Author |
Topic: 2006 Postal Tournament Format (Read 3042 times) |
|
Fritzlein
Forum Guru
Arimaa player #706
Gender: 
Posts: 5928
|
 |
2006 Postal Tournament Format
« on: Nov 18th, 2005, 1:26pm » |
 Quote  Modify
|
If Omar is going to be incommunicado for the first half of January, then he will probably want to finalize the format for the Postal Championship and open registration for it before World Championship has ended. Last year's postal registration deadline was January 23rd, so I thought I would kick off the discussion now.
On the other hand, there might not be very much that needs to be discussed, because last year's format worked so well. The $20 registration and the scoring system to return prize money (designed to insure commitment) seemed to work just fine. I note that KamikazeKing, the 15th seed, got back $19.38 for his spirited play, and while JDB scored $30.57 as the 13th seed.
The pairing ring scheme seemed excellent, as it allowed for a flexible and large number of players. To have overlapping pairings meant that we didn't have to limit registration, we didn't have to split the tournament, and yet we could keep the number of games per player down to a managable ten. I especially liked how everyone (except the top and bottom few seeds) had a shot at five higher-rated players as well as five lower-rated players, so that everyone had a variety of difficulties. I liked that all the extreme mismatches were ruled out.
If there was one serious problem, it was that the time commitment seemed a touch too high for many people. We could fix that by reducing the number of games per person, but that would also reduce the social mixing which is such a nice feature. Instead I would suggest extending the time control from 1d/14d/100/14d/300d to 2d/14d/100/28d/300d. Two days per move would mean that you could pace yourself by moving in only two or three games per day. The 28 day maximum reserve would allow folks to build up to a three-week vacation some time in the year if they moved reasonably quickly at other times. The 14 day initial reserve (rather than 28) would discourage people from signing up without intending to start right away, so that we wouldn't have to wait a month to know whether someone would be in or out.
Otherwise the only glitch I can think of was that games didn't automatically start, i.e. the Gold clock didn't run until Gold made a move, which allowed Gold not to show up for an unreasonably long time. Fortunately nobody abused this loophole, but it should be closed nonetheless.
Mostly, though, if it ain't broke, don't fix it. Let's just do it again this year, and hope it is as fun as it was last year!
What do other folks think?
|
| « Last Edit: Nov 18th, 2005, 1:27pm by Fritzlein » |
 IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: 2006 Postal Tournament Format
« Reply #1 on: Nov 18th, 2005, 2:11pm » |
Quote Modify
|
I agree and emphasize that lower rated players can have a good experience here since they (we?) have many games against each other. I liked having a bot among the contenders. Maybe more than one bot is possible. Is there a problem with bots playing each other in postal?
An issue is the fact that the idea of a "winner" is not enforced well in this format. For me, that's OK. I liked it just for the experience, but people counting their galleons may observe that underrated and improving players do better than overrated players. But consider that last year I won less than half my games and still made a profit. $0.09! That makes me a professional!
I had no problem playing ten games. If a longer time format solves the problems some of the players had, I'm in favor of it.
|
|
IP Logged |
|
|
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2006 Postal Tournament Format
« Reply #2 on: Nov 18th, 2005, 3:40pm » |
Quote Modify
|
We should definitely discuss the format of this year’s tournament, such as:
(a) Manageable number of games ( e.g. 5-8 ) for each player with 1d/14d/100/14d/300d format
(b) Large number of games (10-14) with 2d/14d/100/28d/300d format
(c) Round Robin format with 2d or possibly 3d time limit.
Personally, I wouldn’t be as excited about option (a) because I enjoy playing a large number of games simultaneously in a large tournament. If we have 16-20 players and we go with option (a) then I would also suggest dividing the players into 2 groups; either players choose to join a 1d or 2d per move group, or alternatively, over/under 1800 rating. The final standings simply wouldn’t be meaningful in a large single group if each player played less than half the competitors.
If we’re going to have one large group, I’d prefer option (c) with a 3 day limit such as 3d/30d/50/30d/360d. The 50% of unused time portion would encourage players to play more quickly before going on a prolonged vacation. On the other hand, the 30d bank could allow players to take a couple of weeks off if they feel overwhelmed. I’ve played Correspondence Chess tournaments in the past that have lasted 2+ years and they work just fine despite the long time period, but it’s always nice to have the option to take a break now and then!
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2006 Postal Tournament Format
« Reply #3 on: Nov 19th, 2005, 7:15am » |
Quote Modify
|
on Nov 18th, 2005, 3:40pm, Adanac wrote:| If we’re going to have one large group, I’d prefer option (c) with a 3 day limit such as 3d/30d/50/30d/360d. The 50% of unused time portion would encourage players to play more quickly before going on a prolonged vacation. |
|
I haven't thought much about a 50% time control. My first guess is that not banking all unused time encourages players to move at a regular pace. It's harder to build up the reserve than to drain it, so using exactly the allotted time every move should pay off relative to thinking more on one move and less on another.
As a tournament director, I would be inclined try to make the time per move uniform for the sake of spectators watching live games. When it comes to the postal tourney, I don't see the purpose of not banking 100% of unused time. It seems more like a penalty to people who have an irregular real-life schedule. However, I'm probably missing something here.
One thing that just occurred to me about the time control is that it makes sense to have it accelerate. We need it to be slow at the beginning when everyone has all their games going at once, but games start to end after a while, or become lopsided affairs which need less attention. When each player has fewer games, a shorter time control would not only be tolerable, it would be preferable as a means to getting the tournament over with. With this in mind, we could consider something like two days per move for the first three months of the tournament only, with any games still in progress speeding up to one day per move after that.
Quote:We should definitely discuss the format of this year’s tournament, such as:
(a) Manageable number of games ( e.g. 5-8 ) for each player with 1d/14d/100/14d/300d format
(b) Large number of games (10-14) with 2d/14d/100/28d/300d format
(c) Round Robin format with 2d or possibly 3d time limit. |
|
I know 99of9 was pushing for the round robin format last year, so that the rankings would be more realistic. He felt JDB's eight wins weren't on a par with Belbo's eights wins, since Belbo had tougher opposition. This observation is true as far as it goes, but doesn't persuade me to want to change the format.
I don't think we need to focus so much on determining a winner as on having lots of good games. Looking at last year's bracket, there were 80 games in the actual pairing, whereas a full round robin would have been 120 games. However, 18 of the 40 games that weren't played would have been mismatches of 400 to 750 points. Why should we include a bunch of games like that just to show who is boss? The pairing method from last year let everyone have ten good games, and that's more important in my estimation.
I would rather split into two round-robins than have one gigantic round-robin, but then there is the complication of not knowing how many people will sign up or where the cutoff will be. Also I can imagine that someone rated 1790 might rather "play up" than have an easy time in the under-1800 bracket. The advantage of last year's pairing is that everyone got a mix of tougher and easier games, omitting the far extremes.
|
|
IP Logged |
|
|
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2006 Postal Tournament Format
« Reply #4 on: Nov 19th, 2005, 11:32am » |
Quote Modify
|
on Nov 19th, 2005, 7:15am, Fritzlein wrote:
As a tournament director, I would be inclined try to make the time per move uniform for the sake of spectators watching live games. When it comes to the postal tourney, I don't see the purpose of not banking 100% of unused time. It seems more like a penalty to people who have an irregular real-life schedule. However, I'm probably missing something here.
|
|
Some tournaments can take can much longer than expected because one or two players take multiple vacations. I was thinking of ways to encourage those players to play faster but you’re right, the 50% banking would be more logical in a live game.
on Nov 19th, 2005, 7:15am, Fritzlein wrote:
One thing that just occurred to me about the time control is that it makes sense to have it accelerate. We need it to be slow at the beginning when everyone has all their games going at once, but games start to end after a while, or become lopsided affairs which need less attention. When each player has fewer games, a shorter time control would not only be tolerable, it would be preferable as a means to getting the tournament over with. With this in mind, we could consider something like two days per move for the first three months of the tournament only, with any games still in progress speeding up to one day per move after that.
|
|
Can we set up a time control like that? I didn’t know that was possible. It might be simpler to set up a time control like 1d/90d/100/90d/300d so that players can use as much time as they need in the opening, but they’re eventually forced to play one move per day?
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2006 Postal Tournament Format
« Reply #5 on: Nov 19th, 2005, 7:08pm » |
Quote Modify
|
on Nov 19th, 2005, 11:32am, Adanac wrote:| Can we set up a time control like that? I didn’t know that was possible. |
|
I guess it would require either manual intervention or extra coding by Omar. I was actually imagining him changing all the games by hand, so maybe it wasn't too nice to suggest it!
Quote:It might be simpler to set up a time control like 1d/90d/100/90d/300d so that players can use as much time as they need in the opening, but they’re eventually forced to play one move per day?
|
|
I guess maybe I want too much. I want (A) that people have enough time to play all their games so they don't feel rushed (especially at the beginning) and can take a vacation later if they move quickly at other times and (B) that the last few games don't straggle on forever, and (C) that the games all start promptly, so that someone doesn't sign up intending to start a couple of months later. But maybe (C) is the least important to me, which would make a one day increment with a large starting reserve reasonable solution to (A) and (B).
|
|
IP Logged |
|
|
|
Ryan_Cable
Forum Guru
Arimaa player #951
Gender:
Posts: 138
|
|
Re: 2006 Postal Tournament Format
« Reply #6 on: Jan 29th, 2006, 12:57am » |
Quote Modify
|
I would like to get this figured out ASAP. As long as the PC looms in the not too distant future, I don’t want to start other postal games.
I don’t have much to say about the overall tournament structure beyond a weak preference that there should be some way to declare a Postal Champion at the end. For the time control, I would like to have an increment of at least 1d12h and a reserve limit of at least 14d. With 1d per move, it is too hard to build up the reserve after not logging on for a few days, particularly when time zones line up unfavorably.
|
|
IP Logged |
|
|
|
99of9
Forum Guru
Gnobby's creator (player #314)
 
Gender:
Posts: 1413
|
|
Re: 2006 Postal Tournament Format
« Reply #7 on: Jan 29th, 2006, 2:07am » |
Quote Modify
|
I'd prefer NOT to change the time controls in the middle of the games, but apart from that I'm happy with most of these suggestions.
Fritz I like the idea of splitting into two smaller round robins. My main reason for wanting a round robin is to ensure that all the top players get to play all the other top players. I'm not gunning for mismatches! I just feel that key gaps are disappointing. You and I got great draws last year, but everyone else missed out on at least one other top player.
To get an idea of numbers, can we have an expression of interest? Has anyone decided on a date?
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2006 Postal Tournament Format
« Reply #8 on: Jan 29th, 2006, 11:47am » |
Quote Modify
|
Like Ryan, I have been avoiding starting new postal games in anticipation of the Postal Tourney. I'm ready to go any day, although I think registration should be open at least two weeks. Will the tourney revive flagging server activity? Can we possibly get a turnout as high as last year's 16 players? I hope it's Yes and Yes.
If the turnout is 16 or more, I guess splitting into two round-robins would be all right with me. Seven games each would be enough, barely  . But with fewer than 16 players, I'd prefer to keep it unified and have more games. After the thrill of ten games last year, only six this year would seem paltry. I side with Adanac in wanting a full slate of games, even if we have to slow down the time control to something like 1d/80d/100/0/300d/21d.
Another concern with splitting the tournament by rating is the huge gap in what adjacent seeds get. The #8 seed might not want purely games against tough opponents, and the #9 seed might not want purely games against weak opponents. Also, if prize money is distributed as it was last year, someone near the bubble has a stark monetary incentive to drop rating points before the tournament begins.
Of course, as RonWeasley pointed out, there was already a monetary incentive to be underrated, but it was much more gradual than getting all weaker games versus getting all stronger games. Dumping some points would instead maybe lose you your top opponent and gain you an opponent on the bottom.
Maybe sandbagging is a phantom concern anyway. There have been lots of instances of people pumping up their ratings, but the single instance of point dumping I can recall was someone intentionally losing to Arimaalon a couple of times. Mostly I think everyone will be eager to "play up". Folks will be most excited for the opportunity to play a few of their games against stronger opponents.
Last year's ring pairing had all-play all among the top six anyway, and compared to a round-robin among the top eight it was missing only 6vs7, 5vs8, and 7vs8. I'd rather lose these three games while giving #10 a shot at #2 than segregate the whole tournament.
That said, if we should be so lucky as to get more than 16 players, the bigger the tournament gets, the more I buy into the argument for split round robins.
When we last discussed time controls, I didn't consider the max-time-per-move parameter. That could be used to solve all of our problems with a ten-game schedule. The time control of 1d/80d/100/0/300d/21d gives everyone a huge starting cushion of 80 days for when there are too many games to play at 1d per move. If you average 3d per move, then you will use up your reserve in 40 moves, by which time half of your games will be over, some of the other half may be clearer, and you should be able to keep up at 1d per move in your remaining games.
The key parameter for me is putting a 21d limit on any given move. The huge reserve could lead to people quitting (or not even showing up at the start) and causing their opponents to wait eleven weeks for the clock to drain down. That's unfair to the opponent, and I wouldn't want to permit someone back in after 79 days vacation even if they showed up ready to play then.
The 300d total game length only gaurantees 70 moves given the 80d starting reserve, but if at least one player is moving quickly enough to max out the reserve, then that guarantees 110 moves, so there shouldn't be much threat of one player dragging it out to win on score. With 300d total game time and a February start, straggler games might already overlap with next year's WC, so why not increase the game time to 365d? At least the games will be over in time for the 2007 Postal Tourney! 
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2006 Postal Tournament Format
« Reply #9 on: Jan 29th, 2006, 12:24pm » |
Quote Modify
|
Last year there was some serious interest from developers in having their bots participate in the Postal Championship, as well as much discussion of what conditions should be imposed on bots to keep things fair. In the end only Bomb participated, and Fotland was quite disappointed with the results.
This year I haven't heard interest from any developer in entering a bot, but I have heard several comments from humans that it would be cool to have at least one bot entered. Given that players may be eager and developers lukewarm, it makes sense to me to either
(A) Have Omar sponsor one of the CC bots to play. Of course, Bomb2006 can't be exposed until after the Challenge is over, and anyway it didn't change much from Bomb2005 which played last year, so the logical candidates would be Clueless2006 and Aamira2006, or
(B) Entice developers to participate by making the postal tournament an arena in which experimental changes can be tested. Last year we required that programs not be manually altered after the start of the tournament, because allowing alterations would allow the developer to insert one set of parameters for one game, and a different set of parameters for a different game, thus "coaching" the bot into attacking or defending or doing whatever was necessary in a particular game.
It seems to me that this obstacle could be overcome by changing the rule to "Programs may be manually altered subject to the following restrictions:
* Before a version of a program can be manually altered, that version must make at least twenty-five moves within the tournament (to prevent move-by-move coaching).
* A program may not alter its play based on the identity of the opponent (to prevent game-by-game coaching).
* Programs must respond to their opponents on a first-moved, first-answered basis (to prevent a version from avoiding positions it temporarily doesn't understand).
Even within these restrictions there could be some potential for abuse, but I'd like to think we can trust all of our developers around here, just like I'd like to think we can trust human players not to call their friends for advice between moves.
It seems that 80% of bot development happens in December, and if we could do something to keep developers actively working on their bots for more of the year, I'm all for it.
|
|
IP Logged |
|
|
|
Ryan_Cable
Forum Guru
Arimaa player #951
Gender:
Posts: 138
|
|
Re: 2006 Postal Tournament Format
« Reply #10 on: Jan 29th, 2006, 6:08pm » |
Quote Modify
|
on Jan 29th, 2006, 12:24pm, Fritzlein wrote:It seems to me that this obstacle could be overcome by changing the rule to "Programs may be manually altered subject to the following restrictions:
* Before a version of a program can be manually altered, that version must make at least twenty-five moves within the tournament (to prevent move-by-move coaching).
* A program may not alter its play based on the identity of the opponent (to prevent game-by-game coaching).
* Programs must respond to their opponents on a first-moved, first-answered basis (to prevent a version from avoiding positions it temporarily doesn't understand). |
|
I think these rules are unnecessarily strict, considering the current gap between humans and bots and since I think the PC is somewhat less serious than the WC, CC or Challenge. I suggest the following:
* After any change, the bot must be left unchanged for 1 week (168 hours), before another change may be made.
* A program may not alter its play based on the identity of the opponent.
* Programs must respond to their opponents on a first-moved, first-answered basis.
* The bot must think for either the same amount of time or the same number of ply (ignoring search extensions) for each game.
|
|
IP Logged |
|
|
|
99of9
Forum Guru
Gnobby's creator (player #314)
Gender:
Posts: 1413
|
|
Re: 2006 Postal Tournament Format
« Reply #11 on: Jan 29th, 2006, 8:17pm » |
Quote Modify
|
Fritz wrote:
Quote: * Before a version of a program can be manually altered, that version must make at least twenty-five moves within the tournament (to prevent move-by-move coaching).
* A program may not alter its play based on the identity of the opponent (to prevent game-by-game coaching).
* Programs must respond to their opponents on a first-moved, first-answered basis (to prevent a version from avoiding positions it temporarily doesn't understand). |
|
on Jan 29th, 2006, 6:08pm, Ryan_Cable wrote:
I think these rules are unnecessarily strict, considering the current gap between humans and bots and since I think the PC is somewhat less serious than the WC, CC or Challenge. I suggest the following:
* After any change, the bot must be left unchanged for 1 week (168 hours), before another change may be made.
* A program may not alter its play based on the identity of the opponent.
* Programs must respond to their opponents on a first-moved, first-answered basis.
* The bot must think for either the same amount of time or the same number of ply (ignoring search extensions) for each game. |
|
Ryan I think you just made them even stricter! You added one condition, and as I understand it, you increased another. 1 week is longer than the time taken to play 25 moves if the bot is playing around 8-10 games. Or did you mean 25 moves in each game Fritz?
I think the most effective way of encouraging bots into the tournament is:
1) Exempt them from the registration fee and from the winnings.
2) Someone provide a dedicated computer for each to play on.
Number 1 is easy, and I think should have general agreement. If the players want the bots in the tourney, the bots shouldn't have to pay their way.
Number 2 is much harder, and will put the crucial limit on the number of bots. I suggest that anyone who wants to donate a computer in such a way gets the rights to offer it to whichever bot they want added to the tourney.
Anyway, back to these conditions you want to put on the bots:
Quote:| * Before a version of a program can be manually altered, that version must make at least twenty-five moves within the tournament (to prevent move-by-move coaching). |
|
I don't mind this, but I'd instead make it three moves in every game it is in (so that updates are still possible once the number of opponents diminishes). No human can fully predict 3 moves ahead, so no move-by-move direct coaching would be possible.
Quote:| * A program may not alter its play based on the identity of the opponent (to prevent game-by-game coaching). |
|
I think this is unneccessary, and a mistake. Your first rule already prevented the direct coaching I was worried about last year. Identity-based play is perfectly legitimate, and although to my knowledge none of the bots do this already, I think it would be a very good thing if they did. Why disallow intelligent play?
Quote:| * Programs must respond to their opponents on a first-moved, first-answered basis (to prevent a version from avoiding positions it temporarily doesn't understand). |
|
This is unneccessarily strict. Bomb did not do this last year (it "cycled" to find the next player that had already played - it did not then sort them to figure out who had played first). Are you worried the program would see a hard position, put it on hold for 25 moves, thus waiting for the developers direct instructions? If so, then my variant of your 25 move rule fixes this. The bot would have to play this move before getting an update, because it has to play 3 moves in every game.
And one extra from Ryan:
Quote:| The bot must think for either the same amount of time or the same number of ply (ignoring search extensions) for each game. |
|
I think this one is also unneccessary. The algorithm which decides time control can also only be changed when the bot is changed. So the developer is not going to be able to give the instruction "That's a hard position, think more about that one."
Fixing ply is difficult, because you don't know in advance whether that will cause it to timeout, or conversely whether that will cause it to play way too quickly.
Fixing time is often wasteful of computer resources. For example Gnobot can often tell that it is not going to be able to complete the next ply in time. What's the point in forcing it to use that time? It would be better to let the bot manage its reserve intelligently.
|
| « Last Edit: Jan 29th, 2006, 8:19pm by 99of9 » |
IP Logged |
|
|
|
Ryan_Cable
Forum Guru
Arimaa player #951
Gender:
Posts: 138
|
|
Re: 2006 Postal Tournament Format
« Reply #12 on: Jan 29th, 2006, 8:36pm » |
Quote Modify
|
Oh, I had misread the 25 move rule to be per game, now it makes sense. However, the 3 move per game limit causes problems if one person is on vacation for a week. With the time requirement, I just wanted to keep the bots from moving quickly in some games to focus on others.
|
|
IP Logged |
|
|
|
Janzert
Forum Guru
Arimaa player #247
Gender:
Posts: 1016
|
|
Re: 2006 Postal Tournament Format
« Reply #13 on: Jan 29th, 2006, 11:25pm » |
Quote Modify
|
on Jan 29th, 2006, 8:36pm, Ryan_Cable wrote:| With the time requirement, I just wanted to keep the bots from moving quickly in some games to focus on others. |
|
Why? I assume humans would do this. Splitting there time depending on how critical they think the current position in the game is. I think a bot should be allowed to do the same, although they're going to be much less likely to understand when a position is "critical" or not.
On the rest of the proposed restrictions I pretty much agree with 99of9, except for the three move rule. I had the same thought as Ryan there. Maybe a combination of 3 moves per game or 25 moves total whichever comes first.
Janzert
|
|
IP Logged |
|
|
|
Ryan_Cable
Forum Guru
Arimaa player #951
Gender:
Posts: 138
|
|
Re: 2006 Postal Tournament Format
« Reply #14 on: Jan 30th, 2006, 1:14am » |
Quote Modify
|
on Jan 29th, 2006, 11:25pm, Janzert wrote:| Why? I assume humans would do this. Splitting there time depending on how critical they think the current position in the game is. I think a bot should be allowed to do the same, although they're going to be much less likely to understand when a position is "critical" or not. |
|
I was thinking of preventing the operator from choosing how long the bot thinks about each game out of fairness to the human players given the nature of a round robin. I don’t feel very strongly about this, since I think an order of magnitude change in thinking time is necessary to significantly change a bots strength.
|
|
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print