Arimaa Forum - Print Page


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

        Arimaa >> General Discussion >> plz say more about the design decisions
        
(Message started by: bzipitidoo on May 2^nd, 2013, 1:41am)

Title: plz say more about the design decisions
Post by bzipitidoo on May 2^nd, 2013, 1:41am

Hello, everyone. Yes, I know, this isn't the forum for introductions.

I played chess enough to reach a class A rating in the early 1990s, then stopped with intentions of returning but never did. Computers becoming better than the best people was part of one reason. It's not just that computers are better, it's the brainless, unthinking brute force computation manner in which they triumphed. I felt that what they really demonstrated is that chess is not that interesting. At the least, it showed chess is not such a good measure of natural intelligence, despite its reputation. But I still read about chess from time to time. I was looking at some chess history on Wikipedia a few days ago, and saw Arimaa. So I've played about 20 games now against bots.

A way Arimaa is like chess is that between decent players of equal skill, it's a long grind. 60 to 80 moves to beat some of those stubborn C and B level bots. Perhaps that's unavoidable for any decent strategy game, but I've become accustomed to the fast play of the Euro board games of the past 20 years, games such as Puerto Rico.

A mistake I see a lot of these weak bots making is repetition. That's just a symptom of the real problem of course, which is that they don't plan, they only calculate. They often spend 3 or 4 moves to do something I can undo in 2 or 3. Quite brainless. So I undo it, and use the extra moves to advance some other pieces. Then they compound the brainlessness by doing the same 3 or 4 moves again next turn.

I am wondering how the designer of the game arrived at his decisions. Yes, I read the bit about making it hard for computers by having a high branching factor, and deep planning. yet still easy enough for a child to understand the rules. And to be extremely economical, players can use a standard chess set. Some sorts of ideas I imagine weren't even considered, because they would be obviously impractical, such as allowing mulitple pieces to occupy one square. So far, so good.

But, why turn based? Why not a simultaneous movement? Players note down their next moves in secret, then when both are done, reveal and execute. Would need some kind of resolution mechanism to deal with conflicting moves, of course. But this would make moot all the discussion about whether gold or silver has a starting advantage.

The idea of assigning ranks to the pieces strongly reminded me of Stratego. The business of pushing, pulling, freezing and unfreezing pieces, rather than capturing, is a bit different. No doubt there are games with similar mechanisms, but I don't know them. But the trap squares seem to provide enough means to advance the game to a conclusion. That's a little like the king's palace in Chinese chess. Movement of 1 space at a time is very like Shogi. Rabbits are obviously very like pawns in chess. But also they are like 8 little kings, in that you lose if your last rabbit dies.

But why are there 4 trap squares, and at c3,c6,f3, and f6? Why not a knights tour kind of pattern, like c4,d6,f5,e3? Or, let each player decide where 2 of the traps will go at the start? Maybe the traps should be mobile?

In short, what ideas were tried, and how were they tested?

Title: Re: plz say more about the design decisions
Post by browni3141 on May 2^nd, 2013, 4:52am

The information you seek is here, but I think it's fairly scattered. I recall him saying a few things about how he decided on the current rules. I think omar mostly decided on rules through play testing with his son. They tried different things and tried to figure out what did and didn't work. At one point there were no trap squares, but it was decided that a game without captures was not very fun and it was hard to make progress. You'd have to ask him how he came up specifically with the idea of trapping as a capture mechanism.
Many variants have been proposed, and I think one of them involved moving traps, but I'm not sure. Try using the search function if you want and don't mind wading through a ton of posts.
I'm not sure how he came up with the placement and number of trap squares, but it seems to work really well. Each square on the board is pretty close to one trap.

Title: Re: plz say more about the design decisions
Post by Fritzlein on May 2^nd, 2013, 8:47am

on 05/02/13 at 01:41:34, bzipitidoo wrote:

A way Arimaa is like chess is that between decent players of equal skill, it's a long grind. 60 to 80 moves to beat some of those stubborn C and B level bots.

Not necessarily. At one point we noticed game turning into long grinds. We speculated that as players got better and better, games between equal players would stretch out further and further. However, it didn't happen. It turns out that "away" strategies are as viable as "home" strategies, which keeps games from becoming super-slow rabbit-pulling contests. The average game length between strong players has stayed at about 45 moves across the years. See http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=talk;action=display;num=1262974905;start=0#0 (Note that the final column is for games played to completion, with no resignation. If we included resignations as the chess culture would dictate, it would shave several moves off of average game length.)

Quote:

But, why turn based? Why not a simultaneous movement? Players note down their next moves in secret, then when both are done, reveal and execute.

In games with simultaneous movement, there is an inherent element of luck. It is an observation from game theory that there is usually no dominant pure strategy, and that the Nash equilibrium (the closest thing to the concept of perfect play) requires a mixed strategy of randomizing among moves.

This is not to say that a game must be perfect information to be a good game, only that in games where moves are simultaneous you can't "control your own fate" in the sense that you can in chess and Arimaa.

Title: Re: plz say more about the design decisions
Post by chessandgo on May 2^nd, 2013, 11:55am

Yeah. I don't know how many moves the WC 2013 averaged, mine should be around 35 moves. That's not too long (a lot less than a chess game played to the end).

Title: Re: plz say more about the design decisions
Post by mattj256 on May 11^th, 2013, 2:47am

on 05/02/13 at 01:41:34, bzipitidoo wrote:

A way Arimaa is like chess is that between decent players of equal skill, it's a long grind. 60 to 80 moves to beat some of those stubborn C and B level bots. Perhaps that's unavoidable for any decent strategy game, but I've become accustomed to the fast play of the Euro board games of the past 20 years, games such as Puerto Rico.

Hi bzipitidoo! I wanted to reply earlier but I've been busy with other things. If I'm understanding correctly, part of your question is: "Why did Omar design a game that consistently takes 60-80 moves to play?" First I'll echo what has been said before: an average game doesn't take that long, although I would imagine a person's playing style has something to do with it. I would expect a player who wins by being patient and pulling rabbits has longer games than someone who's aggressive and tries to take control of an enemy trap right off the bat. But I don't have any statistics on that.

I don't have all the games in a database so I'll give you some anecdotal evidence. In the recently completed 2013 Arimaa Challenge (http://arimaa.com/arimaa/challenge/2013/showGames.cgi) the games took:

0-39 moves: 4 games
40-49 moves: 1 game
50-59 moves: 2 games
60+ moves: 2 games

This is against the world's best computer program with two minutes per move plus reserve time. In particular, in round 3 Mathew Brown ("browni3141") gave himself a horse handicap and still won in 37 moves.

In the recently completed World Championship (http://arimaa.com/arimaa/wc/2013/showGames.cgi) (human vs. human), the length of the games in the final 6 rounds was:

38, 43, 21, 50, 67, 44, 26, 72

I don't have the energy to copy-paste the information from earlier rounds (involving weaker players), but you can go to that link and eyeball for yourself how long the games took. (If a board has "g 38" written above it, it means the game took 38 moves. If a game took less than 10 moves it was probably due to someone losing their internet connection, so those games aren't relevant.)

on 05/02/13 at 01:41:34, bzipitidoo wrote:

A mistake I see a lot of these weak bots making is ... [t]hey often spend 3 or 4 moves to do something I can undo in 2 or 3. Quite brainless.

Yes, that's why they're "weak." The strong bots don't do that.
And many beginners find the weak bots quite challenging!

I looked at some of your past games that took many moves. Here's some random advice. (Some of which may be wrong.)

Your Game 264909 (http://www.arimaa.com/arimaa/games/jsShowGame.cgi?gid=264909&s=w): You moved your elephant to f7 on 22g and allowed it to stay there for a very long time, leaving it on the sidelines. In general d6/e6/d3/e3 are the best squares for the elephant, and c4/c5/f4/f5 are ok but not as good. Anything outside the center four-by-four square could be acceptable temporarily if there's a tactical reason, but permanently stationing your elephant outside that square is generally not a good idea. Because your elephant was on the sidelines you were unable to break the camel deadlock around c6. Even though you had strong goal threats you couldn't convert them so you had to win by elimination instead, which takes much longer. I don't want to be rude, but if you had strong goal threats on both sides of the board by 34g and it took you almost 50 moves to convert that into a win, I don't think that's a problem with the rules! (I'm sure a player who's stronger than me can give you better and more specific advice.)
I see you making a lot of the two-for-one moves in your games, exactly the kind of move that you recognize as bad when the computer does it. I have a feeling if you drastically reduce those moves it will improve the quality of your game, and you'll enjoy the game more, and your games will take fewer moves.

I hope you'll keep playing! (Or at least tell me why I'm wrong.) :)
Matthew

Title: Re: plz say more about the design decisions
Post by Hippo on May 12^th, 2013, 1:57am

on 05/11/13 at 02:47:55, mattj256 wrote:

I looked at some of your past games that took many moves. Here's some random advice. (Some of which may be wrong.)

Your Game 264909 (http://www.arimaa.com/arimaa/games/jsShowGame.cgi?gid=264909&s=w): You moved your elephant to f7 on 22g and allowed it to stay there for a very long time, leaving it on the sidelines. In general d6/e6/d3/e3 are the best squares for the elephant, and c4/c5/f4/f5 are ok but not as good. Anything outside the center four-by-four square could be acceptable temporarily if there's a tactical reason, but permanently stationing your elephant outside that square is generally not a good idea. Because your elephant was on the sidelines you were unable to break the camel deadlock around c6. Even though you had strong goal threats you couldn't convert them so you had to win by elimination instead, which takes much longer. I don't want to be rude, but if you had strong goal threats on both sides of the board by 34g and it took you almost 50 moves to convert that into a win, I don't think that's a problem with the rules! (I'm sure a player who's stronger than me can give you better and more specific advice.)
I see you making a lot of the two-for-one moves in your games, exactly the kind of move that you recognize as bad when the computer does it. I have a feeling if you drastically reduce those moves it will improve the quality of your game, and you'll enjoy the game more, and your games will take fewer moves.

I hope you'll keep playing! (Or at least tell me why I'm wrong.) :)
Matthew

I don't consider moving elephant to f7 in that position that bad, as opponent elephant was on g7, but later that become bad. I would try to maintain the frame at f6 without elephant on f7.

The moves pulling opponent piece to the trap when it can leave by one step are strange to me, it could be seen from both sides in the game.
Capturing piece for free is almost always good, especially when there are no immediate goal threats on either side. Capture is not important just for material, but for the trap controll and possibility of other captures in the trap.

At the end of the game there were almost just 2 defenders ... elephant guarding h,g,(f) files and camel guarding a,b,(c) files. I would try goaling through d,e files ...

Title: Re: plz say more about the design decisions
Post by jhoravi on May 12^th, 2013, 5:56pm

@bzipitidoo,

I too have the same curiosity about the placement of the trap squares. But the explanation in the following site satisfies me:
http://www.boardgamereviewsbyjosh.com/2013/03/arimaa-review.html#comment-form

I wonder if the author considered eliminating enemies by simply pushing them outside the board because it seems logical. Or more interestingly, the possibility of ending the game quicker by adding a vulnerable King!

Title: Re: plz say more about the design decisions
Post by omar on May 17^th, 2013, 9:13am

>I am wondering how the designer of the game arrived at his decisions.

The main objects were:
* The game should be playable with a chess set.
* The game should be difficult for computers.
* The rules should be easy to learn.
* The game should be fun and interesting.

As you go down that list it goes from being very clear and objective to being very vague and subjective. So a lot of personal preferences do come into play. With those objectives in mind I tried out various rule sets until I found something that I felt good about. I do feel that I was very lucky and fortunate in settling on the set of rules that I did. If I had to explain or justify why I picked this set of rules over another set that also seems to work I don't think I would be able to do it. There is a lot of parallel to picking a move in Arimaa. If you are asked to explain why you picked a particular move (say in a postal game where you have plenty of time to think) in the end you really can't explain it. Of course you make sure the move is not an obvious blunder (similar to checking that a set of rules are consistent and don't lead to flawed game), that it meets some strategic objective (similar to picking rules that meet objectives) and beyond that it's your personal preference and style which you can't really put in words or justify.