Author |
Topic: Max game time rules (Read 2086 times) |
|
Migi
Forum Senior Member
Arimaa player #4643
Gender: 
Posts: 26
|
 |
Max game time rules
« on: Sep 5th, 2011, 6:25pm » |
 Quote  Modify
|
I am currently implementing time control and time management for my bot, and decided to read up on the rules so I don't make any wrong assumptions, and I was surprised to read the following rule:
Quote:| If a game is stopped due to reaching the time control limit then the winner is determined by scoring the game as follows: the player who currently has or most recently had more pieces left after the completion of a turn wins. Otherwise the player to move second (silver) wins. |
|
This was quite shocking to read! It seems to me like this can easily be exploited: if your bot is playing silver, just make it play ultra-defensively until the time is up. It really shouldn't be too hard to make it avoid any captures at all (or at least only even exchanges) for a few hundred moves. Also, some of the exchanges that we normally see as good or equal (like M for HR) are probably theoretically unsound if there's a game time limit. Here too, the player who's lost fewer pieces should be able to draw out the game to the max time limit while avoiding more material loss.
Not only does this rule completely change which strategies are valid and which ones aren't (compared to having no max time limit), the only theoretically valid strategy for silver is to play the most boring way imaginable and always use as much time on each move as you possibly can, even if it's completely obvious what the right move is.
To me this just seems like a bad rule. Much better would be to divide the max game time between the two players, and if any player uses more than that time, he loses. Unfortunately, it would be really hard to change the rules of the game now, because of all the bots that already exists and would never be patched to account for the rule change. Also, the clients would have to be updated to show how much total time each player has left.
However, did we ever need this rule yet? I mean for serious event games. I could see how it could be useful for a casual blitz game, but wouldn't it be better if we just set the max game time to unlimited for all event/postal games?
|
|
 IP Logged |
|
|
|
dree12
Forum Senior Member
Arimaa player #4082
Gender:
Posts: 27
|
|
Re: Max game time rules
« Reply #1 on: Sep 5th, 2011, 7:36pm » |
Quote Modify
|
on Sep 5th, 2011, 6:25pm, Migi wrote:
However, did we ever need this rule yet? I mean for serious event games. I could see how it could be useful for a casual blitz game, but wouldn't it be better if we just set the max game time to unlimited for all event/postal games? |
|
I believe there were BvB games that required this rule, due to them stalling with no progress.
|
|
IP Logged |
|
|
|
novacat
Forum Guru
Arimaa player #751
Gender:
Posts: 119
|
|
Re: Max game time rules
« Reply #2 on: Sep 6th, 2011, 7:01am » |
Quote Modify
|
The BvB games that had to use this rule were between two bots that were extremely poor players. This has been discussed in a previous thread, but I can't find it at the moment.
The main idea is that we have yet to see a game where such a long defense is possible against an intelligent attacker when both players are trying to win. If the attacker is successful, they win. If they are unsuccessful, it is because the defender has gained a significant advantage (usually by making attacks of their own) and can win without having to stall.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Max game time rules
« Reply #3 on: Sep 6th, 2011, 7:55am » |
Quote Modify
|
on Sep 5th, 2011, 6:25pm, Migi wrote:| It seems to me like this can easily be exploited: if your bot is playing silver, just make it play ultra-defensively until the time is up. It really shouldn't be too hard to make it avoid any captures at all (or at least only even exchanges) for a few hundred moves. |
|
Who is your bot going to exploit in this way? Not a human opponent. Experience has shown that neither player can hunker down defensively and keep everything safe. You must concede at least a rabbit pull to your opponent, if not a decisive swarm. So if you plan on having your bot win any games in this fashion, it will only be against clueless bot opponents.
Back in 2004 when Bomb was the strongest bot, it also was the most defensive-minded. A couple of other bot developers did attempt to exploit this by running out the clock without making progress. Back then the winner of an out-of-time game was decided by the arimaascore formula, but in principle the problem was the same. Fotland responded by adding a new mode to Bomb; if Bomb feels that no progress has been made after 50 moves, it turns up the aggression. This prevented any other bots from exploiting Bomb in this way.
Today's bots have become much more aggressive than Bomb was. I would be quite curious to know whether you could exploit marwin or sharp in the way that you assert is possible. You might be surprised to discover it isn't so easy. But even if you could hurt modern bots in this way, I am entirely skeptical that it work against humans who would catch on to what is happening.
You missed the phase of Arimaa's history when human players were trying to be ultra-defensive. We've been there, done that, and it didn't work.
Quote:| Not only does this rule completely change which strategies are valid and which ones aren't (compared to having no max time limit) |
|
No, it doesn't change which strategies are valid, at least not according to our current understanding of correct strategy. If it were possible to play defensively and keep everything safe against an aware opponent, that would not be a problem with the time control rule, it would be a problem with Arimaa.
Quote:| To me this just seems like a bad rule. |
|
I understand your theoretical concerns that MIGHT make this a bad rule, but for it to actually be a bad rule, certain things must be true about Arimaa that haven't proven to be true. No human vs. human games have ever gotten to a position where neither side could make progress.
Yes, there have been cases where the side with the advantage could have stalemated the position and run out the clock instead of winning on the board, but this is bad manners, not a rule problem.
Yes, there have been very rare cases where a bot had an advantage on a human player, but the bot was too stupid to exploit its advantage, and the human player's position was too weak for the human to make progress. This is rare but unfortunate. The current rule would assign the win to the bot in these few cases, after the bot runs out the clock, which is bad manners by the bot, but a reasonable outcome. Fortunately the human can resign rather than playing out the lost position pointlessly.
Yes, there have been cases where both bots were too dumb to make progress in a position that was not stalemated. I hardly care which side gets the win in such games, although I approve of cutting games off in some way rather than letting the bots waste time trying to avoid repetition until the sun burns out. If these games had any similarity to optimal play, I would be worried, but they don't.
If you really want to make a case against this rule you don't like, you need to prove that it matters.
Quote:| However, did we ever need this rule yet? I mean for serious event games. I could see how it could be useful for a casual blitz game, but wouldn't it be better if we just set the max game time to unlimited for all event/postal games? |
|
No, the rule has never affected a serious game. The first time it does matter, it will spark a huge discussion. When we have a game that is truly stalemated, we will be frantic to figure out what went wrong and how to fix it. Perhaps we will decide that the best course of action is to allow an agreed draw between the players, or to allow one player to claim a draw in a certain situation. Perhaps we will decide that a minor rule tweak will allow Arimaa to remain drawless, although we won't know what that rule tweak should be until we see what caused the draw. Perhaps we will see that the time cutoff in force was too short, and needed to be extended for the game to reach a natural conclusion. In all likelihood, when we get such a game, we will conclude that current rule is a bad rule.
However, until the rule fails even once in a real game, it is impossible to know why the rule is bad and how we should fix it. For example, if your earlier assertion is actually correct (i.e. it should be no problem for Silver to keep all pieces safe from capture), then it would be a terrible solution to have unlimited game time. Imagine a World Championship match that has dragged on for twelve hours with one player doing nothing but defense. It would be better to have the game halted and decided by some means, any means, than to have it continue until one player loses by exhaustion. At least after the game is arbitrarily halted, we will be able to have a discussion about what when wrong and how to fix it.
I concede that there is a potential problem with Arimaa, but my view is that we shouldn't try to fix it until it becomes a real problem. We see that something MIGHT break, but until it actually breaks, it isn't obvious what cure will be necessary. If we try to fix it before it breaks, we will very likely fix it the wrong way.
|
|
IP Logged |
|
|
|
Swynndla
Forum Guru
Arimaa player #1821
Posts: 235
|
|
Re: Max game time rules
« Reply #4 on: Sep 6th, 2011, 6:36pm » |
Quote Modify
|
Nice post Fritz.
Yes, it seems that as arimaa is being more & more understood, the correct play isn't to passively defend a threat, but instead to counter attack. This may in fact turn out to be optimal practical play. If I loosely define "optimal practical play" as play that gives you the highest chance of winning "over the board", then counter attacks become even more important. Example is, if your opponent is organizing a rabbit pull, it may be best (depending on the position) for you to organize your own rabbit pull on the opposite wing (otherwise, trying to stop the rabbit pull may lead to your elephant being tied down, etc, and you're at a disadvantage). This leads to attacks, counter attacks, and piece trading ... all of which increases the sharpness of the game, and shortens the number of moves of the game. This is related to the other post "risk reward? Where are we strategy-wise?"
|
|
IP Logged |
|
|
|
Migi
Forum Senior Member
Arimaa player #4643
Gender:
Posts: 26
|
|
Re: Max game time rules
« Reply #5 on: Sep 6th, 2011, 8:27pm » |
Quote Modify
|
Thanks everyone for the replies.
I can see now why an ultra-defensive strategy would not be as easy as I first imagined. I envisioned it very simply as constantly retreating your units to your home ranks, but I can see now that rabbit-pulling will prevent that. Since I'm still a real beginner at Arimaa (as you rightly say, I missed most of the history of Arimaa), I was thinking more in terms of chess than in terms of Arimaa. And in chess, it's relatively easy to draw against even the top rated engines just by locking the pawn chains and avoid any exchanges at all for 50 moves so you can claim a draw according to the "50 move rule". In chess, a rule like this would have a major impact on the game. But Arimaa isn't chess.
So you are probably right that the rule isn't very exploitable in practice, and that it does more good than harm at the moment. I am still convinced that theoretically, silver is the winning player if the max game time is under 1000 moves (while without game time limits, gold might be the winning side). But yeah, in practice, theory doesn't matter very much 
So all in all I agree: it's best not to fix things that aren't broken. Even though theoretically there is most likely a way to exploit this rule, we should wait and see if someone ever comes up with a practical way to exploit it, and only then then should we change it.
|
|
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print