Topic Rybka Support & Discussion / Rybka Discussion / Evidence against Rybka, the meteoric rise in strength
One of the arguments against Rybka was the meteoric rise in strength. Nobody can suddenly make some huge Elo improvement in a short time.
And yet the Elo improvement continued forward unabated:
Rybka 1.0 64-bit 2823 +16 -16 68.8% -128.6 34.9% 1423
Rybka 1.1 64-bit 2898 +23 -23 74.6% -176.1 35.5% 727 +75
Rybka 1.2 64-bit 2897 +22 -21 74.4% -174.0 32.9% 821 -1
Rybka 2.1 64-bit 4CPU 2985 +38 -37 73.1% -163.0 36.8% 258 +88
Rybka 2.2 64-bit 4CPU 2983 +20 -20 66.1% -109.1 41.1% 815 -2
Rybka 2.3.2a 64-bit 4CPU 3027 +14 -14 65.4% -103.6 42.7% 1707 +44
Rybka 3.0 64-bit 4CPU 3133 +14 -14 72.8% -158.4 39.6% 1795 +106
Rybka 4.0 64-bit 4CPU 3164 +14 -14 60.7% -69.6 48.8% 1633 +31
Rybka 4.1 64-bit 4CPU 3160 +12 -12 56.8% -40.4 56.1% 2065 -4
+337 Elo from 1.0 to 4.1
Keep in mind that it is generally considered a lot harder to improve a strong program than to improve a weak one.
And yet the Elo improvement continued forward unabated:
Rybka 1.0 64-bit 2823 +16 -16 68.8% -128.6 34.9% 1423
Rybka 1.1 64-bit 2898 +23 -23 74.6% -176.1 35.5% 727 +75
Rybka 1.2 64-bit 2897 +22 -21 74.4% -174.0 32.9% 821 -1
Rybka 2.1 64-bit 4CPU 2985 +38 -37 73.1% -163.0 36.8% 258 +88
Rybka 2.2 64-bit 4CPU 2983 +20 -20 66.1% -109.1 41.1% 815 -2
Rybka 2.3.2a 64-bit 4CPU 3027 +14 -14 65.4% -103.6 42.7% 1707 +44
Rybka 3.0 64-bit 4CPU 3133 +14 -14 72.8% -158.4 39.6% 1795 +106
Rybka 4.0 64-bit 4CPU 3164 +14 -14 60.7% -69.6 48.8% 1633 +31
Rybka 4.1 64-bit 4CPU 3160 +12 -12 56.8% -40.4 56.1% 2065 -4
+337 Elo from 1.0 to 4.1
Keep in mind that it is generally considered a lot harder to improve a strong program than to improve a weak one.
This is not really what Bob has been telling us all the time: that Vas copied his way to the top ;-).
hi all!
ChessWar VII
01/04/2005 - 30/09/2005
54 RYBKA 1.6.1 2133 USA
ChessWar VII C 40m/20'
Final standings after round 11:
RYBKA 1.6.1 48th/70 , 5pts/11 bayeselo perf : 2064
ChessWar VIII
01/10/2005 - 10/04/2006
58 RYBKA 1.01 Beta 13 2104 CZE
ChessWar VIII D 40m/20'
Final standings after round 11
RYBKA 1.01 Beta 13 1st/70 , 11pts/11 bayeselo perf : 2961
lol
bye all

http://www.open-aurec.com/chesswar/archive.html
ChessWar VII
01/04/2005 - 30/09/2005
54 RYBKA 1.6.1 2133 USA
ChessWar VII C 40m/20'
Final standings after round 11:
RYBKA 1.6.1 48th/70 , 5pts/11 bayeselo perf : 2064
ChessWar VIII
01/10/2005 - 10/04/2006
58 RYBKA 1.01 Beta 13 2104 CZE
ChessWar VIII D 40m/20'
Final standings after round 11
RYBKA 1.01 Beta 13 1st/70 , 11pts/11 bayeselo perf : 2961
lol
bye all

http://www.open-aurec.com/chesswar/archive.html
It is obvious that the Crafty in 1.6.1 caused a loss of 897 Elo...

not a loss , a gain
chesswar V : rybka 1.5.32 23th/70 perf 2001
after some things of crafty and fruit +900
improved with other sources yes ...
chesswar V : rybka 1.5.32 23th/70 perf 2001
after some things of crafty and fruit +900
improved with other sources yes ...
No. Crafty caused a huge loss in Rybka, not a gain.
The chesswar version of Rybka was a testbed and used Crafty code to fill in the holes. Rybka 1.0 Beta came out a year and a half later, has no traces of Crafty, none whatsoever, and performs 900 Elo better.
If you believe that Rybka 1.0 Beta has anything from Crafty, anything at all, please let us know. Thanks!
The chesswar version of Rybka was a testbed and used Crafty code to fill in the holes. Rybka 1.0 Beta came out a year and a half later, has no traces of Crafty, none whatsoever, and performs 900 Elo better.
If you believe that Rybka 1.0 Beta has anything from Crafty, anything at all, please let us know. Thanks!
> One of the arguments against Rybka was the meteoric rise in strength. Nobody can suddenly make some huge Elo improvement in a short time.
There was another engine that showed significant gains post Fruit 2.1 release and given it actually beat Rybka in a championship match it surprised me that it was not used as a benchmark comparison to show such gains could be made being a true contemporary of Rybka.
Zappa should have been used in the evaluation feature comparison. Zach had the source...
Lots of engines have meteoric rise.
Look at Komodo, since the 'doch' days.
Look at Shredder's rise in the SSDF year after year.
I think that (if anything) failing to have a meteoric rise is a better indication of something strange going on {if you have a strong program}.
If you are smart enough to write a very strong program, a meteoric rise is the rule and not the exception.
If you write a very strong program and then it stagnates, that is an indication of monkey business to me.
Why were you able to create this incredibly strong program, and thereafter you could not improve it?
There can certainly be logical reasons like 'lost interest in chess programming', 'got a real job', etc.
Look at Komodo, since the 'doch' days.
Look at Shredder's rise in the SSDF year after year.
I think that (if anything) failing to have a meteoric rise is a better indication of something strange going on {if you have a strong program}.
If you are smart enough to write a very strong program, a meteoric rise is the rule and not the exception.
If you write a very strong program and then it stagnates, that is an indication of monkey business to me.
Why were you able to create this incredibly strong program, and thereafter you could not improve it?
There can certainly be logical reasons like 'lost interest in chess programming', 'got a real job', etc.
Topic Rybka Support & Discussion / Rybka Discussion / Evidence against Rybka, the meteoric rise in strength
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill