Arimaa Forum (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
Arimaa >> Site Discussion >> Hiding Predictions
(Message started by: 99of9 on Oct 27th, 2004, 12:43pm)

Title: Hiding Predictions
Post by 99of9 on Oct 27th, 2004, 12:43pm
Although I find it very interesting to see what everyone else is predicting, I think it would be better if that was hidden until the game started and my vote was locked in.  That way I can't rely on everyone else's wisdom, and must do the predicting for myself.

Title: Re: Hiding Predictions
Post by Fritzlein on Oct 27th, 2004, 12:54pm
I agree with 99of9.  I think the aggregate predictions shouldn't be made public until everyone's prediction is locked in.  Particularly when there are so few people making predictions, it is possible to infer a great deal about what other predictors are doing.  To really play the prediction game to the fullest under the present system, I actually ought to enter outrageous predictions (for concealment) and change them at the last minute.  But this is risky, and I might not be available at the last minute to change them.

Title: Re: Hiding Predictions
Post by omar on Oct 27th, 2004, 5:29pm
I was also contemplating this. You guys have helped me finalize the decision. I've changed it so that the perdictions will not be viewable until after the game starts.

Title: Re: Hiding Predictions
Post by 99of9 on Oct 27th, 2004, 6:58pm
Thanks Omar.

Title: Re: Hiding Predictions
Post by 99of9 on Nov 1st, 2004, 9:38am
I think that this system is good enough to use even on the early rounds of the world championship.  Because Fritz's least-squares method makes underdog wins even more risky, those preliminary rounds are just as interesting prediction-wise.

Guessing the gamelengths of lopsided contests could be even more interesting now that the winner is aiming for a win in the lowest possible number of moves.

Title: Re: Hiding Predictions
Post by Fritzlein on Nov 1st, 2004, 12:15pm
I also think that the prediction system as it stands is good enough to implement for the World Championship prediction contest.  All games are interesting to predict, whether close or lopsided.

However, I would sound one note of caution:  If I have predicted myself to win by 55%, then the difference to me between winning and losing is 40 prediction points.  But if the game goes longer than 70 moves, I would rather lose than win, in terms of prediction points.  For example, losing in 71 moves gets me (71 - 21) = 50 prediction points, whereas winning in 71 moves gets me (100-71) + 19 = 48 prediction points.  So suddenly we have reintroduced an incentive to lose, albeit only in very long games where you bet as lightly as possible on yourself.  To iron out this wrinkle, I suggest that the winner get at least 30 points for for the moves portion, and the loser get at most 70 points, no matter how many moves it takes for the game to end.

It hasn't come up yet in in the World Cup, but I'll go out on a limb and predict at least one 70-move game in the World Championship, and it's better to resolve this kind of thing in advance than change midstream when there is money on the line.

Title: Re: Hiding Predictions
Post by 99of9 on Nov 1st, 2004, 4:03pm
But as we can see in the EU match, a handful of points hardly matters.  Remember that the WC is knockout - surely the small money prizes for prediction are not worth getting yourself knocked out for, no matter how slim you think your chances of winning are.  And if they're really that slim, how did you manage to duke it out for 71 moves anyway!?

Although I'd be very happy if the game winners were given more bonuses, I don't think it's necessary, and it would seem less clean.

Title: Re: Hiding Predictions
Post by omar on Nov 1st, 2004, 4:35pm
What you say might be OK for the WC tournament, Toby, but in general I have to agree with Karl and I think we should fix it so that losing never gives more points than winning. I don't mind it if the equations are not so clean and elegant. I think sometimes we have to accept exceptions.

I've changed the scoring function so that if the move count for the game is more than 70 it is reset to 70 before computing the score.

Title: Re: Hiding Predictions
Post by 99of9 on Nov 4th, 2004, 5:26am
Anyway, the important thing is not about the points obtained for winning vs points obtained for losing.  It's actually the *relative* points obtained for winning vs *relative* points obtained for losing.  Where *relative* is relative to the rest of the field of predictors.

I could easily throw a game and get WAY ahead in a prediction contest.  All I've got to do is wait until I'm against a really easy opponent, so that everyone will be betting high stakes on me.  Then bet low and throw the game.  Ok, sure, I lose 21 points or whatever, but everyone else except my opponent loses up to 300.

I think you've just got to make the incentives for winning games much higher than the incentives for winning prediction contests.  Then this all becomes a non-issue.

Nevertheless, I don't think your cap to 70 moves hurts much.  It just means that once you get to 70 moves there is no incentive to win quickly.

Title: Re: Hiding Predictionsc
Post by omar on Nov 6th, 2004, 11:08pm
I wonder if maybe we should not allow the players to predict at all on their own games and use the average predictions of confidence and number of moves on that player as that players perdictions. The points given for confidence and number of moves can then be calculated in the same way as for other perdictors. That way in order to get the most prediction points the player has to win and try to win in the number of moves predicted the other players.

Title: Re: Hiding Predictions
Post by 99of9 on Nov 7th, 2004, 9:16am
But what if everyone else predicts the other player to win?

Title: Re: Hiding Predictions
Post by Fritzlein on Nov 7th, 2004, 11:24am
99of9's point about the relative points for winning is serious.  Even if it is worth more points to win than lose, someone could indeed strategically throw a game in the knowledge that it will hurt himself less than it would hurt the other predictors.

On the other hand, if we excessively raise the prediction contest incentives for winning, or raise the penalties for losing, then the prediction contest favors the better players even more than it already does.

I think it is reasonable to leave the system the way it is at present, but if we do change it to address 99of9's concern, then I propose that the best fix would be for the winner and the loser of a game to get the same number of prediction points as each other, and that that number be the average of the points that the non-participants get.  That way a player can't control how well they do against the field; they will always be relatively average.


Title: Re: Hiding Predictions
Post by 99of9 on Nov 7th, 2004, 11:32am

on 11/07/04 at 11:24:18, Fritzlein wrote:
On the other hand, if we excessively raise the prediction contest incentives for winning, or raise the penalties for losing, then the prediction contest favors the better players even more than it already does.


Oh, sorry, I didn't mean increase the prediction contest incentives for winning.  I actually meant the normal incentives for winning (eg the prizes in the main tournament should be more than those in the prediction contest - which they are).



Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB © 2000-2003. All Rights Reserved.