Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> General Discussion >> Win on Score
        
(Message started by: deep_blue on Mar 13th, 2015, 5:48am)
Title: Win on Score
Post by deep_blue on Mar 13th, 2015, 5:48am
I saw there was a discussion in the chat today about the Score win rule so maybe we should start (reopen) a thread on this.
Here's my opinion on some of the suggestions:
1. Browni once suggested to let a bot evaluate the position.
I would definitely disagree on this one since it's not plannable how the bot would evaluate the final position thus Arimaa would become a game with no perfect information anymore.
2. Aaaa's reasoned that the rule should be changed becaues of me exploiting the bots.
I also disagree on this one and agree to Fritzlein who meant that this shows that bots just aren't artificially intelligent. Besides it shouldn't be difficult to implement something against Score wins.
3. Kzb had the idea to do something like 50-moves-rule in chess (I guess that means like 50 moves without a capture).
That seems like a fair solution to me but could have some problems. First problem is a general one, we would include draws again. The second problem is that a too short move number might affect some strategical battle in an unnecessary way. If move number is too high that might easily lead to a R exchange every 50 moves or so so games could take days.
But maybe there's a way to make this rule working I haven't thought about before.
4. Aaaa suggested to count the friendly pieces in opponent's half throughout the game or in the end to calculate who's more active.
I see two problems with this: 1. It would be a disadvantage to play a home game and there's no need to restrict the number of playable styles. 2. Such delaying games like me vs. Z tend to have no "little" pieces in the opponents side from both sides that play.
5. Material could be counted like 1. How many rabbits? 2. How many pieces? and so on.
The problem with this is that rabbits have to be one of the first questions since they are really important. But then one also could just trade M vs. r instead of M vs. h like I did.
6. Lightvector had the idea (though probably not being serious) that both lose on score (e.g. in an event game like the WC).
Although it seems strange to have no winner and probably is somewhat unfair I would still consider it to be an interesting suggestion. One could e.g. unrate unofficial games and in the official ones take that rule. But that's probably just too big change in a tournament when both players can lose.
7. My suggestion: keep the score rule how it is now. Eventually change it to adjucate some trades that are obvious. That would mean in case of even rabbits let the side win that has an obvious material advantage (that even a beginner would see) like M vs. h, MH vs. hh, MD vs. hc but NOT e.g. MD vs. hh since that is somewhat unclear.
The problem with that rule is that it doesn't help in that many cases and would even complicate the rules.
8. A new idea I had: Take the static material values (like implemented in a bot) that seem most accurate (also possibly the wiki page on material balance though that should be expanded then) and adjucate using them. To make it fair there could be a counter that shows how the actual value of the position is. (having this also in the plan window where you can try out lines)
The idea of this is that it's avoiding the problem with imperfect information and still is somewhat more fair than the current system. A problem could be if people disagree what's worth more for a certain material balance but it's probably still fairer than the current system.
9. Like already suggested in 2/7: Keep the Score how it is now. It was like this for a long time and humans and bots have to deal with it. If bots learn to deal with the Scoring function which should not be too difficult there's nothing really unfair. A bad material trade that wins on Score is just "advanced strategy" so to say.

What do you think about those ideas? Do you have any other ones?
Title: Re: Win on Score
Post by cadwallon on Mar 13th, 2015, 6:46am
Not sure we are asking the right question here.   In your games v sharp and z, I doubt the scoring function entered the bots' thoughts, as it would have been too far away to contemplate.  It certainly wasn't what caused Sharp to break the deadlock, as it was Silver and would have won had the game been adjudicated by score.
Title: Re: Win on Score
Post by deep_blue on Mar 13th, 2015, 8:17am
As far as I know sharp and Z both don't have anything about Score implemented (that should change of course). I agree that it wasn't Score that lead sharp to attacking me and that game (in my opinion) isn't part of the discussion.
Title: Re: Win on Score
Post by JimmSlimm on Mar 13th, 2015, 8:43am
We can't change it just because bots are being exploited. Keep the score rule as it is, its the bots' problem to deal with the rules of the game.

Semi offtopic: I'd rather change the repetition rule to repeat-once
Title: Re: Win on Score
Post by browni3141 on Mar 13th, 2015, 2:43pm
What I don't like about score is that I feel it is really ugly game design, although other solutions I've proposed are also kind of ugly (although very much less so, IMO)
Double loss is not hard to implement for elimination tournaments. We already deal with it through double forfeits. Double loss seems very fair, except when both players played very well. It would be funny if the perfect game ended in a double loss :P. I think it is not as good as a draw, but a lot better than the current scoring function. I would personally recommend switching to this and be open to further improvement in the future.

Draw by N move rule with accelerating time controls would both allow a fair result and keep the game length manageable, but we should adapt the WC/WCC for the possibility of a draw before changing to something like that.
Title: Re: Win on Score
Post by rbarreira on Mar 14th, 2015, 10:41am

on 03/13/15 at 14:43:40, browni3141 wrote:
Double loss is not hard to implement for elimination tournaments. We already deal with it through double forfeits. Double loss seems very fair, except when both players played very well.


How would double loss work for the challenge / challenge screening games?


on 03/13/15 at 08:17:32, deep_blue wrote:
As far as I know sharp and Z both don't have anything about Score implemented (that should change of course).


Judging from the chat archive during your game vs Z, neither did you apparently ;) (before the game at least).
Title: Re: Win on Score
Post by deep_blue on Mar 15th, 2015, 4:27am
About the double loss: I think it could be an interesting idea (otherwise I wouldn't have suggested it). I don't know if it was the first time that idea popped up or if there were some discussions on it earlier, maybe arimaa veterans like Fritzlein can tell.
For the Challenge one could just not let it count and play another game instead.
The main problem I see in what Browni already wrote: A perfect (or very good at least) game might end in a double loss which doesn't make sense.
The biggest problem I still see in a game in which there's no winner (and let it only be a psychological problem).
Rbarreira, I am not sure what you meant with your second comment. Did you mean my "tripple checking question" that I am the winner? (I was at least 99% sure I would be since I read that exact rule only two days ago and won on Score vs. Sharp2014blitz and Ziltoid2014fast)
Title: Re: Win on Score
Post by rbarreira on Mar 15th, 2015, 5:17am

on 03/15/15 at 04:27:43, deep_blue wrote:
Rbarreira, I am not sure what you meant with your second comment. Did you mean my "tripple checking question" that I am the winner? (I was at least 99% sure I would be since I read that exact rule only two days ago and won on Score vs. Sharp2014blitz and Ziltoid2014fast)


Yeah I meant that.
Title: Re: Win on Score
Post by CraggyCornmeal on Mar 16th, 2015, 5:46pm
A possibility that hasn't been mentioned: simply remove the time limit on games. This would prevent players from tacitly agreeing to play passively and wait to hit the limit. If a game doesn't have a predetermined stop time, the players will search for opportunities to attack.

I also like the double loss idea. Removing the time limit stops rewarding passivity, but the prospect of a double loss more actively encourages aggression. I don't share Browni's concern that a perfectly played game could end in a double loss. When a game is decided by score, it usually means the players mutually abandoned the idea of attacking, not that they played sterling defense.

The problem with deciding a game by score is that it rewards you for grinding the game to a halt. The deep_blue/bot_Z game is the most boring game of Arimaa I've ever looked through. I congratulate deep_blue for finding a novel way to to defeat bot_Z. But this style of play undercuts the spirit of the game, so I think we should amend the rules to discourage it.
Title: Re: Win on Score
Post by browni3141 on Mar 16th, 2015, 7:46pm

on 03/16/15 at 17:46:04, CraggyCornmeal wrote:
I don't share Browni's concern that a perfectly played game could end in a double loss. When a game is decided by score, it usually means the players mutually abandoned the idea of attacking, not that they played sterling defense.

Assuming the colors are almost exactly even, it is very likely that a perfect game of Arimaa will have a very high move count, far exceeding the point at which score is used right now.

Score can also come into play when both (non-perfect) players are trying to win and even if one of the players is aggressive and some kind of stalemate position is reached. Camel frames and double away frames, both the result of aggression, can become near stalemate positions. In a highly aggressive attacking game of mine I had gotten into a position which was AFAIK a stalemate as the result of a blunder, but the other player broke it and went on to lose.

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=316691
Look at 23s and beyond, and especially after 29g after my opponent eliminated any plans of bringing pieces through the trap for himself.

Note that in a game like this is would be even worse to have no time limit. It would become a contest of waiting out the opponent.

I believe this game might have gotten more attention if it had actually ended on score. Perhaps the scoring function is intrinsically self-preserving. People don't care to worry about it because it "never" comes in to play, but when it does come into play one of the players ends up breaking the stalemate because the rules would have him shuffle for hours to claim a victory, and the game goes unnoticed :P

I think that positions likely to end on score are almost never reached because both players gave up on trying to win. It's just not the nature of humans to do that, especially not both players. Bots certainly have, but I don't care so much about bots being stupid.

For the record, I don't think deep's method of beating Z counts as novel, since I beat one of Z's ancestors this way :P
Title: Re: Win on Score
Post by deep_blue on Mar 16th, 2015, 10:10pm
I agree with browni that some score rule is necessary. There are indeed too many positions that are so stalematet that they would never end without score (except on play "blunders" to play on but that shouldnt be right too).
You may call winning on score boring, I disagree and call it one of many many playing strategies/styles.
My Screening wins weren't showing the score rule has to change, it showed the bots have to change.
If (what I think) perfect play is longer than score limit then there would be a problem with double defeat. If it was not like this (and we eventually found out who loses to give him the "score right") then there would be no point in delaying since one would lose eventually if the oponent plays the correct moves (and delaying doesn't pose any problems to the oponent...).
Title: Re: Win on Score
Post by rbarreira on Mar 17th, 2015, 4:24pm

on 03/16/15 at 17:46:04, CraggyCornmeal wrote:
A possibility that hasn't been mentioned: simply remove the time limit on games.


This seems like the simplest and most elegant solution indeed. It's not like games are getting too long and stalemated with good play, so there's no reason to worry about never-ending games. Even bot vs bot games don't last for very long.

Edit - also, if a game between humans is stalemated and they're encouraged to attack by the score function,  they should get similarly encouraged by the fact they'll fall asleep eventually.
Title: Re: Win on Score
Post by supersamu on Mar 17th, 2015, 5:19pm
I dislike a function that determines the winner when the function can theoretically give the win to a player who is most definitely losing. (Like browni demonstrated is possible in a game against briareus, giving up a goal in 1 seconds before the game time limit was reached (  http://arimaa.com/arimaa/gameroom/comments.cgi?gid=261898 ))

But I would also dislike it if a game is a draw one move before a player has a goal in 1 available, which can also happen if we have an arbitrary cutoff.

- We could have a committee decide whether to continue the game for x more moves (The committee would decide again and again until the game is drawn or someone has won)

- We could have accelerating time controls and a trigger that needs to be activated (for example a capture) for the game to continue for x more moves and not draw.

I would not like an infinite game, because then the win could go to the player that has less IRL obligations.
Title: Re: Win on Score
Post by kzb52 on Mar 17th, 2015, 5:30pm
Just linking to some things for reference:

The game where browni won on score (be sure to look at the ending)

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=261898

The game in the screening where deep won on score

http://arimaa.com/arimaa/gameroom/comments.cgi?gid=331230

Also, many of these ideas (accelerating time controls, alternative scoring functions, human adjuticators, etc. )  were discussed in this forum thread.  I encourage everyone to look through it before continuing the discussion

http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=siteIssues;action=display;num=1366359189
Title: Re: Win on Score
Post by CraggyCornmeal on Mar 17th, 2015, 5:43pm

on 03/16/15 at 19:46:29, browni3141 wrote:
Score can also come into play when both (non-perfect) players are trying to win and even if one of the players is aggressive and some kind of stalemate position is reached.

Good point. I think this is the most common application of the score rule in games that don't involve novice players or bots.

The pertinent question is, how do we want to conclude games that reach a stalemated position? Do we want some way of discerning who the victor is, or do we want to encourage the players to break the stalemate?

I think the biggest virtue of deciding games by score is that when the time limit nears, the losing player should naturally become more aggressive. If Z knew it was about to lose to deep_blue, surely it would have waged an attack late in the game.

However, I'm still concerned about the incentives for the player who is ahead by score. As the time limit nears, their incentive switches from reaching goal to eating up as much of the clock as possible. I would hate to see a game decided by score because the winning player continually uses the maximum time for each of their moves late in the game, preventing the other player from having enough moves to try to break the stalemate.

So if you guys aren't inclined to dramatically change the rule by removing the time limit or by awarding a double loss, I have a more modest proposal. Instead of a game having a time limit of, for example, four hours, I think we should give each player a personal time limit of two hours. When the player who is losing by score reaches their personal two hour limit, the game ends and they lose. However, if the player who is winning by score is the first to reach their personal time limit, the game continues until the losing player either reaches their personal time limit or pulls ahead in score.

The player who is ahead by score would no longer be incentivized to eat the clock. If you're down by score and nearing your personal time limit, you can move quickly each turn to keep the game going, and it doesn't matter how much or little time your opponent uses.

A similarly modest change would be to decide a game by score after x moves instead of after x hours.

Whether the score rule changes or remains the same, I think our goal should be to minimize the incentives players have to let the game sit in a stalemated or stagnant position.


on 03/16/15 at 19:46:29, browni3141 wrote:
I think that positions likely to end on score are almost never reached because both players gave up on trying to win. It's just not the nature of humans to do that, especially not both players. Bots certainly have, but I don't care so much about bots being stupid.

I agree about bots. The deep_blue/Z game illustrates Z's shortcomings a lot more than it illustrates the shortcomings of the score rule.

I disagree about human nature, though. Chess games often end when both players give up on winning and agree to a draw. Arimaa largely avoids this pitfall of human nature by making draws impossible.


on 03/16/15 at 19:46:29, browni3141 wrote:
Assuming the colors are almost exactly even, it is very likely that a perfect game of Arimaa will have a very high move count, far exceeding the point at which score is used right now.

If this is true, it would be a shame for the score rule to cut short a perfectly played game of Arimaa.


on 03/16/15 at 22:10:56, deep_blue wrote:
You may call winning on score boring, I disagree and call it one of many many playing strategies/styles.

Both are true. Your strategy against Z was definitely effective, but the game was definitely boring. When Z gets some code telling it about the score rule, your shuffling strategy will stop being effective and such boring games will stop happening.

As a side note, I like how the repetition rule encourages stalemated positions to break up. In the game browni talked about, arimaa_master may have made some mistakes, but there are several times where the repetition rule gets applied, providing opportunities to break the stalemate.
Title: Re: Win on Score
Post by Fritzlein on Mar 18th, 2015, 12:32pm
I will participate in the discussion in order to try to clarify the reason why I think it is fruitless to have the discussion now, when only games involving bots are ever decided on score.

When we talk about the ideal solution, there are many things we would like to preserve.  We would like any imposed result to be the same as if the game were played to a natural conclusion.  We don't want games to be decided by who has more free time to outlast the other player (no game time limit), but also don't want it to be decided by who can input moves faster (accelerating to sudden death time controls).  We don't want draws, because it raises the specter of agreed draws in positions that aren't stalemates, and makes elimination tournaments awkward to run.

Perhaps we can't have everything we want.  In particular, if and when stalemated positions start to crop up with regularity, we might not be able to think of any way to avoid draws that seems natural enough, and therefore we might go about incorporating draws into the rules and into our tournament structure.  But it is critically important how stalemated positions come about.  It might be that we have no qualms awarding the win to the player who was trying to win, and saddling the player who was just trying not to lose with a loss.  It might be that in all practical cases of stalemate, there is a simple score function that matches our intuition of who is more deserving of a win.  We might in that case decide that imposing an "unnatural" result can be done in a way that is a lesser evil than the evil of allowing draws.

Some proposals are talking past each other, because they are trying to solve two different problems.  If the current score function fails because the game was tending to a natural outcome, and the score picks the wrong winner, the notions of "just keep playing", "accelerate the time control", or "make a more sophisticated score function" come into play.  On the other hand, if the current score function fails because neither player can make progress, i.e. because the position is stalemated, then the above solutions don't apply.  The natural outcome is a draw, and the relevant question is whether we can think of a way to impose a decisive result that is less bad than allowing draws.

At present it is inevitable that the discussion will run around in circles, because the score function hasn't been relevant in enough human vs. human games to give us the necessary experience.  We don't know WHICH problem we are trying to solve, and if it is the stalemate problem that is the true issue, we don't know WHETHER there is a satisfactory way to arbitrarily impose a winner.  We don't know the answers now, because we can't know the answers now.  We won't know what to do until the problem becomes a real problem instead of being essentially a theoretical problem.

Once we have a real problem on our hands, we will have data to inform the discussion and guide our choice of which solution is the least unsatisfactory.
Title: Re: Win on Score
Post by Fritzlein on Mar 18th, 2015, 12:43pm
Just as an example of how data might change our minds, it might turn out some day that stalemates start cropping up in high-level human games, and we notice that in every single stalemate it is critical that the players have two horses.  We might then decide that the best solution is to play tournament games with only one horse on each side!  That is to say, we might even be able to solve the problem of the score function in a way that neither imposes draws nor imposes victory on a stalemated position, but which instead restores the current situation of stalemates never happening in practice.
Title: Re: Win on Score
Post by browni3141 on Mar 18th, 2015, 4:02pm

on 03/18/15 at 12:32:48, Fritzlein wrote:
I will participate in the discussion in order to try to clarify the reason why I think it is fruitless to have the discussion now, when only games involving bots are ever decided on score.


Is that some careful wording? It allows you to ignore games which were likely negatively influenced by the current rules, such as the game of mine I linked to.


Quote:
When we talk about the ideal solution, there are many things we would like to preserve.  We would like any imposed result to be the same as if the game were played to a natural conclusion.  We don't want games to be decided by who has more free time to outlast the other player (no game time limit), but also don't want it to be decided by who can input moves faster (accelerating to sudden death time controls).  We don't want draws, because it raises the specter of agreed draws in positions that aren't stalemates, and makes elimination tournaments awkward to run.

Agreed on all counts, except that "we don't want draws" assumes that we all agree that the downsides of introducing draws outweigh the benefits. I still agree that the reasons listed are among the reasons against introducing draws for the majority of us.


Quote:
But it is critically important how stalemated positions come about.  It might be that we have no qualms awarding the win to the player who was trying to win, and saddling the player who was just trying not to lose with a loss.  It might be that in all practical cases of stalemate, there is a simple score function that matches our intuition of who is more deserving of a win.  We might in that case decide that imposing an "unnatural" result can be done in a way that is a lesser evil than the evil of allowing draws.


I personally think (at least at the moment) it is completely unimportant how stalemated positions come about. At least for me, in a true stalemate there is only one type of result which matches my intuition. A stalemated position is inherently equal, and an equal position deserves an equal result.


Quote:
At present it is inevitable that the discussion will run around in circles, because the score function hasn't been relevant in enough human vs. human games to give us the necessary experience.  We don't know WHICH problem we are trying to solve, and if it is the stalemate problem that is the true issue, we don't know WHETHER there is a satisfactory way to arbitrarily impose a winner.  We don't know the answers now, because we can't know the answers now.  We won't know what to do until the problem becomes a real problem instead of being essentially a theoretical problem.


I object to this argument. The facts that score influencing a game is a rare occurrence and it is difficult to know what the best answer is makes lessening the problem neither undesirable nor impossible at present.

I have been trying to operate under the assumption that we are trying to solve/lessen BOTH problems. Any solution that lessens either or both problems without worsening the other problem or introducing new ones is strictly better than the current scoring function.

I see no reason why we can not come up with something superior to what we have now, and there is no reason we can't later change to something even better. I think that most of the proposed solutions are superior to the current one; it is only a matter of choosing one and implementing it. I would be okay with choosing something I disagree with if it is better than the current scoring function.
Title: Re: Win on Score
Post by Fritzlein on Mar 18th, 2015, 6:13pm

on 03/18/15 at 16:02:04, browni3141 wrote:
Is that some careful wording? It allows you to ignore games which were likely negatively influenced by the current rules, such as the game of mine I linked to.

The reason I am ignoring the HvH games (about three so far?) that didn't end on score but where stalemate became a real consideration at some point is not that they are irrelevant, but that they are insufficient.  I believe I could formulate a score rule that handles all those existing cases properly.  Not perfectly, of course, since we have to give up something, but well enough.  I would award the win to the side that deserved it more in some sense, because I value the absence of draws in Arimaa, and I am willing to add some inelegance to the rules, assuming it applies only very rarely, to keep all games decisive.

The reason I don't propose such a solution is that I am confident you will criticize it on hypothetical grounds.  You would say that my solution would have undesirable consequences in situations which have not arisen yet and may never arise.  And how could I respond?


Quote:
Agreed on all counts, except that "we don't want draws" assumes that we all agree that the downsides of introducing draws outweigh the benefits. I still agree that the reasons listed are among the reasons against introducing draws for the majority of us.

No, I am certainly not assuming that the disadvantages of introducing draws outweigh the benefits.  On the contrary, I am reasonably likely to become an advocate of introducing draws if stalemates become common and no band-aid seems adequate to stop them.  All I am assuming is that draws are undesirable, and that undesirability should be balanced against other factors.


Quote:
I personally think (at least at the moment) it is completely unimportant how stalemated positions come about. At least for me, in a true stalemate there is only one type of result which matches my intuition. A stalemated position is inherently equal, and an equal position deserves an equal result.

Are you staking out an absolute position here?  Namely that any inelegance added to the rules to prevent draws is inherently unacceptable, no matter how minor the inelegance or how seldom it occurs?  It seems like you might be doing what you suggested I was doing before, except that you are doing it on the other side of the argument.

Anyway, even if your position is absolute now, there might be reasons to soften your stance.  Consider the rule that you can't undo the move that your opponent just made.  That rule is inelegant, and it obviously exists only to prevent draws.  Without that rule, many positions (I estimate 1% or so) would be naturally stalemated.  To be consistent with an absolute position that "a stalemated position is inherently equal, and an equal position deserves an equal result," you would have to insist that we repeal this rule and accept all the draws that occur as a result because a draw is the only result that such games deserve.

Conversely, if you think that having a "no undo" rule is an acceptable price to pay to go from about 1% of games ending in draws down to a nearly undetectable level (less than 0.1%, anyway), then I think you have to entertain the possibility that there might be some additional rule that is an acceptable price to pay for eliminating the few remaining draws that the "no undo" rule didn't catch.


Quote:
I object to this argument. The facts that score influencing a game is a rare occurrence and it is difficult to know what the best answer is makes lessening the problem neither undesirable nor impossible at present.

I agree, it is neither undesirable nor impossible to lessen the problem.  But the fact that draws are extremely rare makes it very difficult to decide how to fix the problem and also makes the benefit of changing anything negligible.  The cost/benefit ratio is too high.


Quote:
I have been trying to operate under the assumption that we are trying to solve/lessen BOTH problems. Any solution that lessens either or both problems without worsening the other problem or introducing new ones is strictly better than the current scoring function.

I see no reason why we can not come up with something superior to what we have now, and there is no reason we can't later change to something even better. I think that most of the proposed solutions are superior to the current one; it is only a matter of choosing one and implementing it. I would be okay with choosing something I disagree with if it is better than the current scoring function.

Go right ahead and devise a superior score function.  I won't stand in your way.  I don't oppose changing the scoring function to something better; I merely think it is not worth the effort.  (And by the same token, I didn't think it was worth it for Omar to change from the old score rule to the new one, even though I understand his reasons for doing so.)

I will, however, provide vigorous opposition if you argue that we should introduce draws right now.  In that case the minuscule benefit would have to be weighed against a very tangible harm of having to change our tournament structure.  Not that it is the end of the world to have to deal with draws in our tournaments, but why impose the cost on ourselves with 100% probability when the benefit is so uncertain?
Title: Re: Win on Score
Post by CraggyCornmeal on Mar 18th, 2015, 10:34pm
Let's try not to let this discussion become contentious. I hate when fascinating debates get tainted by adversarial impulses.

I agree with Fritz that we lack a sufficient sample size of stalemated games to write an optimal replacement for the score rule. But that doesn't make discussing it any less intriguing.

Arimaa is a near-perfect game. The necessity of resolving stalemated games (however rare they may be) is one of its few flaws. And I think the best way to understand something is to analyse its flaws. I don't know about everyone else, but that's why I'm keeping up with this discussion. Yeah, it would be nice to improve the score rule, but dissecting the limits of a great game is what rivets me.

...

Are stalemated positions inherently equal? Or can we say that one player is winning, despite the fact they lack an opportunity to capitalize on their advantage?

I'm curious what everyone's intuitions are. Browni and Fritz have touched on this question, but our answer will form the ethical foundation for what kind of score rule we want, so I think it deserves a fuller discussion.
Title: Re: Win on Score
Post by browni3141 on Mar 19th, 2015, 12:26am

on 03/18/15 at 22:34:56, CraggyCornmeal wrote:
Let's try not to let this discussion become contentious. I hate when fascinating debates get tainted by adversarial impulses.

I wouldn't say that's what is happening between me and Fritzlein, although I usually can't tell when I've stepped on his toes until he yells at be for it :P
I still view this as more of a rational-minded discussion than a heated argument, and I hope Fritzlein does also.
Title: Re: Win on Score
Post by browni3141 on Mar 19th, 2015, 1:21am

on 03/18/15 at 18:13:04, Fritzlein wrote:
No, I am certainly not assuming that the disadvantages of introducing draws outweigh the benefits.  On the contrary, I am reasonably likely to become an advocate of introducing draws if stalemates become common and no band-aid seems adequate to stop them.  All I am assuming is that draws are undesirable, and that undesirability should be balanced against other factors.


If this occurs we have bigger problems to worry about than draws :)

My only concern with draws I personally give any weight to is how they would break our current tournament format. I actually like draws. This isn't very much related to the current discussion, but I would argue that necessarily decisive games have the tiny flaw that the result of perfect play by both sides is a not a draw (the same could be true if draws were possible, but necessarily decisive games make this necessarily true). Even if this never actually happens, I find this feature of a game very aesthetically pleasing. I am likely a minority in this respect, though.


Quote:
Are you staking out an absolute position here?  Namely that any inelegance added to the rules to prevent draws is inherently unacceptable, no matter how minor the inelegance or how seldom it occurs?  It seems like you might be doing what you suggested I was doing before, except that you are doing it on the other side of the argument.


I mean that a non-equal result, be it draw, double loss, etc. for a stalemate is against intuition, and therefore inherently distasteful, but not necessarily unacceptable.


Quote:
Anyway, even if your position is absolute now, there might be reasons to soften your stance.  Consider the rule that you can't undo the move that your opponent just made.  That rule is inelegant, and it obviously exists only to prevent draws.  Without that rule, many positions (I estimate 1% or so) would be naturally stalemated.  To be consistent with an absolute position that "a stalemated position is inherently equal, and an equal position deserves an equal result," you would have to insist that we repeal this rule and accept all the draws that occur as a result because a draw is the only result that such games deserve.

Conversely, if you think that having a "no undo" rule is an acceptable price to pay to go from about 1% of games ending in draws down to a nearly undetectable level (less than 0.1%, anyway), then I think you have to entertain the possibility that there might be some additional rule that is an acceptable price to pay for eliminating the few remaining draws that the "no undo" rule didn't catch.


For what it's worth (hardly anything) I would estimate much lower than 1% of positions would change value with repetition being a draw. Maybe 1% of positions' best moves would change, though.

If I had designed Arimaa, I would have made three-fold repetition a draw by similar reasoning as I would have stalemate positions be considered drawn. However, by now I don't think I would like the game being changed so much. The fact that score affects games a lot less than the current repetition rule makes it score much easier to change without messing too much with the game I love.

Also, this is a pretty minor point, but I view the method by which repetition is handled as somewhat more acceptable than the method by which games exceeding the GTL are handled. Three-fold repetition is an impossibility (at least by the server's implementation of the rules for humans), as opposed to a scenario requiring adjudication. Although still a fairly arbitrary rule modification, it would be interesting if we could similarly make stalemate positions impossible, but this seems practically impossible.

The more I think about it, the more I think I would support changing the rules to make three-fold repetition a draw. I'm not so opinionated on this that I feel the need to actively advocate a rule change, though, for reasons mentioned above.


Quote:
I agree, it is neither undesirable nor impossible to lessen the problem.  But the fact that draws are extremely rare makes it very difficult to decide how to fix the problem and also makes the benefit of changing anything negligible.  The cost/benefit ratio is too high.


Score is insurance. The negative result is very unlikely, but extremely undesirable when it happens without a good insurance policy to protect us. I view the benefit of improving on the score function as significant. If we don't have any insurance the worst case scenario is very bad. I think right now the worst case is not as bad as it would be without the score function, but still bad enough that it deserves attention.


Quote:
I will, however, provide vigorous opposition if you argue that we should introduce draws right now.  In that case the minuscule benefit would have to be weighed against a very tangible harm of having to change our tournament structure.  Not that it is the end of the world to have to deal with draws in our tournaments, but why impose the cost on ourselves with 100% probability when the benefit is so uncertain?


I don't plan on doing that in the near future. The day I do, if I do, I would like to be able to support my argument with a new tournament format already designed.

How do you feel about using double loss, also an "equal result?" I really don't like the fact that perfect or mutually good play could then theoretically result in a double loss, but think I prefer it over the scoring function since is doesn't award a win arbitrarily in a true stalemate position. I also wouldn't really like applying double loss in a non-stalemate position where even an ugly score function is more probably fair.

On a side note, what would happen if a double forfeit occured in the last round with each player having one life remaining prior?
Title: Re: Win on Score
Post by Boo on Mar 19th, 2015, 3:51am

Quote:
We can't change it just because bots are being exploited. Keep the score rule as it is, its the bots' problem to deal with the rules of the game.

Semi offtopic: I'd rather change the repetition rule to repeat-once


+1
Title: Re: Win on Score
Post by deep_blue on Mar 19th, 2015, 10:28am
I disagree with browni's idea of a draw being possible with perfect play. The same discussion also was there when draws were abolished. The result was that it is very likely that perfect play wouldn't end in draw anyway.
I think Craggy's idea with each player having an own "score clock" is interesting and should be considered. So far I don't see a disadvantage compared to the current rule while all others have at least some disadvantage.
I really don't like the idea of a draw. I always found it great that arimaa is a game which never can end in a draw and I don't think we should change it for such unimportant things like Score rule. Also I could imagine theoretically two players agreeing to draw in the WC in the finals when money is at stake.
@semi-off-topic: I would keep three fold repetition since I don't see a reason to change it.
Title: Re: Win on Score
Post by browni3141 on Mar 19th, 2015, 3:04pm

on 03/19/15 at 10:28:53, deep_blue wrote:
I disagree with browni's idea of a draw being possible with perfect play. The same discussion also was there when draws were abolished. The result was that it is very likely that perfect play wouldn't end in draw anyway.


I think you are misremembering. Maybe you should provide a link to the thread where this was supposedly concluded. IIRC what was happening in practice was poor play ending in a draw, but this has absolutely no bearing on what would happen with perfect play.

Regarding the separate idea of moving to "repeat once," I think this would be a mistake. The current rule allows players to deviate after the repetition, which makes a generally difficult scenario more forgiving.
Title: Re: Win on Score
Post by deep_blue on Mar 19th, 2015, 3:16pm

on 03/31/09 at 11:48:00, omar wrote:
This is exactly what I used to think. Ideally I wanted a game to allow for draws but with a very low margin. Arimaa fit this pretty well with the rule that if both sides lose all the rabbits the game is a draw. Draws were possible but pretty rare. However, in practice I was finding that for elimination type tournaments (like the finals of our world championship tournament) I did not want there to be any possibility of draws, because another game has to be played to break the draw and that makes the tournament schedule less predictable. So for tournaments we were already using the rabbit elimination rules to not allow draws.

I was pretty content with this situation because I knew that ideally Arimaa would be a draw, but for tournaments we could make it drawless. Then one day Karl suggested to me that even though Arimaa ideally allows for draws a perfectly played game of Arimaa might always be win for one player or the other. A solved game that fits this is Connect4. It allows for draws and well played games often result in draws, but a perfectly played game is always a win for the first player. So just because a game allows for draws doesn't mean that a perfectly played game will be a draw. I'd never be able to prove that Arimaa is not like Connect4 (without solving it :-) ), so why bother carrying two sets of rules when in practice you want the game to be drawless. That convinced me to make Arimaa a drawless game.
Title: Re: Win on Score
Post by browni3141 on Mar 19th, 2015, 4:51pm

on 03/19/15 at 15:16:16, deep_blue wrote:


Thank you, although this does not support your previous post. No one makes a claim of the likelihood of perfect play resulting in a draw, only remarking that games which allow draws are not necessarily drawn with perfect play, which is what I said a few posts ago :)
Title: Re: Win on Score
Post by Fritzlein on Mar 19th, 2015, 5:39pm

on 03/19/15 at 00:26:47, browni3141 wrote:
I still view this as more of a rational-minded discussion than a heated argument, and I hope Fritzlein does also.

Indeed I do; I'm glad I'm not stepping on your toes so far. :)
Title: Re: Win on Score
Post by CraggyCornmeal on Mar 19th, 2015, 6:48pm
Suppose we amend the rules of Tic-tac-toe: if neither player achieves three in a row, the player with the most marks on the board wins.

This would be absurd. Tic-tac-toe needs draws because the game's natural, perfectly played conclusion is a stalemate.

Omar's old post shows that he used to believe Arimaa is like Tic-tac-toe. He wanted draws to be possible because it would be absurd to reward perfect play with a loss when the natural conclusion of Arimaa is a stalemate.

But Connect Four shows that it's not necessarily absurd to issue a loss to a player who has played perfectly. Because the natural conclusion of Connect Four is a victory for the first player.

Thus, it is absurd to reward perfect play with a loss if and only if the game's natural, perfectly played conclusion is a stalemate.

If Arimaa is like Tic-tac-toe, we need draws to be possible. If it's like Connect Four, screw draws.

The problem is that, unlike Tic-tac-toe and Connect Four, we have no solid evidence for what the outcome of a perfectly played game of Arimaa is. Until we learn what kind of game we're dealing with, we should not base our arguments on an assumption that it is one kind or the other. Browni's argument that draws should be possible is premised on an unproven assumption.

We know scarcely anything about perfect Arimaa play. When Omar recognized it might be like Connect Four instead of Tic-tac-toe, he began basing rule changes on what we actually know about the game: how imperfect play unfolds in practice.
Title: Re: Win on Score
Post by browni3141 on Mar 19th, 2015, 7:51pm

on 03/19/15 at 18:48:07, CraggyCornmeal wrote:
Browni's argument that draws should be possible is premised on an unproven assumption.

This is true, but my argument is of course not completely undermined, as there are other reasons to bring draws in than the result of perfect play.
Title: Re: Win on Score
Post by lucky81 on Apr 8th, 2015, 5:15pm
My suggestion would be this: apply the tiebreaker after 200 moves from each side, instead of after a set time limit.

This would at least remove all timing considerations from the pure logical analysis of the game, thus making the game definition more mathematically pure.

Currently, which side has winning strategy might theoretically depend on time control settings. Even worse, a viable strategy is: wait until 1 second before the game timer expires to make your last capture, opening a goal in one for the opponent. Hope that the opponent does not have time to make his response. Game outcome depends on which side can send moves to the server with more precise timing during the last second of the game. My suggestion would eliminate these problems.

200 moves would mean that at 60s/5m time control, the longest possible game would last 6 hours 50 minutes. That's if both players use up all their available time.
Title: Re: Win on Score
Post by clyring on Apr 8th, 2015, 5:20pm
Some of the server documentation suggests this should be possible with the 't' suffix instead of a time in the game length limit field, but this functionality was never actually implemented and the server mostly reads '120t' as 120 hours instead of 120 turns.


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