Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / How to calculate elo of ones own games...
1 2 Previous Next  
- - By Italian81 (****) Date 2008-05-21 23:26
Is there any kind of software to take my "human" played games (pgn database) and calculate my elo strength?

I am not the next Bobby Fischer, but I would like to know how accurate the Playchess server is....I average about 1800 in normal time controls and about 1500 in Blitz
Parent - - By Psilocybe (**) Date 2008-05-22 00:40
There is a function like that in Fritz GUI. It requires that some of the players in the database have an ELO, and a minimum number of games. For exact requirements read Fritz documentation.
Parent - By Italian81 (****) Date 2008-05-22 01:30
Not very helpful :P
Parent - - By Felix Kling (Gold) Date 2008-05-22 01:47
Your rating should be about 1300-1500 Elo then. Anyway, you can play against a stronger player and ask him what he thinks about your play. Then you get a good evaluation in most cases.
Parent - - By Mark Mason (***) Date 2008-05-22 06:44
Italian,

Here is an online test to assess an estimate of your elo based on solving some test positions:

http://www.chessmaniac.com/ELORating/ELO_Chess_Rating.shtml

Mark
Parent - - By Uly (Gold) Date 2008-05-22 10:28 Edited 2008-05-25 00:54
I got 1525; I don't have a FIDE rating but I think that's pretty close to what it would be.

Now, just for fun, I let Rybka 2.3.2a 32bit examine each position for 30 seconds (Here are the FENs if anyone is interested):

r1b3k1/6p1/P1n1pr1p/q1p5/1b1P4/2N2N2/PP1QBPPP/R3K2R b KQ - 0 1


2nq1nk1/5p1p/4p1pQ/pb1pP1NP/1p1P2P1/1P4N1/P4PB1/6K1 w - - 0 1


8/3r2p1/pp1Bp1p1/1kP5/1n2K3/6R1/1P3P2/8 w - - 0 1


8/4kb1p/2p3pP/1pP1P1P1/1P3K2/1B6/8/8 w - - 0 1


b1R2nk1/5ppp/1p3n2/5N2/1b2p3/1P2BP2/q3BQPP/6K1 w - - 0 1


3rr1k1/pp3pbp/2bp1np1/q3p1B1/2B1P3/2N4P/PPPQ1PP1/3RR1K1 w - - 0 1


r1b1qrk1/1ppn1pb1/p2p1npp/3Pp3/2P1P2B/2N5/PP1NBPPP/R2Q1RK1 b - - 0 1


2R1r3/5k2/pBP1n2p/6p1/8/5P1P/2P3P1/7K w - - 0 1


2r2rk1/1p1R1pp1/p3p2p/8/4B3/3QB1P1/q1P3KP/8 w - - 0 1


r1bq1rk1/p4ppp/1pnp1n2/2p5/2PPpP2/1NP1P3/P3B1PP/R1BQ1RK1 b - - 0 1


Only one of my moves matched Rybka's. And the Rybka's rating is...

2110!
Parent - - By M ANSARI (*****) Date 2008-05-22 11:24
Hehehe ... I got 1840 but I thought the examples were terrible .... I was just about to do what you did and was very curious to see what Rybka would play.  I am not surprised at all that it only got 2110 ... which shows you just how inaccurate this whole thing is.  For example in this position I played Be4

2nq1nk1/5p1p/4p1pQ/pb1pP1NP/1p1P2P1/1P4N1/P4PB1/6K1 w - - 0 1


and I thought that this would for sure be the strongest since it brings the only inactive piece in play and taking it is suicide because if the Bishop is taken then Knight takes with tempo that looks totally devastating with either mate or queen loss because a check on f6 looks like almost mate.  But it seems that this site thinks that Nxh7 is stronger.  When I put Nxh7 I get a higher rating than Be4 ???  Although it seems that Nxh7 wins ... a strong human would much more likely go for Be4 because it brings another piece into attack with tempo ... and Rybka agrees since it also chooses Be4. To get penalized for Be4 is just silly and shows this is not accurate.  I am curious to see what they thought was strongest in all these positions.
Parent - - By M ANSARI (*****) Date 2008-05-22 12:35 Edited 2008-05-22 12:40
r1b3k1/4p1p1/P1n2r1p/q1p5/1b1P4/2N2N2/PP1QBPPP/R3K2R b KQ - 0 1


In this position I chose Bxa6 ... Rybka also thinks it is playable since it wins a pawn and could possibly impede white castling.  BUT .... apparently you get the most points for playing Rxf3 ??  I ofcourse thought about that but quickly saw that white simply takes with bishop.  Rxf3 only works if white takes with pawn .... but white can simply take with bishop and Knight is hanging and rook is pinned .... what they think is best is actually losing !!!  I am now wondering what the rating of the player who used these examples has?
Parent - - By dareapa (**) Date 2008-05-22 14:36
my rating is 1620 ....i'm wondering how far off am i from my true rating....is it higher or lower?
Parent - - By Mark Mason (***) Date 2008-05-22 15:30
As I was the one who posted the link to the elo calculator I thought I ought to give it a try.  I responded to the 10 positions spending about 2 minutes on each and scored elo 1730. 

I don't play over the board chess regularly these days so have no official rating to compare this with but my blitz rating on the Playchess server is around 1620. Therefore this seems to be fairly close to the mark, as I would expect to be stronger at slower time controls than at blitz.

M Ansari - I was interested how you knew which move was highest rated by the site's authors as I couldn't find any information on this - it just gives an overall elo at the end with no breakdown, unless I'm looking in the wrong place ?

Mark
Parent - By Italian81 (****) Date 2008-05-22 18:01
I did the test spending about a minute on each diagram and got 1695
Parent - - By M ANSARI (*****) Date 2008-05-22 18:42
Well I left all the setting exactly as it was when I got the 1840 ... then I changed one of moves trying to figure out what the best move was, then I recalculated.  By doing that I found out what I did in the above post.  Try it and you will see.  I usually am quite good at solving tactical positions and I was expecting to score a lot higher.  Still if Rybka only managed a little better then it should be OK :P
Parent - - By Lukas Cimiotti (Bronze) Date 2008-05-22 21:44
Btw. if you put in the worst moves you can find (queen sacs...) you get 1000 Elo ;)
Parent - By Uly (Gold) Date 2008-05-22 22:35

> Btw. if you put in the worst moves you can find (queen sacs...) you get 1000 Elo ;-)


Hm, so the rating for a player that gives away the queen when possible is 1000 and Rybka is 2100, interesting ;)
Parent - - By Vempele (Silver) Date 2008-05-23 09:06
I had no idea random move generators were so strong. :)
Parent - By Uly (Gold) Date 2008-05-23 09:29

> I had no idea random move generators were so strong. :-)


Well, this wasn't a random move generator, this was a worst move generator ;)
Parent - - By Vempele (Silver) Date 2008-05-24 14:21 Edited 2008-05-24 14:27

> what they think is best is actually losing !!!


Black gets two minor pieces and a bishop pair for a rook and a pawn. If you don't know the elementary piece values, you're clearly not qualified to disagree with the solution. :)

Edit: the fact that you got the position wrong may have had something to do with it. The pawn on e7 should be on e6!
Parent - By Uly (Gold) Date 2008-05-25 00:57

> Edit: the fact that you got the position wrong may have had something to do with it. The pawn on e7 should be on e6!


Well, oops! That was my fault

I manually rebuilt those positions on Winboard and pasted them here; I had them side by side and could not spot that the pawn was on the wrong position. I apologize for the confusion that my mistake could have caused (I've fixed the position now.)
Parent - - By Sesse (****) Date 2008-05-23 19:41
If you look at the JavaScript source, it's trivial to find what they consider the strongest move (which would average around 2500-2600):

Position 1: Rxf3
Position 2: Bf4
Position 3: c6
Position 4: e6
Position 5: Bc5
Position 6: Bxf6
Position 7: Nh7
Position 8: Bd8
Position 9: Bd4
Position 10: Qd7

It's pretty simplistic: They have a few rated moves for each position, and any others get 1000. Then they average the score over all ten positions.

I got 1350 spending twenty seconds or so per position, which I think is slightly too high. :-)

/* Steinar */
Parent - - By turbojuice1122 (Gold) Date 2008-05-24 05:42
The ones I missed were positions 2, 6, and 10.  In number 2, I had this crazy idea about sacrificing the bishop on d5 so that I could push e6, even though I couldn't see anything good at the end--I just couldn't see anything with other moves, either.  On position 6, I wanted to put a knight on d5.  With position 10, I wanted to play ...Bf5, and I really don't see how ...Qd7 is better than that--white is threatening to play f5 and then launching a nasty kingside attack.
Parent - - By M ANSARI (*****) Date 2008-05-24 07:09 Edited 2008-05-24 07:40
On position 2 it takes about 3 seconds to see that taking the bishop is impossible due to Knight takes and check on f6 so in reality there is no bishop sac.  Therefore if you can add a piece to a direct attack with impunity usually that is the best move and it takes a little more calculation to see that black is either getting mated or losing a lot of material.  On position 1 ... if you played Rxf3 thinking you have discovered a nice tactical shot... well that is simply wrong and is losing immediately since white can simply take with bishop ... and then what ??.  On position 10 ... what a silly position ... there are so many possibilities that you really can't give a move as being better than others, it is more a matter of what plan to take ... personally I think black has to avoid at all costs d4 ... because with d4 then a4 is almost a free move since it comes with tempo and white is totally passive and probably lost since black totally controls the center and white basically has no play for any of his pieces ... I would think immediately exchanging in the middle and trying to activate his pieces would be the best plan otherwise white is dead.  There are ofcourse some positions which are correct but really most are wrong .... besides ... anything that gives Rybka a 2110 rating is just plain and simple wrong.

To be honest I think that the best way to estimate your ELO rating is to play on Playchess ... see what your rating stabilizes at ... and then subtract 200 or 300 ELO points and you should be about right.
Parent - By turbojuice1122 (Gold) Date 2008-05-24 13:52
In position 1, I would say that ...Rxf3 is clearly the correct move--I saw it instantly, it is the general theoretical thematic move in these types of positions, and it's really not all that much of a sacrifice--black takes with the pawn on d4 with the next move and wins back material.  I still haven't checked--does Rybka not see that?!  If so, then I would agree that its positional understanding is roughly 2100 level.  The idea here is that Rybka's "understanding" of chess is about the same as that of a 2100-rated player, give or take.  I would agree with this, and I think that most other people would, too.
Parent - By Italian81 (****) Date 2008-05-25 01:01

> and then subtract 200 or 300 ELO points and you should be about right.


Yah, if you are a monkey!! hah
Parent - - By M ANSARI (*****) Date 2008-05-24 06:50
On position 2 Bf4 is not possible.  Is that a typo or are the positions shown up here not in order?
Parent - - By turbojuice1122 (Gold) Date 2008-05-24 06:53
I was assuming that it is Be4
Parent - By Sesse (****) Date 2008-05-26 00:52
g2-e4, yes.

(I wonder, will my flag change now?)

/* Steinar */
Parent - By Carl Bicknell (*****) Date 2008-05-24 15:33
I got 1985 - I don't like this test!
Parent - - By turbojuice1122 (Gold) Date 2008-05-24 05:37
I scored 2150 on first attempt--and that was after a very long day and feeling very sleepy.  However, many of these don't require what I would call extreme concentration--it's more a matter of sitting back for about 30 seconds and then suddenly seeing it and then calculating it out and confirming it.

Rybka scoring 2110 isn't all that surprising--since not all of these are tactical shots, and we know that Rybka's initial evaluation of positions is below IM level (and some think far below), it shouldn't be shocking that Rybka would score so "low."  There is, of course, the accuracy issue.
Parent - - By Uly (Gold) Date 2008-05-24 06:11
Turbo, I think you missed some posts up there, that prove that the test is flawed and that some moves given as best, aren't.
Parent - - By turbojuice1122 (Gold) Date 2008-05-24 06:48
I didn't miss some posts.  It's a little flawed, though not critically flawed.  It's not so much a test that predicts your elo performance in a tournament, but more of a test that tries to equate understanding to that of players with such and such elos.  Question: does Rybka find Nh7 in that King's Indian position?  That is the only decent move there, I think, and I also think that many engines will underrate it.
Parent - - By Uly (Gold) Date 2008-05-24 08:00
It does, then it switches to Be4. Here's a Multi-PV 2 on my computer:

2nq1nk1/5p1p/4p1pQ/pb1pP1NP/1p1P2P1/1P4N1/P4PB1/6K1 w - -


[Snip]
10   0:00   +1.52   1.Nxh7 Nxh7 2.hxg6 fxg6 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 a4 (32.108) 46
10   0:00   +1.13   1.hxg6 hxg6 2.Qh8+ Kxh8 3.Nxf7+ Kg7 4.Nxd8 a4 5.Bf1 Na7 6.f4 (37.810) 48
-----
11   0:01   +1.54   1.Nxh7 Nxh7 2.hxg6 fxg6 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 a4 7.Qxd5 Qxd5 (54.356) 47
11   0:01   +1.10   1.hxg6 hxg6 2.Qh8+ Kxh8 3.Nxf7+ Kg7 4.Nxd8 Na7 5.f4 Bd7 6.Kf2 Nb5 (81.246) 51
-----
12   0:02   +1.53   1.Nxh7 Nxh7 2.hxg6 fxg6 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 Qg5 7.Qc8+ Qg8 (115.091) 51
12   0:03   +1.21   1.hxg6 hxg6 2.Qh8+ Kxh8 3.Nxf7+ Kg7 4.Nxd8 Na7 5.f4 Bd7 6.Bf1 Nb5 7.Bxb5 Bxb5 (185.214) 56
-----
13   0:04   +1.26   1.Nxh7 Nxh7 2.hxg6 fxg6 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 Qc7 7.Qh6 Bc6 (242.547) 55
13   0:08   +1.26   1.hxg6 fxg6 2.Nxh7 Nxh7 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 Qc7 7.Qh6 Bc6 (500.286) 58
-----
14   0:48   +1.88   1.Be4 Be8 2.hxg6 fxg6 3.Nh5 Qe7 4.Nf6+ Kh8 5.Bd3 Nd6 6.Nfxh7 Nxh7 7.Nxh7 Qxh7 (2.618.536) 54
14   0:10   +1.32   1.Nxh7 Nxh7 2.hxg6 fxg6 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 Qc7 7.Qh6 Bc6 (572.057) 57
14   0:10   +1.32   1.hxg6 fxg6 2.Nxh7 Nxh7 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 Qc7 7.Qh6 Bc6 (610.516) 58
-----
15   0:57   +2.10   1.Be4 Nd6 2.hxg6 hxg6 3.exd6 dxe4 4.Qh8+ Kxh8 5.Nxf7+ Kg7 6.Nxd8 Bd3 7.Nb7 Kf7 (3.060.234) 54
15   0:59   +1.32   1.Nxh7 Nxh7 2.hxg6 fxg6 3.Qxg6+ Kh8 4.Qxe6 Ne7 5.Nf5 Nxf5 6.gxf5 Qc7 7.Qh6 Bc6 (3.154.732) 54
best move: Bg2-e4 time: 1:04.513 min  n/s: 54.557  nodes: 3.426.596
Parent - - By turbojuice1122 (Gold) Date 2008-05-24 13:56
I was referring to the King's Indian position that is still in the opening, number 7; the position you posted above looks more like a Benoni.
Parent - - By Vempele (Silver) Date 2008-05-24 14:28

> the position you posted above looks more like a Benoni.


Looks French.
Parent - By turbojuice1122 (Gold) Date 2008-05-24 20:34
Yes, it could be an Advance French.
Parent - - By Uly (Gold) Date 2008-05-25 01:11
Sorry about that...

r1b1qrk1/1ppn1pb1/p2p1npp/3Pp3/2P1P2B/2N5/PP1NBPPP/R2Q1RK1 b - -


On this postion, Nh7 is not even in the top 4:

[Snip]
-----
10   0:00   +0.19   1...Qe7 2.h3 Rd8 3.Qc2 a5 4.a3 Nb6 5.a4 c6 6.Bg3 Qc7 (38.476) 62
10   0:00   +0.19   1...b6 2.h3 Bb7 3.Qc2 Qe7 4.b4 a5 5.a3 Rfd8 6.Bg3 Rdc8 (38.477) 62
10   0:01   +0.22   1...Rb8 2.b4 Qe7 3.Qc2 b6 4.h3 Bb7 5.a3 Rfe8 6.Bf3 (100.252) 60
10   0:01   +0.22   1...Qd8 2.b4 b6 3.Qc2 Bb7 4.a3 Qe7 5.h3 Rfd8 6.Bf3 (110.257) 58
10   0:01   +0.23   1...a5 2.a3 Nb6 3.f3 c5 4.Qb3 Ra6 5.Nb5 Qe7 (109.059) 59
10   0:00   +0.28   1...Nb6 2.a4 a5 3.f3 Bd7 4.Bf2 Qd8 5.Qb3 (47.974) 56
-----
11   0:01   +0.19   1...Qe7 2.h3 Rd8 3.Qc2 a5 4.a3 Nb6 5.a4 c6 6.Bg3 Qc7 (110.258) 58
11   0:01   +0.19   1...b6 2.h3 Bb7 3.Qc2 Qe7 4.b4 a5 5.a3 Rfd8 6.Bg3 Rdc8 (110.259) 58
11   0:02   +0.22   1...Rb8 2.b4 Qe7 3.Qc2 b6 4.h3 Bb7 5.a3 Rfe8 6.Bf3 (163.006) 62
11   0:02   +0.22   1...Qd8 2.b4 b6 3.Qc2 Bb7 4.a3 Qe7 5.h3 Rfd8 6.Bf3 (163.006) 62
-----
12   0:02   +0.19   1...Qe7 2.h3 Rd8 3.Qc2 a5 4.a3 Nb6 5.a4 c6 6.Bg3 Qc7 (163.007) 61
12   0:02   +0.19   1...b6 2.h3 Bb7 3.Qc2 Qe7 4.b4 a5 5.a3 Rfd8 6.Bg3 Rdc8 (163.008) 61
12   0:05   +0.22   1...Qd8 2.b4 b6 3.Qc2 Bb7 4.a3 Qe7 5.h3 Rfd8 6.Bf3 (312.562) 61
12   0:05   +0.23   1...Rb8 2.b4 Qe7 3.Qc2 b6 4.h3 Bb7 5.a3 Rfe8 6.Bf3 (312.561) 61
-----
13   0:09   +0.19   1...Qe7 2.h3 Rd8 3.Qc2 a5 4.a3 Nb6 5.a4 c6 6.Bg3 Qc7 (488.000) 54
13   0:09   +0.19   1...b6 2.h3 Bb7 3.Qc2 Qe7 4.b4 a5 5.a3 Rfd8 6.Bg3 Rdc8 (488.000) 54
13   0:09   +0.22   1...Qd8 2.b4 b6 3.Qc2 Bb7 4.a3 Qe7 5.h3 Rfd8 6.Bf3 (488.001) 54
13   0:20   +0.23   1...Rb8 2.b4 c5 3.a3 b6 4.Bg3 Bb7 5.h3 Qd8 6.Bf3 (1.100.371) 55
-----
14   1:26   +0.24   1...b6 2.a3 Bb7 3.b4 Rc8 4.Qb3 Qd8 5.f3 Qe7 6.Bd3 c5 (4.254.057) 50
14   1:44   +0.24   1...Qd8 2.b4 a5 3.a3 c5 4.Qb3 Qb6 5.Na4 Qc7 6.Bg3 axb4 (5.192.786) 51
14   1:12   +0.25   1...Qe7 2.a3 Nb6 3.f3 a5 4.Qb3 a4 5.Qb4 Ra6 6.Nb5 Rd8 (3.586.098) 50
14   2:01   +0.25   1...a5 2.a3 Nb6 3.f3 Qe7 4.Qb3 a4 5.Qb4 Ra6 6.Nb5 Rd8 (6.088.644) 51
14   1:52   +0.26   1...Rb8 2.b4 c5 3.a3 b6 4.Bg3 Bb7 5.h3 Qd8 6.Bf3 Qc7 (5.634.080) 51
-----
best move: b7-b6 time: 2:20.251 min  n/s: 50.172  nodes: 6.871.917
Parent - - By turbojuice1122 (Gold) Date 2008-05-25 04:07
That's why Rybka's "chess understanding" is not much higher than 2100 elo level. :-)  If you look at some of those moves, like ...Qd8? and ...Qe7?, in this position, and that they're actually rated higher than ...Nh7! is absolutely astounding.  It shows that computers have a very long way to go in dealing with the opening and early middlegame.
Parent - - By Banned for Life (Gold) Date 2008-05-25 05:55
It seems more likely that the computer is playing with consistent strength in the opening, early middle game, and middle game, but people have an advantage until the early middle game when they are familiar with the relevant theory.
Parent - - By turbojuice1122 (Gold) Date 2008-05-25 15:24
But the opening, middlegame, and endgame have very different conceptual aspects from the programming point of view, and programming for the middlegame is going to be the easiest and most straightforward because more raw calculation goes into that part than any other point in the game--or, I should perhaps say, one gets more reward from raw calculation in that part of the game than at any other point in the game.
Parent - By Banned for Life (Gold) Date 2008-05-25 17:55
Vas has stated on multiple occasions that he is philosophically opposed  to putting in rules just for the opening (for instance to prevent early knight moves to c3 or c6 or to prevent moving bishops in front of pawns. Of course Larry is the expert on the changing values of pieces as a function of the material and characteristics of the position but its not clear that he can't do this as well in the opening or early middle game as he does in the late middle game.

I'm not sure I agree with your statement about getting more reward per unit processing in the middle game either. There are some middle game positions where this is true and others where it isn't, just as there are in the opening and in the end game.

The question is how much worse does an engine look in the end game when we can compare its moves to exhaustive search (tablebases) than it would in another position where it would take a huge amount of time and effort to show that the engine's output was seriously flawed? I'm not convinced that Rybka 2.3.2a's middle game play won't look just as flawed ten years from now when we play it against the best at that time.

Regards,
Alan
Parent - - By Italian81 (****) Date 2008-05-22 17:42
I beat 1800-2000 players consistently.  I think I just need to play more games.....In blitz and bullet matches I am at a severe handicap using my laptop and my finger as the mouse
Parent - - By M ANSARI (*****) Date 2008-05-25 07:35 Edited 2008-05-25 07:45
This server is not very good and the problems are terrible ... anything that gives Rybka a rating of 2100 ELO must be off by quite a bit.  If you want a good server for problems go to this site ... I highly recommend it.  It has chess problems setup in a very nice way and although I don't think you will get a good ELO estimate of your strength you will get a general idea of how you rate against others.  There is a good chance that a lot of the people there cheat by using computer assistance, so don't get discouraged if you score low ... but still it is very useful if you want to train your tactical skills.

http://chess.emrald.net

or

http://chesstempo.com/

The problems are setup in a way which makes it feel like a real game ... you get to see your opponents move and then react.  Ofcourse it is time based as well so you get credit for solving a problem quickly.  It will dramatically improve your tactical awareness and thus probably also improve your overall chess ELO.
Parent - - By SR (****) Date 2008-05-25 08:27
I got elo 2300 white is the elo I had when I stopped playing serious chess.  I think the test probably test chess understanding pretty well even though I fail to understand why Bd4 (that I missed) as well as Qd7  are such strong moves. I cannot see how the test can work for a +2300 player? Are they supposed to find these moves? To me a7-a5
(or maybe Nc6-a5!? or Nc6-e7!?) looks like reasonable moves in Q10. 
Parent - By Italian81 (****) Date 2008-05-25 17:35
Come back and play chess, I need a trainer!!

Ha, seriously though I have never had someone professionally teach me how to play chess and I want to be better!!
Parent - - By turbojuice1122 (Gold) Date 2008-05-25 15:28
anything that gives Rybka a rating of 2100 ELO must be off by quite a bit

But Rybka's chess understanding is on the lower side of the 2000 to 2600 range, and the test was more of a test of chess understanding, so its results for Rybka are fairly consistent, within error bars (which I would guess are on the order of 200 elo points), with what is known about Rybka.  Larry puts Rybka's positional understanding at about 2300 elo, while Vas puts it a bit lower; I recall estimates in the 1900 to 2100 range, though I don't know his current estimate.
Parent - - By M ANSARI (*****) Date 2008-05-25 18:33
That theory is totally false ... because it is not comparing apples with apples.  Humans can look at a position and evaluate relatively accurately ... but the assumption that no calculation is made to make the evaluation is not true ... humans will most certainly go through several quick calculations.  When Rybka is not allowed to to calculate and is supposed to simply look at a position and evaluate it simply with no calculation ... ofcourse it will rate low ... positions need calculation no matter how simple or complicated ... opening game, middle game, end game makes no difference ... calculation is needed.  There are very very few positions where a human might do a better job at evaluating a position ... the reason is simply because the computer can look very deeply into the position while the human is limited to a tunnel vision view.  A good example is that the human is in a totally dark room and looking at objects with a very narrow beam of a flash light .... while the computer has all the lights in the room on.  This has been proven several times by excellent results of computers playing on their own in Freestyle tournaments.  Although it is not allowed anymore ... when it was allowed computers once endowed with a good book and decent hardware will more than hold their own against the very best humans using computer assistance.  And these are not any humans ... these are humans that know very well the weaknesses of engines and have a lot of experience in exploiting those weaknesses.  I think GM's that played Freestyle events quickly learned to swallow their pride and minimize their overruling of computer moves ... those that didn't were quickly eliminated.
Parent - - By turbojuice1122 (Gold) Date 2008-05-25 21:18
We're still talking about minimal calculation even on the part of Rybka.  Rybka's rating at 1 ply must be weaker than 2100.  The critical positions involved here are ones where additional ply accumulation will not help.  In particular, in the position in question where ...Nh7 is really the only move, I would expect something with a "chess understanding" of over 2000 to be able to see that.  Rybka doesn't see that, and I would guess that many other top programs don't see that, either--it's just not a part of the game where they have very much strength.
Parent - - By M ANSARI (*****) Date 2008-05-25 23:21
r1b1qrk1/1ppn1pb1/p2p1npp/3Pp3/2P1P2B/2N5/PP1NBPPP/R2Q1RK1 b - - 0 1


In this position I see many reasonable moves, and although Nh7 might be playable I am not so sure it is the best.  Personally I see that black needs to develop his bishop and although Nh7 with the idea of Nd7 f6 might release the bishop ... so will the simple b6 and then Bb7.  There are other flexible waiting moves also possible that are probably just as strong and are more dependent on the style of the player or how much risk he wants to take ... there is no one definitive best move ... which is why such a test is useless in figuring out the strength of a player.  Qe7 which was considered strongest by Rybka looks bad because it pins the Queen ... but I am sure that Rybka sees the pin as harmless and black still has all options open after this waiting move.  I will guarantee one thing though ... that Rybka will prove that any of its top 10 moves in this position are winning against any of us here ;)
Parent - - By Bill Forster (**) Date 2008-05-25 23:48
In my database strong players always play Nh7 in this position. The queen has gone to e8 to free the knight from the pin. This enables the knight to move out of the way of the f pawn so black can play the thematic f5. It has nothing to do with the Bc8 which is not in need of immediate "development" in a closed position like this. Besides, there is more to development than nominally moving pieces off the back row. The bishop would be stupid after b6 and Bb7 looking at granite. Actually the long term prospects of this piece on the c8-h3 diagonal are very important. Black seeks a massive buildup that often culminates in a triumphant break through sacrifice of the bishop. White often seeks to defang black's attack by landing a knight on b6 and forcing the bishop's exchange. Of course this type of super long range planning is still a weak area for computers. My guess is that eventually chess engines will improve in this area and then they will start to invent the Spanish Opening and the QGD etc. when they play without a book, rather than the strange stuff we see today.
Parent - - By SR (****) Date 2008-05-26 00:54 Edited 2008-05-26 00:59
I think the point is that in this type of position there is basically a race between white and black. White only way of progress is to break through on the queen side and black only way to progress is to drum up play on the king side. If black hesitate the white attack on the queen side become too strong. It is paramount that Black play f7-f5 sooner rather than later. A move like Qe7 completely miss the point, and shows that engines still have a lot to learn.  The engines failure to play moves like Nh7 illustrate why an unassisted engine have no chance against a human+engine in correspondence chess.  After Qe7 I think black is in serious trouble. For a human it should be pretty obvious that Nf6-h7 is the most efficient way to create play on the king side. 
Parent - - By M ANSARI (*****) Date 2008-05-26 07:25
I used to always believe that unassisted engines will have no chance against a human assisted engine ... I don't anymore.  I think although the human can help the computer decide the best move in certain positions ... he will go wrong due to lack of deep calculation or simply error more times than none.  Ofcourse he will fare much better if he actually tests a position on a variation board and waits for an engine agreement that position A is better than position B ... and if he avoids any move that doesn't drop the evaluation substantially.  Not doing so has caused the elimination of many GM's very quickly.

Kasparov always said that computers have their own understanding of chess and that it does not necessarily match with ours ... and is not necessarily more correct than ours.
Up Topic The Rybka Lounge / Computer Chess / How to calculate elo of ones own games...
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill