Author |
Topic: Endless Endgame Event? (Read 84766 times) |
|
clyring
Forum Guru
Arimaa player #6218
Gender: 
Posts: 362
|
 |
Endless Endgame Event?
« on: May 12th, 2012, 8:59am » |
 Quote  Modify
|
For an overview of this event, please see its wiki page.
Original post content:
I've read a few times that computers in Arimaa are much stronger in the endgame (and have seen enough of them to know why). I propose that this be changed- although we may never catch up to computers in the endgame, I see no reason for us not to try. I don't want to believe that the endgame is entirely intractable for humanity! >:(
As for how I propose this done, the thread title should be enough of a hint. What better way to foster the development of endgame theory than to encourage players to play from endgame positions?
The first issue to resolve is scheduling games with custom starting positions, but Omar's last post in the 'mob vs gang' thread seems to imply that this is possible. Other issues may arise, though...
Old rules: (Left here for my own reference when I write up new official rules)
-Randomized initial setups (first 2 ranks?), to add to variety and ensure that the endgame theory and skills that are likely to develop are broad and can apply in games of regular Arimaa rather than just a few specific positions.
-Use the same setup for every game within a given round.
-What material balances? I choose: "Randomly choose the number of every piece type except the elephant, discarding setups with more than 12 or fewer than 4 pieces per side. (That is: 50% chance of a camel on each side, randomly any of 0-2 horses/side, any of 0-2 dogs/side, any of 0-2 cats/side, and any of 1-8 rabbits/side)"
-Reflection vs rotation for silver setup... I would guess that since your strong wing is conversely your opponent's weak wing when using rotation, I would guess play would focus more on exploiting weaknesses than on creating them, at least relative to using reflection. However, both creating and using weaknesses are important. I will for now simply randomly choose which the setup is.
-Balance. Simply limiting gold's first move to two steps will probably work well enough. If gold violates this, automatic forfeit.
-Pairings (every other Monday at 2359 UTC). I would ideally like to see a 'WHREndgame' and use it to perform pairings... ;)
---I would, as a player, like to see a variety of opponents, not only in name, but in skill. Further, I believe that mismatches are important as learning experiences, but not so much when taken to extremes. (That is, a mismatch of 400 or more elo is ridiculous.)
---The system currently in use is this: Select an unpaired player at random, then generate a normally distributed random number with standard deviation 150 centered around their rating and select from the list of unpaired opponents (excluding the last 3* played) the one whose rating is closest to the normally distributed number as the first player's opponent. If there is no legal opponent, restart the pairing process from the beginning. If there are an odd number of players, when the second-to-last bot is paired, the remaining bot will immediately be paired against a dummy player unless this bot also did not play in the previous 3* rounds, in which case the pairing process will be restarted. After all players have been paired but before the pairings are posted, the pairs will be sorted by total current WHREND rating. *If there are no legal pairings using 3 rounds, I will use the highest possible value for the number of rounds since the most recent non-playing round of the bot eventually paired to the dummy (up to 3 rounds) and then the highest possible value for the number of rounds since the last repeat pairing (up to 3 rounds) in strict priority.
-Time control: 2m/0/100/0/6h/6m (or equivalent)
-All games should, of course, be unrated and affect neither the gameroom ratings nor the WHR ratings for regular games (WHRH, WHRP, and WHRE). Games will still affect my own WHREND ratings.
Is there interest in something like this?
Are there suggestions for improving this?
|
| « Last Edit: Feb 16th, 2015, 10:20am by clyring » |
 IP Logged |
I administer the Endless Endgame Event (EEE). Players welcome!
|
|
|
browni3141
Forum Guru
Arimaa player #7014
Gender: 
Posts: 385
|
|
Re: Endless Endgame Event?
« Reply #1 on: May 12th, 2012, 1:54pm » |
Quote Modify
|
I think at the very least players should be able to setup and play any position as an unrated game.
On the topic of engines and Arimaa endgames, I've never been impressed with their play. Perhaps I haven't played strong enough bots or let the board get empty enough for them to shine.
While it makes sense that bots should perform better, it also makes sense in a way that they might not.
|
|
IP Logged |
|
|
|
chessandgo
Forum Guru
Arimaa player #1889
Gender:
Posts: 1244
|
|
Re: Endless Endgame Event?
« Reply #2 on: May 12th, 2012, 2:08pm » |
Quote Modify
|
I'd like to be able to play endgames-only arimaa games as well.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Endless Endgame Event?
« Reply #3 on: May 12th, 2012, 2:48pm » |
Quote Modify
|
I'm interested in endgame play; I'm tying to think what would be the most plausible code modification. Omar is unlikely to prioritize this, but the client code is open, so if someone modified the client in a way that left relatively simple server-side modifications, then it might actually happen.
The simplest thing I can think of is a client that allows setting up with fewer than the full set of pieces. Maybe reserve the single click for swapping piece locations, and let a double-click zap a piece off the board. Double-clicking on an empty home square could bring back a piece that was earlier zapped. Also to prevent the feature from annoying people who don't want it, make the ability to remove pieces during setup available only when "expert mode" is turned on, and only if the game is unrated. While we're at it (if expert mode is turned on and the game is unrated) allow pieces to be set up anywhere on the board.
I don't think it is necessary to code alternate rules about the setup. Let the players decide what their alternate rules are, and enforce the rules themselves by leaving the game if they can't agree. Better to give the players great flexibility, so that when they decide some other variant rules are even better, no new coding has to happen.
I have a hunch that modifying the client setup code is the most realistic code path than a tool to create games with arbitrary setup, not least because a non-Omar person could do the bulk of it.
|
|
IP Logged |
|
|
|
clyring
Forum Guru
Arimaa player #6218
Gender:
Posts: 362
|
|
Re: Endless Endgame Event?
« Reply #4 on: May 12th, 2012, 4:30pm » |
Quote Modify
|
Fritzlein, I think you've missed too much of what I am suggesting happen by focusing instead on how I initially thought of implementing it. Please see the thread title. If all I wanted were a more convenient way for players to start with reduced material, I would say nothing and instead continue with the mutual agreement to sacrifice pieces that has worked so well so far. I am looking to create a motivator in the form of an endless (or continuous, if you prefer  ) endgame event such that more people actually play these endgame positions and the beginnings of endgame theory can develop sooner rather than later.
|
| « Last Edit: May 12th, 2012, 5:00pm by clyring » |
IP Logged |
I administer the Endless Endgame Event (EEE). Players welcome!
|
|
|
rbarreira
Forum Guru
Arimaa player #1621
Gender:
Posts: 605
|
|
Re: Endless Endgame Event?
« Reply #5 on: May 13th, 2012, 3:57am » |
Quote Modify
|
on May 12th, 2012, 1:54pm, browni3141 wrote:
While it makes sense that bots should perform better, it also makes sense in a way that they might not. |
|
When humans routinely miss goal-in-2 opportunities (both for and against them), which bots can often find in less than one second, it's hard to believe that humans have the edge in sharp endgames. Even some rare goal-in-1 moves can be hard to see optically without shuffling pieces on the board...
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Endless Endgame Event?
« Reply #6 on: May 13th, 2012, 9:47am » |
Quote Modify
|
on May 12th, 2012, 4:30pm, clyring wrote:| Fritzlein, I think you've missed too much of what I am suggesting happen by focusing instead on how I initially thought of implementing it. |
|
The tournament is a good idea. I hope it happens. I hope a lot of people participate in it.
[EDIT]
I shouldn't have responded so defensively, but clyring quoted me before I thought better of it! 
|
| « Last Edit: May 13th, 2012, 1:27pm by Fritzlein » |
IP Logged |
|
|
|
clyring
Forum Guru
Arimaa player #6218
Gender:
Posts: 362
|
|
Re: Endless Endgame Event?
« Reply #7 on: May 13th, 2012, 11:42am » |
Quote Modify
|
on May 13th, 2012, 9:47am, Fritzlein wrote:
Before posting, I did read your suggestion in its entirety. Do you think there could have been a good reason that I focused on only one part in my response? Or must my limited response have been due to my failure to comprehend the full glory of your proposal?
|
|
As you made no mention in your earlier post of the event itself, it was unclear from what context you were writing. I do not mean to insult you, only to ensure that you were considering it from the context of an entire event and not just one game. The second-to-last paragraph of your previous post in particular confuses me when looking at it from the perspective of a tournament- I could hardly call it a competition if, in each matchup, the players use completely different rules.
on May 13th, 2012, 9:47am, Fritzlein wrote:
The harder something is to implement in code, the less likely it is to happen, regardless of how fun it would be. I have participated in too many "wouldn't it be neato if..." discussions that were then dropped with nothing happening, because the expectation was that Omar would do all the work to implement the entire, exact feature set that had been proposed. Yes, it matters how nifty your idea is, but also it matters who is going to make it happen and how. My thoughts are guided to separate the part that you (or some other enthusiastic volunteer) can make happen without Omar from the part that requires Omar to write new code himself.
|
|
I don't want to participate in a "wouldn't it be neato if" discussion, but I suppose I deserve this one for listing it as "the first issue to resolve" when in fact it interferes only with the convenience of the players- as things are, the players could simply suicide pieces en masse to create a position I give them before the game and then begin play, but that would be a great hassle for all involved and something that I would like to avoid if possible.
on May 13th, 2012, 9:47am, Fritzlein wrote:
Furthermore, there have been other variant proposals before yours and there will be others after. I think there is merit in a client modification that subsumes many variant setups all at once, allowing players to experiment not just with (say) symmetrical endgame setups, but also asymmetrical ones, and arbitrary positions they feel like playing out. Is it wrong of me to muse that, if a code change has to be made, we might make it more general for no extra work?
|
|
I never wrote that any code changes must be specific to this event- I would prefer that they be adaptable as well. Adding in specific code to tell the client to generate randomly a setup according to an arbitrarily chosen selection procedure that is only used for one event is more work and gets less done than just allowing a setup editor either in the client or in the scheduling tool. I would not knowingly ask anyone to do such a thing.
on May 13th, 2012, 9:47am, Fritzlein wrote:
If the only obstacle to your event were pairings, you could do them by hand for a small number of players, as I initially did for the Continuous Tournament that I ran. I didn't ask Omar to implement my pet pairing scheme. That enabled the Continuous Tournament to get off the ground with less waiting on Omar. Before long, woh stepped in with a program to automate the maintenance of standings and pairings for me according my peculiar scheme (never used before or since, I might add). My last post didn't focus on your pairing proposal, because that isn't the biggest technical obstacle to your event happening.
|
|
I agree that the pairings are not the biggest obstacle here.
on May 13th, 2012, 9:47am, Fritzlein wrote:
I thought my response was constructive, in that I was brainstorming about the best way to get over the biggest hurdle, but if it was unwelcome then I will bow out. |
|
Your suggested client modification would also speed things along and is a good alternative that I would just as much like to see.
EDIT:
on May 13th, 2012, 9:47am, Fritzlein wrote:
The tournament is a good idea. I hope it happens. I hope a lot of people participate in it. |
|
Thank you.
|
| « Last Edit: May 13th, 2012, 12:07pm by clyring » |
IP Logged |
I administer the Endless Endgame Event (EEE). Players welcome!
|
|
|
browni3141
Forum Guru
Arimaa player #7014
Gender:
Posts: 385
|
|
Re: Endless Endgame Event?
« Reply #8 on: May 13th, 2012, 7:00pm » |
Quote Modify
|
on May 13th, 2012, 3:57am, rbarreira wrote:
When humans routinely miss goal-in-2 opportunities (both for and against them), which bots can often find in less than one second, it's hard to believe that humans have the edge in sharp endgames. Even some rare goal-in-1 moves can be hard to see optically without shuffling pieces on the board... |
|
I was thinking about the point where the board is fairly empty, but goal threats and tactics don't completely dominate. There are still strategic elements. I was thinking that bots might not be so strong at this point because their evaluation is tuned for fuller boards, and different things become important as pieces come off the board. I don't know how accurate this is and it certainly will depend on the bot. How does briareus handle this?
I don't think goals in two are as big a deal for humans as you make them out to be. Most goals in two are fairly easy to see. At least in my opinion it only starts to get difficult beyond that. I hope to reach the point where it is almost unheard of to miss a goal in two
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Endless Endgame Event?
« Reply #9 on: May 14th, 2012, 12:19am » |
Quote Modify
|
on May 13th, 2012, 7:00pm, browni3141 wrote:| I hope to reach the point where it is almost unheard of to miss a goal in two |
|
I just hope that I don't lose any more World Championship games by missing goal in two. 
|
|
IP Logged |
|
|
|
rbarreira
Forum Guru
Arimaa player #1621
Gender:
Posts: 605
|
|
Re: Endless Endgame Event?
« Reply #10 on: May 14th, 2012, 3:13am » |
Quote Modify
|
on May 13th, 2012, 7:00pm, browni3141 wrote:
I was thinking about the point where the board is fairly empty, but goal threats and tactics don't completely dominate. There are still strategic elements. I was thinking that bots might not be so strong at this point because their evaluation is tuned for fuller boards, and different things become important as pieces come off the board. I don't know how accurate this is and it certainly will depend on the bot. How does briareus handle this?
I don't think goals in two are as big a deal for humans as you make them out to be. Most goals in two are fairly easy to see. At least in my opinion it only starts to get difficult beyond that. I hope to reach the point where it is almost unheard of to miss a goal in two |
|
Briareus is not one of the best bots when it comes to "advancing rabbits" strategy, I think Marwin is better at that. It's only when the position gets sharp that I think it starts playing better.
It was very interesting when jdb had bot_clueless playing NoMHH (i.e. no camels/horses on either side) games online, and I've been thinking about doing the same with briareus. Watch out for that (maybe later today now) if you're interested.
As for my comments about goals-in-2, they're based on the fact that I recall several instances of players missing them, sometimes even in postal games.
|
| « Last Edit: May 14th, 2012, 3:59am by rbarreira » |
IP Logged |
|
|
|
hyperpape
Forum Guru
Arimaa player #7113
Gender:
Posts: 80
|
|
Re: Endless Endgame Event?
« Reply #11 on: May 14th, 2012, 7:50am » |
Quote Modify
|
Quote:| -Randomized initial setups (first 2 ranks?), to add to variety and ensure that the endgame theory and skills that are likely to develop are broad and can apply in games of regular arimaa rather than just a few specific positions. |
|
Why not just grab a bunch of interesting positions from actual games? Take a game that had a complicated endgame and back up 5, 10 or 15 moves to reach a point where you're not just looking for goal in two threats, and play from there.
That way, you ensure that the positions are ones that plausibly arise in real games.
There is a worry that it may be hard to find even games, and therefore it might be hard to build a rating system. In my opinion, the rating system is the least interesting part of this exercise, but I suspect others would differ.
|
|
IP Logged |
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: Endless Endgame Event?
« Reply #12 on: May 14th, 2012, 9:27am » |
Quote Modify
|
Another idea might be to set up a mob vs bot postal game with reduced material. That way everyone can learn collectively about a bot's strengths and weaknesses in the endgame at a slow pace.
I think this idea will give the most "bang for the buck" without a great deal of setup and customization that a more detailed proposal such as yours would require.
|
|
IP Logged |
|
|
|
Hippo
Forum Guru
Arimaa player #4450
Gender:
Posts: 883
|
|
Re: Endless Endgame Event?
« Reply #13 on: May 14th, 2012, 9:37am » |
Quote Modify
|
on May 14th, 2012, 3:13am, rbarreira wrote:
As for my comments about goals-in-2, they're based on the fact that I recall several instances of players missing them, sometimes even in postal games.
|
|
Yep, I agree. I often see a goal in 2 which actually can be defended ... . At least in the "almost mentioned" postal with Nombril.
|
|
IP Logged |
|
|
|
browni3141
Forum Guru
Arimaa player #7014
Gender:
Posts: 385
|
|
Re: Endless Endgame Event?
« Reply #14 on: May 14th, 2012, 6:42pm » |
Quote Modify
|
no_MHH briareus didn't play very well in my first game with it, but I must say I'm impressed that it apparently sees forced goal before I even know it is close (31g).
http://arimaa.com/arimaa/gameroom/comments.cgi?gid=230109
|
| « Last Edit: May 14th, 2012, 6:44pm by browni3141 » |
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print