GetClub Easy Level now as good as Rybka at 5 sec / move
Sanny
taking 5 sec / move.
I gave GetClub 8 times longer time. And the two were playing equal.
So Rybka is just 8 times stronger than GetClub.
GetClub takes 8 times longer to think a move which Rybka takes to make
same move.
I acheved this strength 3 months back. But after modifications the
game get detoriated.
I removed all new modifications and find GetClub is playing as good as
earlier. Now its just 8 times weaker than Rybka.
Bye
Sanny
Play Chess at: http://www.GetClub.com/Chess.html
=============================================
Play Chess: http://www.GetClub.com/Chess.html
A Descision making Game
Enjoy & Chat: http://www.GetClub.com
Talk with Computer
Earn $1600/ month: http://www.getclub.com/salesjob.html
Get $200-$400 per sale.
Business Planning Software: http://www.softtanks.com/
3 Versions of Software
1. Small Version: for Owner of Small Shops/ Companies
2. Medium Version: For: Managers working in Small/ Medium sized
Companies or CEO/ Business Owners.
3. Large Version: For: Executives in Large Companies/ Business Owners
=============================================
Poutnik
>
> I played GetClub against Rybka. GetClub taking 40 sec / move and Rybka
> taking 5 sec / move.
>
> I gave GetClub 8 times longer time. And the two were playing equal.
>
> So Rybka is just 8 times stronger than GetClub.
I would say this math does not work.
Why not to compare ELO points ?
--
Poutnik
The best depends on how the best is defined.
Sanny
> > taking 5 sec / move.
>
> > I gave GetClub 8 times longer time. And the two were playing equal.
>
> > So Rybka is just 8 times stronger than GetClub.
>
> I would say this math does not work.
> Why not to compare ELO points ?
According to Rybka site If a program is Twice Stronger it has +50
rating.
As when a computer program is played on is Twice faster computer It
plays at +50 Rating.
Assuming the Rybka version I am playing is having a rating of 2600.
Since it is taking 1/4th the time Its eating is 2500.
GetClub Easy Level is taking 8 times longer So 3*50 = 150 rating lower
than Rybka.
So Easy Level is 2500- 150 = 2350 Rated.
So Easy Level is playing Like 2350 Rated Player.
I dont know if above hypothesis is correct or not. If correct the
GetClub has below ratings.
So Rough Rating to all levels are.
Beginner: 2100+
Easy: 2300+
Normal: 2500+
Master: 2700+
Play Chess at: http://www.GetClub.com/Chess.html
=============================================
Play Chess: http://www.GetClub.com/Chess.html
A Descision making Game
Enjoy & Chat: http://www.GetClub.com
Talk with Computer
Earn $1600/ month: http://www.getclub.com/salesjob.html
Get $200-$400 per sale.
Business Planning Software: http://www.softtanks.com/
3 Versions of Software
1. Small Version: for Owner of Small Shops/ Companies
2. Medium Version: For: Managers working in Small/ Medium sized
Companies or CEO/ Business Owners.
3. Large Version: For: Executives in Large Companies/ Business Owners
=============================================
Bye
Sanny
taylor....@comcast.net
>
>
> So Rough Rating to all levels are.
>
> Beginner: 2100+
> Easy: 2300+
> Normal: 2500+
> Master: 2700+
Sanny neglects to mention that in converting GetClub ratings to Elo
ratings, one must take off about 1,000 to 1,200 points.
help bot
Granted, before their removal, Mr. Kingston's
assessment would be an accurate one, but now
that all the disimprovements have been removed,
the GetClub program is somewhere near the
midpoint between Sanny's guesstimates and
those of Mr. Kingston. I am happy to see that
after years of coaching, Sanny has now adopted
the position that there is, say, 200 points between
each of the different levels.
It is very dificult to assign accurate estimated
ratings to the GetClub levels on account of the
fact that its human opponents have no real time
limit for considering their moves. However, one
gets a vague idea of the power of resistance to
force applied, just by playing the program a few
hundred times, in many different types of chess
positions, and comparing that to the resistance
"felt" in tournament play against humans of
various ratings.
My feeling is that at times, the program can be a
dangerous opponent tactically -- just like ordinary
chess engines -- but this must be balanced
against GetClub's awful endgame skills, inferior
openings knowledge, and the occasional glitch in
mid-game, along with not handling en passant
captures properly.
On the whole, remembering how important
tactics are in chess, I would put GetClub very
near the mid-point-- just subtract five or six
hundred points from Sanny's numbers above.
Of course, this assumes the human or other
opponent will /not/ spend an inordinate amount
of time considering his moves, although there
is nothing to stop him from doing so. Without
the ticking away of strictly limited thinking time,
GetClub is handicapped by its own liberal
policy-- just as it would be if it allowed move
take-backs when the opponent blundered.
-- help bot
ph...@ideastakingshape.co.uk
I spend less time on my moves than I would in a OTB match.
Does anyone know what my rating would be in ELO? That should provide
an upper limit for estimations of GetClub's strength.
Phil.
Sanny
You used to win when the GetClub Program was bad. Recentrly the game
has been improved a lot.
Try playing with Normal Level: 5 min / move and I am surte you will
find it very difficult to win the games.
Last 2 months trhe program detoriated as the improvements were not in
right directioin. Now you will face a very strong opponent.
Well I agree you are strong player. Still Normal Level will beat you.
As the GetClub game is playing wonderful.
Bye
Sanny
taylor....@comcast.net
> I played GetClub against Rybka. GetClub taking 40 sec / move and Rybka
> taking 5 sec / move.
>
> I gave GetClub 8 times longer time. And the two were playing equal.
This is utter nonsense. Just from idle curiosity, I pitted my Rybka
3 UCI in blitz mode against GetClub's easy level. The result was a
slaughter:
GC (easy) - Rybka
1. e2-e4 [2] e7-e6 [40]
2. Ng1-f3 [0] c7-c5 [12]
3. Nb1-c3 [190] d7-d6 [14]
4. Bf1-c4 [150] Ng8-f6 [10]
5. d2-d4 [52] c5-d4 [10]
6. Qd1-d4 [32] Bf8-e7 [18]
7. Bc1-e3 [36] Nb8-c6 [16]
8. Qd4-d3 [42] Nf6-g4 [18]
9. Ke1-c1 [80] a7-a6 [22]
10. h2-h3 [56] Ng4-e3 [14]
11. Qd3-e3 [20] b7-b5 [30]
12. Bc4-d3 [30] Qd8-c7 [18]
13. Nf3-h2 [64] Ke8-g8 [18]
14. Nh2-g4 [86] h7-h5 [18]
15. Ng4-h2 [26] b5-b4 [14]
16. Nc3-a4 [68] Nc6-e5 [16]
17. Kc1-b1 [50] Ra8-b8 [20]
18. Rh1-g1 [238] Bc8-d7 [14]
19. b2-b3 [20] Ne5-d3 [16]
20. c2-d3 [52] Bd7-a4 [16]
21. b3-a4 [24] b4-b3 [18]
22. a2-b3 [198] Rb8-b3 [10] 0-1
The numbers in brackets are the time used in seconds. For White, the
numbers are accurate, showing GC's contemplation time. For Black,
deduct about 10-15 seconds from each move to account for the time
spent manually entering moves to and from GetClub by cursor. The 40
seconds for move one was the time it took to boot Rybka up, which is
necessary; if GetClub is not started first it fails to boot. Rybka
replied instantly up until 6.Qxd4.
It is not possible to set my version of Rybka to a strict 5 seconds
per move TL, but it did indeed average about 5 seconds per move.
GetClub spent an average of 1 minute 9 seconds per move, about 14
times longer than Rybka, yet it still got beaten like meringue by a
Sunbeam mixer. It is nowhere near as strong as Rybka. Sanny, as usual,
is full of it.
help bot
This was a good experiment, for the strange
opening nixed Rybka's presumably overwhelming
advantage in openings book "knowledge", and an
attempt was made to balance the two programs'
strengths a bit by severely restricting Rybka's
thinking time. However, we were not informed
as to whether or not Rybka thinks while Sanny's
monstrosity is cranking away, and in such time-
handicap games this could be very significant.
Personally, I have a hard time imagining a
large enough time handicap by which GetClub
could be made Rybka's equal, and this is why I
usually suggest material odds. However, if
Rybka were /not/ allowed to think on the oppo-
nent's time, then perhaps Advance level versus
five-seconds-per-move could prove interesting.
I cannot conduct such an experiment because
of technical difficulties (GetClub freezing up).
-- help bot
Sanny
> 3 UCI in blitz mode against GetClub's easy level. The result was a
> slaughter:
Today there was a further improvement in GetClub. So try a new game.
Lets see how GetClub faces it.
When seeing that game I was wondering how TK was able to find such a
long combination. But since that program was Rybka I can understand it
can calculate to that depth.
Today I played against Jester. Giving 12 sec / move to Jester and
GetClub taking just 10 sec / move.
Here is the game
Recorded Game: http://www.getclub.com/playgame.php?id=DM45499&game=Chess
(Jester 12 sec / move ) - (GetClub Beginner 10 sec / move)
1.e2-e4 e7-e5
2.Ng1-f3 Nb8-c6
3.Bf1-b5 Bf8-c5
4.Ke1-g1 Ng8-e7
5.c2-c3 Ne7-g6
6.d2-d4 e5xd4
7.c3xd4 Bc5-b6
8.Nb1-c3 a7-a6
9.Bb5-c4 Nc6-a5
10.Bc4-d5 Qd8-f6
11.Bc1-g5 Qf6-d6
12.Nc3-a4 Bb6-a7
13.Qd1-d2 Na5-c6
14.Na4-c3 Ke8-g8
15.Ra1-c1 h7-h5
16.Nc3-e2 Nc6-e7
17.Bd5-b3 Rf8-e8
18.Ne2-f4 h5-h4
19.Rf1-d1 Ng6xf4
20.Bg5xf4 Qd6-b6
21.Bf4xc7 Qb6-f6
22.e4-e5 Qf6-g6
23.Nf3-g5 Kg8-f8
24.Bb3xf7 Qg6-f5
25.Rc1-c3 Ba7xd4
26.Qd2xd4 Qf5xg5
27.Rc3-f3 Ne7-f5
28.Qd4-a4 Kf8xf7
29.Qa4-e4 g7-g6
30.Rf3-f4 Kf7-f8
31.Qe4-b4+ Kf8-g8
32.Rf4-g4 Qg5-h5
33.h2-h3 a6-a5
34.Qb4-c4+ Kg8-g7
35.Rd1-d5 a5-a4
36.Qc4-b5 Nf5-e7
37.Rd5-d6 Ne7-c6
38.f2-f4 Nc6-e7
39.Rg4-g5 Qh5-h8
40.Bc7-a5 Kg7-h7
41.Rg5-g4 Kh7-g8
42.e5-e6 Ra8xa5
43.Qb5xa5 d7xe6
44.Qa5xa4 Kg8-f8
45.Qa4-b4 Qh8-f6
46.Qb4-b5 Ne7-f5
47.Rd6-d2 e6-e5
48.Qb5-c5+ Kf8-g7
49.Qc5-c7+ Kg7-h8
50.f4xe5 Qf6-g7
51.Qc7xg7+ Kh8xg7
52.Rg4-c4 g6-g5
53.Rc4-c7+ Kg7-g6
54.a2-a3 Nf5-g7
55.Rd2-d6+ Kg6-h7
56.Rd6-b6 Re8-d8
57.Rb6-f6 Rd8-g8
58.a3-a4 b7-b6
59.e5-e6 Bc8-a6
60.e6-e7 Ng7-e8
61.Rc7-a7 Ba6-c8
62.Rf6xb6 Bc8-f5
63.Rb6-f6 Bf5-d3
64.Rf6-f7+ Rg8-g7
65.Rf7-f8 Bd3-g6
66.b2-b4 Bg6-f7
67.b4-b5 Bf7-g6
68.Kg1-f2 Bg6-h5
69.Ra7-d7 Rg7-f7+
70.Rf8xf7+ Bh5xf7
71.b5-b6 Ne8-g7
72.b6-b7 Bf7-e8
73.Rd7-d8 Be8xa4
74.b7-b8 Kh7-g6
75.Rd8-d6+ Kg6-h7
76.Qb8-f8 g5-g4
77.Qf8-f4 g4-g3+
78.Kf2-e3 Kh7-g8
79.Rd6-g6
Recorded Game: http://www.getclub.com/playgame.php?id=DM45499&game=Chess
So you can see Jester took 80 moves to win the game.
With recent improvements even Rybka will take time to win the Easy
Level.
Sanny
> five-seconds-per-move could prove interesting.
> I cannot conduct such an experiment because
> of technical difficulties (GetClub freezing up).
Long time not seen your game.
The Freezing problem was solved yesterday.
I played a game and it freeze. Now I knew why it was freezing and
corrected the problem.
Now, I think it will not Freeze even with Higher LKevels.
Beginner is playing as good as Jester. So You will find it very
challenging to win the Beginner Level.
Play a few quick games with Beginner Level and tell us how you feel.
Beginner Level plays in 10 sec / move So you can finish a game in
10-15 min only.
Martin Brown
Getclub was only losing gradually until the fateful time waster
18. Rhg1 (ranked about 11/33 moves)
White is already losing but 18. b3 and then 19. Nb2 would hold the
position together for a lot longer by creating a fortress around the
king and seems obvious so I wonder why the evaluation didn't find it
until later.
After that the serious mistakes come thick and fast. The worst in terms
of not having adequate lookahead or king protection was.
20. cxd3 allowing black to nab the bishop free of charge
Unfortunately GetClub cannot see deep enough to reallise this and breaks
his own king position wide open as part of a helpmate!
Qxd3 or Rxd3 are the only two plausible moves here and they are poor,
but they are much better than the move played.
21. bxa4 is unbelievably bad and guarantees total annihilation.
22. axb3 #3 continues the helpmate theme.
Considering that GetClub did not resign at this point I guess we can
deduce that it could not yet see the inevitable forced mate in 3.
Any of Rc1, Rd2, Qd2, Qc1 would be plausible delaying tactics.
Everything else is a mate inside the lookahead horizon (or should be). A
less than 6 ply lookahead when the key moves have the king in check is
lamentable.
Here is a challenge.
What is the oldest historical chess program that can beat GetClub?
Regards,
Martin Brown
taylor....@comcast.net
> > This is utter nonsense. Just from idle curiosity, I pitted my Rybka
> > 3 UCI in blitz mode against GetClub's easy level. The result was a
> > slaughter:
>
> Today there was a further improvement in GetClub. So try a new game.
> Lets see how GetClub faces it.
Sanny, this is always your response. Every time your program gets
walked over like a rug, you say "I just fixed it, try it now."
Bullshit. You haven't been able to improve it substantially for what,
two years now? And yet you expect us to believe you just made it
Rybka's equal in less than a day? Rrrriiight.
taylor....@comcast.net
wrote:
That's a bit like saying "The man was only drowning gradually until
the water covered his head." GetClub played stupidly well before
18.Rhg1. For example moves 13-16 were not only time-wasting, but ended
up posting the knights on rim squares were they were completely
ineffective.
> White is already losing but 18. b3 and then 19. Nb2 would hold the
> position together for a lot longer by creating a fortress around the
> king
A fortress easily breached by a6-a5-a4-xb3 etc. It also leaves the c-
pawn weak and vulnerable on the open file, and the c3 square weak and
ripe for occupation, especially since White has no dark-square bishop.
Your suggestion is probably the least of evils at that point, but the
fact that White has only a choice of evils indicates he's already
losing.
> and seems obvious so I wonder why the evaluation didn't find it
> until later.
>
> After that the serious mistakes come thick and fast. The worst in terms
> of not having adequate lookahead or king protection was.
> 20. cxd3 allowing black to nab the bishop free of charge
> Unfortunately GetClub cannot see deep enough to reallise this and breaks
> his own king position wide open as part of a helpmate!
>
> Qxd3 or Rxd3 are the only two plausible moves here and they are poor,
> but they are much better than the move played.
>
> 21. bxa4 is unbelievably bad and guarantees total annihilation.
So does anything else, since it would leave White a piece down.
> 22. axb3 #3 continues the helpmate theme.
> Considering that GetClub did not resign at this point I guess we can
> deduce that it could not yet see the inevitable forced mate in 3.
>
> Any of Rc1, Rd2, Qd2, Qc1 would be plausible delaying tactics.
For only a few moves, and at great cost. If 22.Rc1 bxa2 23.Kxa2 Qa5
24.Rc4 d5 -+, or 22.Qd2 bxa2+ 23.Ka1 Bg5! 24.Rc1 (if 24.Qxg5 Qc3+)
24...Qb7-+, or 22.Qc1 bxa2+ 23.Kxa2 Qb6 etc. -+.
> Everything else is a mate inside the lookahead horizon (or should be). A
> less than 6 ply lookahead when the key moves have the king in check is
> lamentable.
Sanny's program is lamentable, but far worse is his continual hyping
of it as great, when it has been proven many times to be trash. The
programmer is far more lamentable than the program.
> Here is a challenge.
> What is the oldest historical chess program that can beat GetClub?
Hmmmmm. Ruy López (1530?-1580?), maybe?
Javert
I'd think the Crafty source could would be far far better than
Getclub, though it is for personal use only. I'd think the chess
source code that came with Borland Pascal 7 would also probably be
better.
Why_am_I_Here
There are quite a few public domain chess programs around. Something to
work with and some of them are probably at least as good as GetClub....
> source code that came with Borland Pascal 7 would also probably be
> better.
I have that... (Well, actually mine is from Pascal 6.) Packed away in
the basement somewhere.
I also have the C version.
And the Zorland Chess program. (Zorland later changed their name to
Zortech.) This was a bitboard program that was very unsuited to the
8088 systems of the time. (It also used CGA graphics.)
I'd also suggest something like Ken Thompson's old (and public domain)
T.Belle program from the 70's.
Or John Stanback's public domain program from the 80's. (A little
buggy, though.)
Or Perhaps Sargon 1. It's in Z-80, but surely somebody could port it to
C? (I have to admit, I would really like to see a C version of Sargon.
I really wonder what it could do on modern systems. It'd be no match
against any modern program, since it was fixed depth search and a static
exchange evaluator, but it'd be fun to play with anyway.)
Or perhaps Chess 0.5 from 1978. It's in Pascal. It's a classic
bitboard approach to chess. Rather limited though.
Still, you do bring up a good idea about BP7 Chess. I had forgotten it.
I wonder what the legal status is? Borland abandoned most of that
stuff from back then. Including their TurboVision text interface. (They
used to make TV's source freely available on their ftp site years ago.
So I think TV was probably considered public domain or at least
copyrighted freeware.)
I seriously doubt they'd care if anybody used BP Chess for
non-commerical use. However converting BP Chess' interface to something
more modern might take a little work.