Arimaa Forum - Print Page


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

        Arimaa >> General Discussion >> Bots
        
(Message started by: Valueinvestor on Apr 4^th, 2007, 2:33am)

Title: Bots
Post by Valueinvestor on Apr 4^th, 2007, 2:33am

Hi, I'm a new player (I joined a while ago but I didn't play until now) and I've been beating the bots on the ladder just by moving a few pieces into the corner and forcing a rabbit through, while occasionally making a few token defensive efforts. Now I don't know what I'm doing and the wikipedia site suggests this is a bad strategy. So are the bots teaching bad habbits to newcomers by allowing this strategy to work? I assume I would be killed if I tried this against a human player.

Title: Re: Bots
Post by Soter on Apr 4^th, 2007, 4:06am

Hello,

I'm a beginner just like you and i'm sure you'll soon get high quality information from best players. The answer to your question, however, is IMVHO simple and straightforward: test your strategic ideas against a CC bot... "paradigm shift" experience guaranteed :)

Good luck with your games!

Title: Re: Bots
Post by Valueinvestor on Apr 4^th, 2007, 7:46am

Well it doesn't work against botbomb2005p1 so I will have to try a defensive strategy.

Title: Re: Bots
Post by Fritzlein on Apr 4^th, 2007, 8:53am

on 04/04/07 at 02:33:28, Valueinvestor wrote:

So are the bots teaching bad habbits to newcomers by allowing this strategy to work?

Hi, Valueinvestor. I probably wrote the Wikibook page to which you are referring. I haven't made substantial edits to the strategy pages in a long time, even though Arimaa theory has continued to advance, so what you read on line is outdated. Here's my take on historical development of strategy.

When the Arimaa server first started in late 2002, there was all kinds of crazy experimentation, since nobody knew how to play. When Don Dailey came out with the first bot, Occam, it was as good as any human, so there was a lot of human focus on learning how to beat that bot. Occam charged forward with pieces and rabbits, and folks learned you could let Occam come to you, and to tear apart its feeble attacks. I suppose that defensive play gained some credence just because it was the safest, easiest way to beat Occam. (By now we know many ways to beat ShallowBlue and Arimaazilla, the descendants of Occam.)

Just when humanity was patting itself on the back for learning how to beat the first bot, Fotland came out with early versions of Bomb in mid-2003. Bomb again was as good as any human, and better than most, despite humanity's experience with Occam. It looked like humanity might lose the Arimaa Challenge in the first year, so the focus of play suddenly shifted to trying to beat Bomb.

Bomb didn't recklessly advance rabbits like Occam, but it was very aggressive with pieces. It would try to advance all of its heavy pieces and take over your traps, not with an eye to immediate goal, but as a means to winning material. This worked really well, and humans had a hard time coping with it until they discovered the camel hostage strategy. Once the knowledge of camel hostages permeated the community, folks started beating Bomb by taking a camel hostage in every game.

Fotland was forced to respond by making Bomb more defensive, and Bomb has never been the same since. However, we should bug Omar to put bot_Arimaanator (an early version of Bomb) back on line so you newcomers can get a taste of what it was like.

Anyway, by the time I joined in mid-2004, it was already well-established that a lone-elephant attack was the ONLY sound opening. Due to Occam we "knew" that advancing rabbits was bad, and due to Bomb we "knew" that advancing pieces other than the elephant was bad. My strategy articles reflect this conventional wisdom.

Since then, however, many new players have defied conventional wisdom, and had great success doing so. In 2005, Robinson scored brilliant successes with elephant + horse attacks, and Adanac did well with elephant + camel + rabbits attacks. PMertens has found ways to advance rabbits behind his pieces to totally tie up your position, not for a quick goal, but for a strategic win. In the 2006 postal tourney, 99of9 went undefeated, often using an elephant + 2 horses attack. Chessandgo has steamrollered the opposition with an elephant + horse + delayed camel attacks.

I happen to remain a lone-elephant attacker, but that is more from habit and laziness than from a deep conviction that it is the only right way to play. I am extremely grateful that all the new players haven't read my Wiki articles and accepted it as gospel. If the bot ladder of today teaches folks the "bad habit" of attacking play, that's good, because it is at least different from the habit of defensive play that we learned from the early bots. I love it every time I see a new player having success by defying conventional wisdom in any way. (For example Brendan likes to keep his camel on the third rank, rather than the second, which is proving difficult to take advantage of, and gives him extra attacking options.)

At the end of the day, the correct strategy is the one that works. We don't know what that strategy is. Hopefully you can help us figure it out!

Title: Re: Bots
Post by IdahoEv on Apr 4^th, 2007, 2:40pm

I agree with what's been said that the strategy you describe will only work against the early bots. Bomb will kick your ass if you try it.

A thought sparked by Fritzlein's post: I spent a good bit of last fall studying attacking strategies like E-H and E-H-H. In January I got to where I could beat Bomb with them reliably.

BUT - I found I could *not* beat Occam/Arimaazilla with them. If I rushed forward to take control of one of his traps, he simply steamrollered me with his aggressive advance. That bot I can beat easily with a defensive strategy + nickel-and-dime, but I just can't make offense work against him, because he responds by flooding the traps with small pieces instead of being lured into decentralizing his phant.

This has me wondering, then: an occam-like inexorable push of all pieces is a terrible offensive strategy, but could it be be an effective response to the E-H, E-M-R, and E-H-H aggressive strategies employed by Robinson, 99, PMertens, and others? If Arimaazilla can beat my offense with it, could Fritzlein beat 99of9's offense with it?

I'd love to see some of the top players demonstrate offensive technique against an advancing bot like Occam, in case it's just that I'm an insufficiently strong player to make it work. And I'd love to see some top players try Occam-like strategies as a counterpunch to attacks.

Title: Re: Bots
Post by Fritzlein on Apr 4^th, 2007, 9:54pm

This may not be exactly what you are getting at, IdahoEv, but Robinson used to play in a very similar spirit. If you launched an E+H attack against him, he would launch an even bigger attack against you, usually E+H+small pieces. This strategy had me completely stymied for a while.

Eventually (too late for the 2006 World Championship) I hit upon the plan of pulling out of my E+H attack as soon as Robinson had committed any little pieces. Then I would have no vulnerabilities in his camp, whereas I would usually get good play against some non-horse piece of his in my camp. Of course, I can't say this is any kind of definitive answer, because I haven't played Robinson enough since then. But maybe he doesn't try that technique so much any more precisely because my counter-strategy worked in a few games.

Since then I've noticed that PMertens and chessandgo in particular (and probably all the other top players too) are very willing to turn on a dime. If you see PMertens' attack coming, and you decide to attack him harder instead of defending, he can fall back in an instant, and suddenly he has your pieces hostage in a devilish way.

In general, I don't think it can be correct to decide that you are going to go on the attack come hell or high water. If your attack is sound, then the other guy can attack more recklessly and force you to defend. Conversely, if your attack is so forceful that the other guy can't top it, then your attack must be too reckless to be sound. (Plug your ears, Value_investor, or at least don't believe it just because I say it.)

Title: Re: Bots
Post by Valueinvestor on Apr 4^th, 2007, 10:00pm

I'll just have to experiment with different strategies. It's possible that certain attacks seem to fail because no one has figured out how to play them properly.

Title: Re: Bots
Post by NIC1138 on Apr 6^th, 2007, 4:34pm

I love to defy common wisdom!! 8) None of the "that doesn´t work" rules-of-thumb are completely true! There is always the context......

You just have to be careful, learn when to stop, learn to see when you are prone to an attack... That is the real learning. Just like in martial arts, where the "kata", or "forms" are just the path to deeper undestanding of the complete dynamics of things.

The rules-of-thumb mean not that the strategy is flawed, but that it´s difficult to perform them in a safe way!... They are not inherently flawed!... (unless, of couse, strange strategies like killing your phant, camel and a horse in the first move. ::) This is worse then the "null strategy"!)

Title: Re: Bots
Post by PMertens on Apr 10^th, 2007, 5:42pm

on 04/04/07 at 22:00:01, Valueinvestor wrote:

I'll just have to experiment with different strategies. It's possible that certain attacks seem to fail because no one has figured out how to play them properly.

absolutely .... so be bold and try new variations and have fun experimenting :-)