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

        Arimaa >> General Discussion >> Plan for attack?!
        
(Message started by: Gerenuk on Jan 23rd, 2008, 8:47am)
Title: Plan for attack?!
Post by Gerenuk on Jan 23rd, 2008, 8:47am
Hi,
I've recently started to play Arimaa, but as strategy and bots get more complex I seem to be lost. I've read the wikibook. Well, I'm not confident enough with all move outcomes (main losing reason), but working on it. Simple positional arguments are familiar to me, but

What is the plan for an attack?

My general idea is to use EH / CH on the wings, but my impression is that attacking doesn't pay. Pulling horses can be undone.

What if he plays very defensively, undos drags and avoids complications?

I get the impression that it's about waiting for a mistake by the opponent.
Or is it all about pulling rabbits?
Swarm a trap at some point?
Reposition pieces have to strongest local piece?
Keep dragging pieces until one is caught?
People must have some general recipe?!

And often in the opening I find myself not using up all 4 steps. Is that a mistake?
Title: Re: Plan for attack?!
Post by Fritzlein on Jan 23rd, 2008, 11:24am
Ha ha, funny you should ask that.  The Arimaa community has been trying to find the answer to that question for over five years, and still nobody knows.  Apparently it takes a skill level of at least 2700 before you can find out ;-)

The three leading contenders (with lots of sub-variations) are

1. Attack with only your elephant.  Use it to pull something to your side, a rabbit if nothing else, and that will give you an advantage.

2. Attack with elephant and horse.  You might be able to do big-time damage, but more likely the horse will have to retreat with just a rabbit pull.  Rabbit pulls are good.  The key is that the opposing elephant can't afford to take your attacking horse hostage, because you will punish him if he does.

3. Attack with elephant and horse, but not hoping to get a rabbit pull.  Instead you hope to gain shared control of a trap for the rest of the game, cement that control with a swarm on that wing, and eventually share control of the opponent's other trap as well.  This type of elephant-horse attack is always supported by additional attacking pieces.

In fact, something other than one of these three or variations might be true as well, but suppose for a moment the answer to your question is one of these three.  How can you tell which one is true?

The difficulty is that a better player can beat you regardless of strategy.  You can agree before the game that you will play (2) while he plays (1), or that you will play (1) while he plays (2), and it turns out that he beats you either way.  It doesn't matter who had the better opening strategy if he has better tactics, better understanding of the midgame, and better understanding of the endgame.

(1) is the strategy I used to be the top-rated player for almost two years.  Over that span nobody had a winning record against me.  But then chessandgo started beating me with (2).

(2) is the current darling of the mainstream, including the reigning World Champion.  But does chessandgo win with it because it is sound, or because he is the best at midgame and endgame play?

(3) was used by robinson to win the 2006 World Championship, and has at times been effectively used by PMertens.  I'm currently using (3) myself just because it is fun to play that way and I want to see whether I can make it work.

I recommend trying a mix of playing styles to help you learn, as froody is doing now.  You want a broad experience.  Arimaa is too deep for any easy answers, at least so far.  If it were that simple, Arimaa wouldn't be a very interesting game.

For example, you can't just file it away that holding a camel hostage is good.  In the World Championship second round, Omar took PMertens' camel hostage, and PMertens punished him for it.  You need to play on both sides of that type of position, find out when it works, when it doesn't, and why.  You need to understand that holding a camel hostage is good if (fill in condition) because (fill in reason).

A lot of the attention Arimaa gets is because the best humans can beat the best computers.  The secret that we regular players know is that Arimaa is incredibly deep and interesting for humans.  If there were an easy way to describe the "right" way to play, then everyone would play that way, and we would get bored pretty soon and play something else instead.  The reason I stick around is that I am still learning something every game after more than a thousand games.
Title: Re: Plan for attack?!
Post by Gerenuk on Jan 23rd, 2008, 1:09pm
Thanks for the summary. I hope many other players find that useful too.

Hmm, generally, if you notice one mistake, is it possible to correct just the last move or the one before? Or can a mistake be burrowed deeper?

1. So after one rabbit, do I go for another one? Is there any use hostaging a horse more than 2 square away from the trap?

2. How will I punish? We both have camels and additionally my horse only is locked. Is there a way to use the slightly higher elephant mobility? What would be a standard setup?

3. What if he protects with the elephant? I swarm the other trap? I surround the elephant?

I guess I need to be able to trap pieces at his place before I can make a goal threat?

I hope I can judge the game better after getting used to the moves. It was quite easy up to bomb2 at least.

I wouldn't stress that humans beat computer so much. It might give the wrong idea to people with less computer knowledge. It's actually human players beating human programmers, because programmers can't decribe what they think that well.
Computers will be much better if programmed with a new approach (that I don't know either :) ).
I see it a statistical data analysis: if done correctly computers can be much better than human judgeing by eye.
Title: Re: Plan for attack?!
Post by Fritzlein on Jan 23rd, 2008, 8:54pm

on 01/23/08 at 13:09:45, Gerenuk wrote:
Thanks for the summary. I hope many other players find that useful too.

You are most welcome.


Quote:
Hmm, generally, if you notice one mistake, is it possible to correct just the last move or the one before? Or can a mistake be burrowed deeper?

Takebacks are not allowed in rated games.  In unrated games, with version one of the client, you should be able to take back more than one move, but I've never had success with that.


Quote:
1. So after one rabbit, do I go for another one? Is there any use hostaging a horse more than 2 square away from the trap?

Usually don't pull a second rabbit until the first is in as good a position as possible, i.e. framed or a solid hostage two squares away.  When you do go for the second pull, try to pull it on the other side of the board so you can threaten something in two traps at once.  Pulling a second piece on the same side is no help because his elephant is defending that side anyway.

If you have a horse hostage more than two squares from the trap, it should be temporary as you try to get that horse into a better position.


Quote:
2. How will I punish? We both have camels and additionally my horse only is locked. Is there a way to use the slightly higher elephant mobility? What would be a standard setup?

Sometimes you can't punish someone for taking your horse hostage with his elephant.  We are not sure when you can and can't.  The ideal attacking strategy is if your horse is hostage on a6 with his elephant on b6, you want your elephant on d6, and you try to attack g6 with your camel since his elephant can't get to your camel.  This is very effective for the attacker, but a smart defender isn't going to wait around for it.  The defender will try to frame your horse, or hold it hostage from c5, not b6.  If you try to stop his elephant from being on c5 by putting your elephant there, the defender might be able to bring his camel through c6 to hold the horse hostage in place of the elephant, putting you in a bad situation.


Quote:
3. What if he protects with the elephant? I swarm the other trap? I surround the elephant?

Yes, you can try to immobilize his elephant, and free your elephant to roam the rest of the board.  Once you get his elephant sufficiently stuck (not necessarily totally blockaded, but committed to defense) while your elephant still has mobility, you are in control.  In the mean time, while you are fighting over mobility, you can try to flip out pieces towards your home trap (for your supporting pieces to capture; your elephant stays on the attack!) so that his elephant can't defend his home trap and your home trap at the same time.


Quote:
I guess I need to be able to trap pieces at his place before I can make a goal threat?

Usually, but not always.  You can look at the 2007 Postal Tournament, chessandgo's victory over me, for an example of someone winning without making captures.


Quote:
I hope I can judge the game better after getting used to the moves. It was quite easy up to bomb2 at least.

There is a lot to learn about Arimaa.  Even after mastering all the bots, you will find that human players surprise you with strategies you have never seen.


Quote:
Computers will be much better if programmed with a new approach (that I don't know either :) ).
I see it a statistical data analysis: if done correctly computers can be much better than human judgeing by eye.

I eagerly await the new computer approach!
Title: Re: Plan for attack?!
Post by Gerenuk on Jan 24th, 2008, 1:44am
Are there more annotated games, than the one in the wikibook?

And that question still puzzles me:
Do good players ever skip part of their moves? Sometimes at the beginning I feel like there's nothing to improve. If a particular pull can be undone, so why do it?
Title: Re: Plan for attack?!
Post by clauchau on Jan 24th, 2008, 7:53am

on 01/24/08 at 01:44:07, Gerenuk wrote:
If a particular pull can be undone, so why do it?


Your opponent may need more steps than you to completely undo what you did. You might like the position slightly better after her partial or complete undo. After my 4-steps pulling a dog on Move 6w in my current postal game with arimaa_master, my opponent had to move a rabbit forward prematurely and my elephant was then closer to the dog, so that I pulled it again with only 3 steps and got my horse 1-step closer on Move 7w.

Or you may simply want to prevent her from making 4 related steps in a row elsewhere by forcing her to answer your pulling.

Even in non decisive moves there is a diluted sense of initiative in Arimaa as in Go.
Title: Re: Plan for attack?!
Post by RonWeasley on Jan 24th, 2008, 9:13am
In answer to another part of your question, even good players will occasionally use less than four steps in a move. This happens in some endgames where pieces must be placed just right.  Sometimes you see this in quiet openings or in blockade positions.
Title: Re: Plan for attack?!
Post by Fritzlein on Jan 24th, 2008, 11:36am
Yes, three steps is occasionally best in endgame positions and rarely best in midgame and opening positions.  However, I can't recall any position in which I thought two steps was better than four.  Usually the sides are in some sort of race, whether a race to goal or a race to achieve positional objectives, and using fewer than four steps loses you time in the race.

Don't stress out and run down your clock looking for a good fourth step if you don't see one quickly; sometimes the gain from a fourth step is worth less than ten seconds on the clock.  That said, if you find that you are often making three-step moves, or ever making two- or one-step moves, you need to stop and evaluate why you can't think of anything to do.  Is there nothing worth achieving anywhere on the board?  Is every single piece you control in the best possible position?  If you are often waiting around for your opponent to do something, you are not playing to win, you are playing to lose slowly, and even that probably won't work.
Title: Re: Plan for attack?!
Post by Fotoelektrika_Angela on Feb 6th, 2008, 5:35am
Games from top players are good examples.


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