Arasan goes online
Jon Dart
I've been playing the latest version of Arasan (4.0) on FICS over
the past few weeks, using the handle "ArasanX". (I have also given
a copy of the program to the owner of handle "Arasan", and he's
run it a little bit).
First of all, I'm very happy to be able to be able to put the program
on line for testing, and want to thank the FICS admins for allowing
it, and Tim Mann for doing Winzippy, which I'm using to connect.
Arasan has been playing mostly blitz games. It doesn't allow lightning
(less than 5 minute) games. Currently its blitz rating is 2065. It is
probably a bit stronger than that, since when I first put it on line, it
was not using its opponent's time, and it also resigned several games due
to crashes - both these things are fixed now.
Arasan has generally done quite well against human players. It has not
faced any IM or GM players, but has won against several Experts and a
couple of candidate Masters. I remember Bruce commenting some time back
that playing Experts will really test your program's king safety
scoring, and I think that's true. Arasan got into a number of games
where it looked like its king was in serious danger. Sometimes it lost,
but it has also pretty often escaped when the opponent misplayed the
attack or overlooked a cheapo. I have put in some more penalties for
having the king open to attack and that seems to help.
Generally I think trying to out-play the computer tactically is not a
good strategy: the computer won't fall for a shallow combination
but you might, especially if you are short of time. But sometimes it
works: the computer weakens its position and doesn't see the danger
until it is too late.
I have also had to tune the time management code quite a bit. Arasan
will now use more time when it is ahead on the clock, and use a little
less when it is behind. It has seldom lost on time. One player made it
lose a 5 0 game by just moving back and forth very rapidly in a closed
position. Since Arasan currently doesn't allocate time in increments of
less than 1 second, it eventually ran out of time and lost. I consider
this a form of abuse, but it hasn't been repeated, so I'm not worrying
about it right now. If I ever decide to allow lightning games, though,
I'll have to start accounting for time in fractions of a second.
Arasan 4.0 should be released in a few months. Right now it is still
experimental and I expect to be changing it as it gets more testing.
--Jon
Robert Hyatt
Jon Dart (jd...@best.com) wrote:
: I've been playing the latest version of Arasan (4.0) on FICS over
Warning #1. You had *better* allocate time in 1/100th's of a second
or *less*. There are several players that will prey on you once your
rating is high enough to make you "attractive". I can show you at least
one game of "*" vs Crafty where in a 3 0 bullet game Crafty lost after
240 moves were played... yes, the human averaged 80 moves per minute.
I've fixed Crafty so that is is basically impossible to do this, and I
have tested it by having it play hundreds of games vs itself, with a
time control of game/1 second, and it easily reaches 100 moves at times,
if one side doesn't lose before that. However, you need to be ready for
this stuff. :)
Warning #2. Experts are dangerous, but they aren't the most dangerous
you'll see. IM's love to attack, and they don't miss many cheapo moves,
so you have to learn to "retreat 15 yards, dig in, and save the equipment"
at times. This was the point that caused Bruce and I to both use the
famous asymmetric king safety code. GM's generally don't attack, *unless*
they notice you are weak at defending. Then they will do it every game,
and they *don't* make many weak moves in doing so. Try chess.net and
ask Roman to give you a whirl to see what I mean. Bottom line is long-
term kingside attacks are the only achilles heel that I see with any
regularity in any of the better programs, unfortunately I see it too
often in all of them.
I suspect you are going to end up investing a huge amount of time in
a good king safety analysis. I did. Bruce did. I still modify mine
all the time trying to catch more problems. In doing this, you will
also lose the ability to attack if you aren't careful, because you tend
to become so defensive after seeing the Stonewall every time you play
black and castle right into a horrific king-side attack, that you really
play the kingside "close to the vest" and don't push anything.
You will also find that the common "h3" or "h6" move to make luft for
your king is going to get you killed more times than it prevents a back
rank mate. Push that pawn and you give your opponent an instant target
that just asks for the g-pawn to advance and open files.
When you have time, stop by ICC or chess.net as well. I'll add you to
crafty's "auto-kibitz" list when you tell me your handle there and you can
compare what Crafty's seeing to what you see. I find that usefule when I
play Bruce so we can see what each other's program is searching. If he's
faster, I can't do much about it, but if he is seeing something way earlier
than I do, I get interested and can find a way to help that case...
Bob
lensp...@aol.com
>
>Arasan 4.0 should be released in a few months. Right now it is still
>experimental and I expect to be changing it as it gets more testing.
Will you be releasing the source code as with the earlier versions?
Jon Dart
Yes, the source code will be included. In response to Bob's comments:
I appreciate the advice, and will work on the time control problem.
I'm not really aspiring to beat GM's at this point though. I tend
to measure progress relative to where my program has been, rather than
relative to other programs or the whole universe of chess players. It
has gone from buggy and weak to less buggy anad not so weak, so I'm
happy. I also tend to look on ICS play as just another testing route
rather than an end in itself - it tells me things about the program that
I can't learn from automated tests. I'm not going to get too hung up
on what its rating is, although I'd like to see some improvement over time.
In article <19970107083...@ladder01.news.aol.com>,
Stheno
Are you sure you were not put up against another computer? I dont see
how its possible for any human to constantly play that quickly. Your
hand/eye co-ordination would have to be incredible.
Richard
Ronald de Man
Stheno <rls...@ix.netcom.com> writes:
>Robert Hyatt wrote:
>> Warning #1. You had *better* allocate time in 1/100th's of a second
>> or *less*. There are several players that will prey on you once your
>> rating is high enough to make you "attractive". I can show you at least
>> one game of "*" vs Crafty where in a 3 0 bullet game Crafty lost after
>> 240 moves were played... yes, the human averaged 80 moves per minute.
>>
>Are you sure you were not put up against another computer? I dont see
>how its possible for any human to constantly play that quickly. Your
>hand/eye co-ordination would have to be incredible.
>Richard
Not necessarily. Against computers it is often not too difficult to get
a position in which you can just move your rook up and down without
thinking. Against gnuchess the following often works:
1. e3 e5 2. Ne2 d5 3. d3 Nc6 4. Nd2 Nf6 5. g3 Bc5 6. h3 Bf5 7. Bg2 O-O 8.
b3 Bd7 9. a3 a6 10. Bb2 Re8 11. Rh2 Bf5 12. Rh1 h6 13. Rh2 b5 14. Rh1 Re6
15. Rh2 a5 16. Rh1 Qe7 17. Rh2 Rd8 18. Rh1 Qe8 19. Rh2 Qd7 20. Rh1 Qe7 21.
Rh2 Qe8 22. Rh1 Kh8 23. Rh2 Qg8 24. Rh1 Bg6 25. Rh2 d4 26. e4 Bh5 27. Rh1
Qh7 28. Rh2 Ree8 29. Rh1 Qg8 30. Rh2 Rd7 31. Rh1 Re6 32. Rh2 Qc8 33. Rh1
Kg8 34. Rh2 Qe8 35. Rh1 Rd8 36. Rh2 Qd7 37. Rh1 Kh8 38. Rh2 Rde8 39. Rh1
Kg8 40. Rh2 Qd8 41. Rh1 Qc8 42. Rh2 Qd7 43. Rh1 Rd8 44. Rh2 Kh8 45. Rh1 Qe8
46. Rh2 Kg8 47. Rh1 Qe7 48. Rh2 Qd6 49. Rh1 b4 50. a4 Qd7 51. Rh2 Kh8 52.
Rh1 Qc8 53. Rh2 Kg8 54. Rh1 Qa8 55. Rh2 Qb7 56. Rh1 Kh8 57. Rh2 Ree8 58.
Rh1 Kg8 59. Rh2 Qa8 60. Rh1 Kh8 61. Rh2 Qa6 62. Rh1 Re6 etc.
Ronald de Man
Robert Hyatt
Stheno (rls...@ix.netcom.com) wrote:
: Robert Hyatt wrote:
: >
: Are you sure you were not put up against another computer? I dont see
: how its possible for any human to constantly play that quickly. Your
: hand/eye co-ordination would have to be incredible.
: Richard
I'm positive. I've talked with the guy, and watched many games. He is
*very* fast, and likes to reach locked positions and "shuffle" by clicking
on a piece that has only one move. Many GUI's will then move the piece
instantly, requiring only one click to do so...
Bob
brucemo
Jon Dart wrote:
>
> I've been playing the latest version of Arasan (4.0) on FICS over
> the past few weeks, using the handle "ArasanX". (I have also given
> a copy of the program to the owner of handle "Arasan", and he's
> run it a little bit).
Hooray! If everyone would do this, the world would be a better place.
>[snip]
> I have also had to tune the time management code quite a bit. Arasan
> will now use more time when it is ahead on the clock, and use a little
> less when it is behind. It has seldom lost on time. One player made it
> lose a 5 0 game by just moving back and forth very rapidly in a closed
> position. Since Arasan currently doesn't allocate time in increments of
> less than 1 second, it eventually ran out of time and lost. I consider
> this a form of abuse, but it hasn't been repeated, so I'm not worrying
> about it right now. If I ever decide to allow lightning games, though,
> I'll have to start accounting for time in fractions of a second.
I've done a lot of thinking about this, as this has come up in a zillion
different forms.
I think that if the problem has to do with deficiencies in how your program
plays chess, it's not abuse. It might be unsporting to take advantage of a
program like this, but it's not abuse.
This is clearly a case you can fix. Start recording time in milliseconds.
If people still flag you, use timestamp/timeseal/accuclock, and if this still
doesn't work, you'll simply have to make it move faster.
Remember also that if you don't like a tactic someone is using against you,
you are free to put them on a noplay list, or to hack Zippy so that it simply
ignores challenges from certain people. I haven't seen it stated on any
chess server that player A has a RIGHT to play player B, if player B doesn't
want to play, he/she/it can simply decline or ignore a challenge.
bruce
brucemo
Stheno wrote:
>
> Robert Hyatt wrote:
> > Warning #1. You had *better* allocate time in 1/100th's of a second
> > or *less*. There are several players that will prey on you once your
> > rating is high enough to make you "attractive". I can show you at least
> > one game of "*" vs Crafty where in a 3 0 bullet game Crafty lost after
> > 240 moves were played... yes, the human averaged 80 moves per minute.
> >
>
> Are you sure you were not put up against another computer? I dont see
> how its possible for any human to constantly play that quickly. Your
> hand/eye co-ordination would have to be incredible.
Some interfaces will let you pick up a piece, move it over the square you
want to put it on, and drop it the instant you get a move from the server.
In some positions it's pretty safe to do this, especially against a computer
that's just moving its rooks around on the back row. And if your opponent
surprises you, you just put the piece back and move another one, or move it
to a different square.
These are chess players we are talking about, and often chess players who
have some computer experience, so some of them are not only chess players,
they are hackers. Some of these people are incredibly competitive, they
will submit themselves to almost any trial in order to win a contest
of any sort between themselves and some other entity.
An example of how far people will go is "wild 6" on ICC. When you start a
game of wild 6, there are no pieces on the board. The idea is that you
aren't supposed to play someone a game of wild6, you are supposed to match
yourself, which puts you in a mode where you can add pieces to the board.
Some people figured out that if you match some programs at wild 6, they will
accept the game, discover that they don't have any pieces, and hang until
they flag. These people were playing tons of wild 6 games against computers
in order to steal "wild" rating points. Incredible, because there is
absolutely no pretense that this has anything to do with chess anymore,
it's some other sort of contest entirely.
bruce
lensp...@aol.com
>
>Yes, the source code will be included. In response to Bob's comments:
>I appreciate the advice, and will work on the time control problem.
>I'm not really aspiring to beat GM's at this point though. I tend
>to measure progress relative to where my program has been, rather than
>relative to other programs or the whole universe of chess players. It
>has gone from buggy and weak to less buggy anad not so weak, so I'm
>happy. I also tend to look on ICS play as just another testing route
>rather than an end in itself - it tells me things about the program that
>I can't learn from automated tests. I'm not going to get too hung up
>on what its rating is, although I'd like to see some improvement over
time.
Sounds great! Hopefully I will be able to see it on fics.onenet.net
someday.
Robert Hyatt
brucemo (bru...@nwlink.com) wrote:
: Stheno wrote:
: >
: > Robert Hyatt wrote:
: > > Warning #1. You had *better* allocate time in 1/100th's of a second
: > > or *less*. There are several players that will prey on you once your
: > > rating is high enough to make you "attractive". I can show you at least
: > > one game of "*" vs Crafty where in a 3 0 bullet game Crafty lost after
: > > 240 moves were played... yes, the human averaged 80 moves per minute.
: > >
: >
: > Are you sure you were not put up against another computer? I dont see
: > how its possible for any human to constantly play that quickly. Your
: > hand/eye co-ordination would have to be incredible.
: Some interfaces will let you pick up a piece, move it over the square you
: want to put it on, and drop it the instant you get a move from the server.
: In some positions it's pretty safe to do this, especially against a computer
: that's just moving its rooks around on the back row. And if your opponent
: surprises you, you just put the piece back and move another one, or move it
: to a different square.
: These are chess players we are talking about, and often chess players who
: have some computer experience, so some of them are not only chess players,
: they are hackers. Some of these people are incredibly competitive, they
: will submit themselves to almost any trial in order to win a contest
: of any sort between themselves and some other entity.
: An example of how far people will go is "wild 6" on ICC. When you start a
: game of wild 6, there are no pieces on the board. The idea is that you
: aren't supposed to play someone a game of wild6, you are supposed to match
: yourself, which puts you in a mode where you can add pieces to the board.
: Some people figured out that if you match some programs at wild 6, they will
: accept the game, discover that they don't have any pieces, and hang until
: they flag. These people were playing tons of wild 6 games against computers
: in order to steal "wild" rating points. Incredible, because there is
: absolutely no pretense that this has anything to do with chess anymore,
: it's some other sort of contest entirely.
: bruce
However, I've seen the following happen as a result of playing like the
above:
The position gets pretty well locked up. White (human) starts the shuffle,
and black (Crafty, ahead a couple of pawns) starts moving pieces around
anywhere to avoid a draw. This goes on at high-speed for nearly 50 moves,
then crafty makes a pawn move that sacs a pawn, to avoid the 50-move draw,
but the opponent is playing shuffle-shuffle-shuffle so quickly he doesn't
recognize the hung pawn, shuffles one more time, then crafty plays something
like PxR, there's a 20 second "oh dear" pause, and a resignation. :)
It's dangerous to do the above, but a limited few like to try. About all I
can say is that beating Crafty on time is impossible now, since I spent a
good bit of time to solve the problem inside Crafty, rather than going the
+noplay route as a permanent solution. Now I only have to solve the problems
caused by the opponent playing better than Crafty. :)