Arimaa Forum - Print Page


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

        Arimaa >> General Discussion >> Endgame Puzzle 6
        
(Message started by: jdb on Apr 3^rd, 2011, 6:20pm)

Title: Endgame Puzzle 6
Post by jdb on Apr 3^rd, 2011, 6:20pm

Gold to play and win

1w Da1 Rb2
1b cd1 ca2 rb4

http://arimaa.com/arimaa/games/planGame.cgi

Title: Re: Endgame Puzzle 6
Post by ginrunner on Apr 3^rd, 2011, 8:48pm

toughest one yet ... but i think I got it.

spoiler?
The first 2 moves are more or less forced for both sides for the game to go on for any amount of time.
Da1e Rb2e Db1n Db2n
cd1n cd2n Rc2e ca2n
then as of now i figure the best move for gold would be rb4e Db3n rc4n Db4e
after that silver has 2 possible saves and each lead to slightly different lines where the gold dog capture the silver cat on d3 and moves up to d4; the game turns into silver trying to protect the rabbit and goal in the west and the gold dog stopping the threat in the west (threatening rabbit capture the whole time) and moving the gold rabbit to the east to threaten and eventually goal.

I just found a line where silver tries to goal in the east and gold tries to goal in the west but the basic premise still holds and gold still easily has it after the first 2 moves.

Ive noticed a trend in all of your puzzles. If the units were positioned in such a way that pieces were in maybe moved 1 space in a certain direction it would go to the other color.

Title: Re: Endgame Puzzle 6
Post by Fritzlein on Apr 3^rd, 2011, 9:18pm

I got the same answer, ginrunner. We can't both be wrong, can we? ;-) Assuming we are right, then this was the easiest puzzle for me so far; the others either took me longer or I got the wrong answer. I think it is because controlling traps and threatening capture comes way more naturally to me than threatening/blocking goal does. I might be generally better at "win by elimination" puzzles than I am at "win by goal" puzzles. But maybe I just got lucky this time, or maybe it wasn't such a hard puzzle. :)

As always, I thank jdb for the challenge; these puzzles are a great contribution to the community.

Title: Re: Endgame Puzzle 6
Post by ginrunner on Apr 3^rd, 2011, 9:26pm

I think the reason that this one was the easiest for fritz and probably one of the tougher ones for me is that this one is almost a forced set of moves. More seasoned players will realize that they *must* take certain action in this one and newer players like me won't necessarily see that they are forced to move in any way. The other puzzles do not contain any type of obvious forcing and can lead to many disputed lines which is why it may have taken people at Fritz's caliber a bit longer and people at mine either not as much time or we simply don't see the solution. PLayers at my level will not go down every line which allows us to find maybe the correct way a little faster. (Worse players prune better?:P)

Title: Re: Endgame Puzzle 6
Post by robinz on Apr 4^th, 2011, 7:49am

I also got ginrunner's solution, and would agree with his assessment that it is quite hard - but mainly because I had to check quite a lot of possible responses from silver at various points. In hindsight, I do agree that much of it is forced - but, as he says, for an inexperienced player (such as me) this is far from obvious at first.

Title: Re: Endgame Puzzle 6
Post by lightvector on May 15^th, 2011, 3:07am

I've just now been working through these puzzles with my bot, as a way of detecting goal search and hash table bugs, etc. These endgame puzzles are really nice for this, since they go really deep, require precise play regarding win/loss conditions, and often have lots of repeating states, as well as zugzwang. I've encountered and fixed more than one bug in the process.

Anyways, sharp's answer for this one amused me. Apparently it doesn't bother trying to goal, and just goes for a clever forced elimination.

Anyone see any holes in the tactics?

1w Da1 Rb2
1b cd1 ca2 rb4
2w Da1e Rb2e Db1n Db2n
2b cd1n ca2n cd2n Rc2e
3w rb4e Db3n rc4n Db4e
3b ca3n ca4n ca5e cb5n
4w Dc4e Dd4n Dd5s rc5e
4b cb6n cb7e cc7e cd7s
5w Dd4e rd5s rd4w De4w
5b cd6n cd7w pass
6w Dd4n cd3n cd4e Dd5s

1w Da1 Rb2
1b cd1 ca2 rb4
2w Da1e Rb2e Db1n Db2n
2b cd1n ca2n cd2n Rc2e
3w rb4e Db3n rc4n Db4e
3b ca3n ca4n ca5e rc5e
4w Dc4e cd3w cc3x Dd4s Dd3n
4b cb5e rd5e re5e rf5e
5w Dd4n Rd2e Dd5e De5e
5b cc5e cd5e pass
6w Df5s rg5w rf5n rf6x Df4n

Title: Re: Endgame Puzzle 6
Post by tize on May 15^th, 2011, 12:39pm

The only move that could be different is the 5b in your second line, it doesn't change anything more than pushing the elimination one move further away. Marwin verifies all other moves in your solutions, so if we doesn't have the same bugs the lines seems to be correct. :)

Title: Re: Endgame Puzzle 6
Post by Fritzlein on May 15^th, 2011, 4:08pm

on 05/15/11 at 03:07:29, lightvector wrote:

Anyways, sharp's answer for this one amused me. Apparently it doesn't bother trying to goal, and just goes for a clever forced elimination.

We mere carbon-based intelligences also came up with the same plan of attack; it seems to be forced by the position rather than being sharp's idiosyncrasy. Indeed, from a human perspective one might say that realizing that one should play for elimination rather than playing for goal is the key to solving the puzzle. (which is why I hid that information in my first comment ;))

Title: Re: Endgame Puzzle 6
Post by lightvector on May 15^th, 2011, 6:17pm

Ah, yes, I missed that.

Anyways, done testing and fixing endgame search bugs. Sharp now solves all of the endgame puzzles correctly, as far as I can tell. And does it within CC time controls ;D, with all but 2 of them in less than 12 seconds.

Bots are scary at the deep endgame...

Title: Re: Endgame Puzzle 6
Post by Fritzlein on May 15^th, 2011, 6:35pm

on 05/15/11 at 18:17:51, lightvector wrote:

Bots are scary at the deep endgame...

Agreed. This may be partly due to humans having studied endgames less to date, but I expect mostly it is a permanent disadvantage to humanity that will only get worse.

Title: Re: Endgame Puzzle 6
Post by Tuks on Jun 6^th, 2011, 7:12am

i think i found it in two minutes...if my solution ends up being correct

one thing to keep in mind though is that a bot reevaluates each position form the start whereas humans remember tactics and strategies formed from prior moves so yes, it takes a minute or two to solve this now but had this shown up in a game i probably could solve it in 30 seconds or less because i will have already gone over potential position strengths (like which is better to go for, elimination or goal)

i'm not debating that bots are better at end games though :)