Author |
Topic: Postal Tournament (Read 6763 times) |
|
99of9
Forum Guru
Gnobby's creator (player #314)
Gender:
Posts: 1413
|
|
Re: Postal Tournament
« Reply #45 on: Jan 26th, 2005, 10:24am » |
Quote Modify
|
I think you underestimate yourself Ron. Arimaazilla searches 8, and you routinely beat it. When you rushed across on move 8w of our game to save your horse, you must've been seeing the horse getting killed at 14 ply, so instead you were willing to sacrifice your cat at 16 ply. So I'd say that in some lines you look to a depth of at least 16.
|
|
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: Postal Tournament
« Reply #46 on: Jan 26th, 2005, 1:45pm » |
Quote Modify
|
OK, 8 then. And that's my final offer.
|
|
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: Postal Tournament
« Reply #47 on: Jan 26th, 2005, 4:18pm » |
Quote Modify
|
On reflection, this brings up an interesting point. Most of the comments about bot strength have been about search depth. Aarima was designed, successfully I think, to be a difficult search problem due to the large number of nodes. My interest comes from the personal experience of not really seeing the game as a tree of steps. When I see a brilliant chess move, or an aarima move (always made by somebody else), I see a sequence of tactical or strategic ideas. Examples: attack multiple areas, block access to a vulnerable area, pin the piece protecting the square needed en route to successful force projection around the king. This kind of human thinking doesn't always rely on plies. I hope someday to create a bot that emphasizes this approach. It may look many plies in parts of the tree, but its evaluation should use these higher level concepts. I realize this may just be a shift of the cognitive load more to the static evaluation function. Perhaps I'm just proposing more detail there. However, my experience makes me expect that there is something more. Unfortunately for aarima, I got involved with a new business this year. Group W is doing well! No time for bot_aardvark development anytime soon. Sorry I can't share this now. As for my own play, I'm pretty bad at tactics. I don't see many possibilities before they're played. I'm better at seeing them AFTERWARDS. That element of surprise is rewarding. So I try to use the higher level concepts. Like in the game with 99of9, I saw losing a horse versus losing a cat and possibly losing a horse and dog for his camel. I didn't think farther. It was all based on proximity and control. For me it was about a 5-ply decision. My hope is that a bot that thinks this way would be the new type of AI Omar had in mind when he started the challenge. Snape thinks this is all rubbish.
|
|
IP Logged |
|
|
|
99of9
Forum Guru
Gnobby's creator (player #314)
Gender:
Posts: 1413
|
|
Re: Postal Tournament
« Reply #48 on: Jan 26th, 2005, 5:21pm » |
Quote Modify
|
on Jan 25th, 2005, 9:54am, 99of9 wrote: Since I had the time to do it, I counted how deep I analysed one line of my last move against Paul. It got to a depth of 7 moves each (56 steps). But that was only one very slender variation... so chances are we will deviate within one or two moves. By the way Paul, if we follow all 14 moves... we end up equal but having exchanged a few pieces . I'll let you know when we deviate, and publish my analysis. |
| Ok here goes - I'm not sure anyone's interested in this analysis, but I said I'd write it down - so I will. We did indeed deviate after 12 steps. 6b Md3n ee3w Hb4n mb3n 7w Ef4w Md4n Ee4w Ed4w 7b ed3n Md5n ed4n hb7s (played up to here) 8w mb4s Ec4w Hb5e Ce1n 8b Hc5n ed5w ra7e dd8s 9w mb3e (x) Eb4s Eb3e Ec3e 9b ec5e ed5s Md6s (x) ef6e 10w Ed3w Ec3n Ec4n Md5e 10b ed4n Me5n ed5e dd7s 11w Ec5e dd6w Ed5n Me6n 11b ee5n cc7e dc6n rc8e 12w Ed6s Ed5e Ee5e Ef5e 12b ee6e Me7s ef6s Me6e (x) 13w hg4w Eg5s hf4s (x) Eg4w Paul deviated onto a path that I had considered, but not to the same depth. You can probably see why I had to go to this depth for this line, because it is sharp the entire time, and all the biggest pieces are at risk. At the same time there are often only one or two decent choices at each move, so it's possible to go to such a depth.
|
« Last Edit: Jan 26th, 2005, 5:22pm by 99of9 » |
IP Logged |
|
|
|
fotland
Forum Guru
Arimaa player #211
Gender:
Posts: 216
|
|
Re: Postal Tournament
« Reply #49 on: Jan 27th, 2005, 12:59am » |
Quote Modify
|
Some questions about bomb and the postal tournament. Right now I have it running in a script that tries each game in turn, and thinks two hours for each move. I haven't touched the computer its running on since Monday. What will happen if the computer crashes or there is a power failure or I abort the script? It will be processing a "bot move" command at the time. Will it be able to redo that move later? I assume that the game is not affected, and the clock keeps running, and the bot can play the same move later. I just realized that I can have it play much slower, since opponent time doesn't against it. It could probably meet the time control with 3 hours per move, since the average opponent takes a half day to move. I could do a more sophisticated script that plays first in the game with the least time left, rather than playing them in order. I don't want to change the time limit, but if I did, would the rules allow it? It seems against the spirit of the rules at least.
|
|
IP Logged |
|
|
|
fotland
Forum Guru
Arimaa player #211
Gender:
Posts: 216
|
|
Re: Postal Tournament
« Reply #50 on: Jan 27th, 2005, 1:08am » |
Quote Modify
|
on Jan 26th, 2005, 4:18pm, RonWeasley wrote: When I see a brilliant chess move, or an aarima move (always made by somebody else), I see a sequence of tactical or strategic ideas. Examples: attack multiple areas, block access to a vulnerable area, pin the piece protecting the square needed en route to successful force projection around the king. This kind of human thinking doesn't always rely on plies. I hope someday to create a bot that emphasizes this approach. It may look many plies in parts of the tree, but its evaluation should use these higher level concepts. |
| This is the way bomb works. The evaluation function sees the concepts you mention (hostages, frames, advanced rabbits with protection, elephant mobility, trap control, etc.) It doesn't search to a fixed depth. Interesting sequences are searched more deeply. In the postal games it's only looking 8 steps on some sequences, looking 14 steps typically, and looking 25 steps in the more tactical lines. As I improve bomb, I make the strategic parts of the evaluation more accurate, and the search more unbalanced. David
|
|
IP Logged |
|
|
|
Paul
Forum Junior Member
Arimaa player #888
Gender:
Posts: 8
|
|
Re: Postal Tournament
« Reply #51 on: Jan 27th, 2005, 4:35am » |
Quote Modify
|
I've considered during a longtime playing mb4s Ec4w Hb5e Ce1n as 8th move. It was probably better than taking the camel directly. The interessing point here is that i've chosen my actual move (Ec4n mb4e mc4s Ec5s) because I'm playing Toby : I would have certainly choosen the other way against a weaker opponent. Now I've a somewhat inferior game (Toby has the initiative), but also a more stable and flat game : with the camel exchange so soon in the game, a lot of tactical possibilities are gone. I've played exactly the other way against bomb : the horse -elephant attack is probably unsound so soon in the game, but I know Bomb puts a too high value in horse hostage in the begining of the game. All of this to emphazise the psychological dimension of arimaa in h-h games. Taking a look at recently played H-H games, I was surprised by how often a player begins to play inferior moves once he got a material advantage.
|
|
IP Logged |
|
|
|
99of9
Forum Guru
Gnobby's creator (player #314)
Gender:
Posts: 1413
|
|
Re: Postal Tournament
« Reply #52 on: Jan 27th, 2005, 5:47am » |
Quote Modify
|
on Jan 27th, 2005, 12:59am, fotland wrote: I just realized that I can have it play much slower, since opponent time doesn't against it. It could probably meet the time control with 3 hours per move, since the average opponent takes a half day to move. I could do a more sophisticated script that plays first in the game with the least time left, rather than playing them in order. |
| You never know. We humans might conspire to all play quickly and run bomb out of time. You would really need an adaptive time control. Regarding rules of the tournament - I have no idea, but I don't think changing the time control is really a change of algorithm.
|
|
IP Logged |
|
|
|
99of9
Forum Guru
Gnobby's creator (player #314)
Gender:
Posts: 1413
|
|
Re: Postal Tournament
« Reply #53 on: Jan 27th, 2005, 6:13am » |
Quote Modify
|
I was just saying to naveed that the level of aggression in arimaa has dramatically increased in the last few months. I've already been under intense fire in nearly all of my postal games. I've split them into two categories - EH attacks, and elephants going around the back of my traps to pop out my horses. EH attacks mv 2, robinson mv 3, belbo mv 4, ronweasley mv 5, adanac around back mv 3, paul mv 3, fritz mv 6, naveed Against all these onslaughts, I have only managed to play two attacks, an EH against Ron on mv 7, and an around back against naveed on mv 8. Both were counterattacks. I think this tournament may prove that arimaa theory has passed me by.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: Postal Tournament
« Reply #54 on: Jan 27th, 2005, 7:18pm » |
Quote Modify
|
on Jan 27th, 2005, 6:13am, 99of9 wrote:I was just saying to naveed that the level of aggression in arimaa has dramatically increased in the last few months. [...] I think this tournament may prove that arimaa theory has passed me by. |
| Far from theory passing you by, you are still writing the book. Opening with the camel back and center is brilliant, and I expect it has a good chance of becoming as standard as your rabbit structure. I'm only glad you uncorked it in time for me to use it in a couple of my late-starting games. Do you remember how I wasn't going to tell you my potential realization with elephant play? It was nothing more than being aware of opportunities to go around behind a trap in the opening. Big secret, eh? Given that two others played it against you too, it must be an old and/or obvious idea. Or maybe it's just the general mentality that has changed. For a while folks seemed to think that attacking was more dangerous to the attacker than the defender, but no more. IMHO, the current spate of aggression is good for Arimaa at all levels. It should make the games more interesting to play, more interesting to watch, and shorter. Unfortunately, I think only one or the other of the types of attack you mention can be correct. If the elephant/horse attack is sound, then going behind a trap to get a horse hostage is useless. It's going to a lot of trouble to get a situation (i.e. using your elephant to hold an enemy horse hostage near a home trap) that the opponent is essentially willing to give you for free with the elephant horse attack. The results of this tournament will probably dramatically affect opening theory in the future. For example, if using the camel against the exposed horse of an elephant/horse attack is effective, it may make folks abandon the attack altogether. Similarly if the around-the-back attack seems to pay dividends, then folks will have to open more conservatively with silver to prevent it from happening. Whoever suggested the postal tourney (MrBrain?), my hat is off to you. It has already been a smashing success, and we are just underway. I love being able to log in at any time and have an interesting position to analyze. I love the chance to test new ideas concretely, as opposed to just hypothesizing about them. I love being able to take a day off without penalty. Ten games seems a perfect maximum; more could start to get burdensome, but to have ten means there is always something to do. And not only is it great fun to be playing in the tournament, it is good for the game. I'll bet that these 80 games will collectively deepen our understanding of Arimaa more than the previous 1000 played on the server. If it turns out (as it now appears possible) that this tournament wraps up in under three months, we should definitely get another one going before the next World Championship cycle comes around.
|
« Last Edit: Jan 30th, 2005, 10:32am by Fritzlein » |
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: Postal Tournament
« Reply #55 on: Jan 28th, 2005, 5:49pm » |
Quote Modify
|
on Jan 27th, 2005, 12:59am, fotland wrote:Some questions about bomb and the postal tournament. Right now I have it running in a script that tries each game in turn, and thinks two hours for each move. I haven't touched the computer its running on since Monday. What will happen if the computer crashes or there is a power failure or I abort the script? It will be processing a "bot move" command at the time. Will it be able to redo that move later? I assume that the game is not affected, and the clock keeps running, and the bot can play the same move later. I just realized that I can have it play much slower, since opponent time doesn't against it. It could probably meet the time control with 3 hours per move, since the average opponent takes a half day to move. I could do a more sophisticated script that plays first in the game with the least time left, rather than playing them in order. I don't want to change the time limit, but if I did, would the rules allow it? It seems against the spirit of the rules at least. |
| There shouldn't be any problem if the computer crashes while bomb is thinking about it's move. It can run again later to make the move. The state of the game will not change on the game server until bomb sumbits the move. I don't see any problem with allowing bomb to run for longer on each move. You still are limited by the time control of the game and using one computer to handle all the games. If you were running 10 computers each with a copy of Bomb it probably would not be fair to the human opponents who can't make copies of themselves; one for each game
|
|
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: Postal Tournament
« Reply #56 on: Jan 28th, 2005, 5:58pm » |
Quote Modify
|
on Jan 24th, 2005, 9:07am, Fritzlein wrote:Shouldn't the clock of the Gold player be started in each game where no moves have yet been made? I can understand having a short grace period, but the starting date of the tournament was publicized in advance, and already players have 1 day to move plus 14 days reserve time. It seems logical for the clocks to start ticking now for exactly the same reasons they will run throughout the rest of the game. |
| You're right, for scheduled games the clock should automatically start for the first player when the scheduled time arrives. However this could be a problem for interactive games. So I think it should only be done for postal games. We have not played many scheduled postal games so this never really came up. For the next postal tournament I will set it up to auto start the clocks.
|
|
IP Logged |
|
|
|
fotland
Forum Guru
Arimaa player #211
Gender:
Posts: 216
|
|
Re: Postal Tournament
« Reply #57 on: Jan 29th, 2005, 2:29am » |
Quote Modify
|
I increased Bomb to three hours per move. It's happy with all of its positions except against Belbo.
|
|
IP Logged |
|
|
|
fotland
Forum Guru
Arimaa player #211
Gender:
Posts: 216
|
|
Re: Postal Tournament
« Reply #58 on: Jan 29th, 2005, 10:18am » |
Quote Modify
|
How much time are the the strong players using per move in this contest? I bet it's under 10 minutes per move. If so, Bomb has a 20 to 1 time advantage compared to the world championship. So this tournament lets us see how strong Bomb would be in a regular game if it were running on a 340 GHz CPU. If Bomb doen't win all of its games, it means I have to do a lot more than just wait for faster computers, to win the human-computer prize.
|
|
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: Postal Tournament
« Reply #59 on: Jan 29th, 2005, 11:54am » |
Quote Modify
|
on Jan 29th, 2005, 10:18am, fotland wrote:How much time are the the strong players using per move in this contest? |
| When you view the game after it is over, it shows the average move time for each player in the chat area; you have to scroll down a little to see it. It is currently calculated based on looking at only when the players sent the moves; so it assumes that the whole time between when your opponent moved and when you sent the next move was spent thinking by you. This is OK for interactive games, but I need to change that for postal games to look at when the player is present and when they are not and use that info as well in determining the actual thinking time.
|
|
IP Logged |
|
|
|
|