 http://wordpress.org/?v=3.0.1 By: Dave Bacon http://dabacon.org/pontiff/?p=1365#comment-127259 Dave Bacon Mon, 23 Oct 2006 15:23:14 +0000 http://dabacon.org/pontiff/?p=1365#comment-127259 "this only works if you don’t know who the best expert is?" Yes. "And it seems odd that, in the algorithm, since you’re essentially tracking the “record” of the experts, that you wouldn’t eventually know who the best expert is" Well you are "learning" who the best expert is (the one with the higher weight). Of course once you identify the best expert, you could just go with that expert. But the beauty of this algorithm is that you don't need to do this. This comes in handy when, for example, the best expert starts to go wrong. That being said, the success guarantee isn't that great for this algorithm. I'm not sure how it would do for fantasy football, but it would be fun to try. Someone's got to have tried this out for porfolio management, right? “this only works if you don’t know who the best expert is?” Yes.

“And it seems odd that, in the algorithm, since you’re essentially tracking the “record” of the experts, that you wouldn’t eventually know who the best expert is”

Well you are “learning” who the best expert is (the one with the higher weight). Of course once you identify the best expert, you could just go with that expert. But the beauty of this algorithm is that you don’t need to do this. This comes in handy when, for example, the best expert starts to go wrong.

That being said, the success guarantee isn’t that great for this algorithm. I’m not sure how it would do for fantasy football, but it would be fun to try. Someone’s got to have tried this out for porfolio management, right?

Like or Dislike: Thumb up 0 Thumb down 0

]]> By: Joe Fitzsimons http://dabacon.org/pontiff/?p=1365#comment-126999 Joe Fitzsimons Sun, 22 Oct 2006 15:12:18 +0000 http://dabacon.org/pontiff/?p=1365#comment-126999 Sorry, meant to say probability 0.156. So it happens 15.6% of the time. Sorry, meant to say probability 0.156. So it happens 15.6% of the time.

Like or Dislike: Thumb up 0 Thumb down 0

]]> By: Joe Fitzsimons http://dabacon.org/pontiff/?p=1365#comment-126998 Joe Fitzsimons Sun, 22 Oct 2006 15:10:37 +0000 http://dabacon.org/pontiff/?p=1365#comment-126998 Well, presumably if you are taking the weighted average of a number of experts you can do better than any one expert. If 3 experts have a probability of each being right 75% of the time, and these numbers are independant, then each should be weighted the same. If you followed the advice of the bes expert you would be wrong 25% of the time, but taking the average would meaan two experts must be wrong for you to make a wrong decision, which happens with probability 15.6%. So you can do better than any one expert. Well, presumably if you are taking the weighted average of a number of experts you can do better than any one expert.

If 3 experts have a probability of each being right 75% of the time, and these numbers are independant, then each should be weighted the same. If you followed the advice of the bes expert you would be wrong 25% of the time, but taking the average would meaan two experts must be wrong for you to make a wrong decision, which happens with probability 15.6%. So you can do better than any one expert.

Like or Dislike: Thumb up 0 Thumb down 0

]]> By: Ian Durham http://dabacon.org/pontiff/?p=1365#comment-126990 Ian Durham Sun, 22 Oct 2006 14:34:15 +0000 http://dabacon.org/pontiff/?p=1365#comment-126990 Let me clarify something (now that my head is a little clearer - though football once again looms on the horizon since it is Sunday): this only works if you <i>don't</i> know who the best expert is? And it seems odd that, in the algorithm, since you're essentially tracking the "record" of the experts, that you wouldn't eventually know who the best expert is. And, if you do, just pick what he/she picks and you'll do better than the algorithm guarantees. Which all means that I must be missing something. What am I missing? Let me clarify something (now that my head is a little clearer – though football once again looms on the horizon since it is Sunday): this only works if you don’t know who the best expert is? And it seems odd that, in the algorithm, since you’re essentially tracking the “record” of the experts, that you wouldn’t eventually know who the best expert is. And, if you do, just pick what he/she picks and you’ll do better than the algorithm guarantees. Which all means that I must be missing something. What am I missing?

Like or Dislike: Thumb up 0 Thumb down 0

]]> By: Ian Durham http://dabacon.org/pontiff/?p=1365#comment-126668 Ian Durham Sat, 21 Oct 2006 11:35:09 +0000 http://dabacon.org/pontiff/?p=1365#comment-126668 I might have another potential application for this: fantasy football predictions! Seriously, no joke. If you're simply betting on wins and losses in a standard "pick 'em" format, it's no different than "up/down." In this case the experts are the usual prognosticators at ESPN, CNNSI, etc. Now the question is how you could adapt it to a fantasy football <i>league</i> in which you pick players and experts project the rough number of points a given player will get in a given week. Since it's not a binary-like prediction it doesn't seem (at least at first glance) that you could use the algorithm in its present form. I would be curious to see if it could be modified to some form useful for this. Sorry, it's Saturday which means I've got football on the brain... I might have another potential application for this: fantasy football predictions! Seriously, no joke. If you’re simply betting on wins and losses in a standard “pick ‘em” format, it’s no different than “up/down.” In this case the experts are the usual prognosticators at ESPN, CNNSI, etc. Now the question is how you could adapt it to a fantasy football league in which you pick players and experts project the rough number of points a given player will get in a given week. Since it’s not a binary-like prediction it doesn’t seem (at least at first glance) that you could use the algorithm in its present form. I would be curious to see if it could be modified to some form useful for this. Sorry, it’s Saturday which means I’ve got football on the brain…

Like or Dislike: Thumb up 0 Thumb down 0

]]> By: rrtucci http://dabacon.org/pontiff/?p=1365#comment-126030 rrtucci Thu, 19 Oct 2006 16:10:56 +0000 http://dabacon.org/pontiff/?p=1365#comment-126030 Bayesian => cool. quantum? Bayesian => cool.
quantum?

Like or Dislike: Thumb up 0 Thumb down 0

]]>