Spacious_Mind wrote:I have played some 5 ply games
If you mean depth=5, no wonder Rybka is ahead,
since the equivalent depth for the Mystery engines would be 8 or 9, and yet they are constrained to 5 !!!!
No Gerold, the results are pasted from Arena and the score is after the name. Mystery A has 11 from 20 points.
rgds
Nick
Ok i didn't look close enough.
So Playinig at 5 ply Rybka wins a lot more.
Test it without changing any setting and game in 10 mins.
This may give a diff score.
Spacious_Mind wrote:I have played some 5 ply games
If you mean depth=5, no wonder Rybka is ahead,
since the equivalent depth for the Mystery engines would be 8 or 9, and yet they are constrained to 5 !!!!
No Gerold, the results are pasted from Arena and the score is after the name. Mystery A has 11 from 20 points.
rgds
Nick
Ok i didn't look close enough.
So Playinig at 5 ply Rybka wins a lot more.
Test it without changing any setting and game in 10 mins.
This may give a diff score.
Best.
Gerold.
I already had, so I know it's strength. The point I am making is that it is hard to find a resemblance to Rybka. It is stronger in normal games, but even on the lowest end it behaves differently. The programmers must be amazingly good con artists to make these two programs behave so differently at all ends of the spectrum.
regards
Nick
ps.... Rybka irritates me because it doesn't follow rules
Spacious_Mind wrote:I have played some 5 ply games
If you mean depth=5, no wonder Rybka is ahead,
since the equivalent depth for the Mystery engines would be 8 or 9, and yet they are constrained to 5 !!!!
No Gerold, the results are pasted from Arena and the score is after the name. Mystery A has 11 from 20 points.
rgds
Nick
Ok i didn't look close enough.
So Playinig at 5 ply Rybka wins a lot more.
Test it without changing any setting and game in 10 mins.
This may give a diff score.
Best.
Gerold.
I already had, so I know it's strength. The point I am making is that it is hard to find a resemblance to Rybka. It is stronger in normal games, but even on the lowest end it behaves differently. The programmers must be amazingly good con artists to make these two programs behave so differently at all ends of the spectrum.
regards
Nick
ps.... Rybka irritates me because it doesn't follow rules
Spacious_Mind wrote:
Cool are you saying that Rybka minimum is 8 ply ? Because that would explain the big difference.
btw Mystery A is not the one being discussed in todays topics, that only leaves..........
No it means 5 ply from Rybka equals 8 from other engines. If you use 8 then Rybka is really using 11. Rybka plys are +3 compared to other engines. In fact you could start playing Rybka at -2 check it yourself.
In conclusion you are giving Rybka the advantadge playing fixed plys.
Spacious_Mind wrote:
Cool are you saying that Rybka minimum is 8 ply ? Because that would explain the big difference.
btw Mystery A is not the one being discussed in todays topics, that only leaves..........
No it means 5 ply from Rybka equals 8 from other engines. If you use 8 then Rybka is really using 11. Rybka plys are +3 compared to other engines. In fact you could start playing Rybka at -2 check it yourself.
In conclusion you are giving Rybka the advantadge playing fixed plys.
Thanks, It seems to be the case. I played some 2 ply games with Rybka and it seems to work.
M ANSARI wrote: I guess Wael or someone else that is good at tuning engine parameters can probably figure out a better setting for default R3 and you will find that R3 is actually equal if not slightly stronger at all time controls ... only need to change the time management of R3. R3 time management was I guess designed against non Rybka engines. I will try to see how changing the time management of R3 will affect results against Robbolito.
Larry did 80k games on a cluster even for a smallest change, and you say TM in R3 is not well tuned...
TM in R3 is great, Robbo with R3 TM is 30 ELO stronger then current e4 version .
Moreover, if you just check the code of both (and you should be able to, concerning your relationship with Vas), you will see that Robbo has more aggressive TM and there is no way that R3 on average can have more time in later stages of the games.
M ANSARI wrote: I guess Wael or someone else that is good at tuning engine parameters can probably figure out a better setting for default R3 and you will find that R3 is actually equal if not slightly stronger at all time controls ... only need to change the time management of R3. R3 time management was I guess designed against non Rybka engines. I will try to see how changing the time management of R3 will affect results against Robbolito.
Larry did 80k games on a cluster even for a smallest change, and you say TM in R3 is not well tuned...
TM in R3 is great, Robbo with R3 TM is 30 ELO stronger then current e4 version .
Moreover, if you just check the code of both (and you should be able to, concerning your relationship with Vas), you will see that Robbo has more aggressive TM and there is no way that R3 on average can have more time in later stages of the games.
I think when they did thousands of games they tried not to use an identical engine as a sparring partner as that could cause incest and thus could skew results. This was also the case for adding contempt ... I don't think they would have added a contempt factor if they were planning to play against an identical engine. The idea was to design the engine to play extremely strong against non Rybka opponents. If you play 2 identical Rybka's with one of them a 15 contempt you will find that the 15 contempt will do worse. Also I noticed a setting in Robbolito that is very effective against Rybka 3, which is to move immediately on ponder hits. This gets Rybka 3 way behind on time in fast time control games.
There is no doubt IMHO that Robbolito's strength comes from having Rybka evaluation and search code. If you manage to incorporate these 2 things into ANY engine, you will have an extremely strong engine.
M ANSARI wrote:If you play 2 identical Rybka's with one of them a 15 contempt you will find that the 15 contempt will do worse.
I think we have now test results with Rybka contempt set to 0 and no significant difference from those with contempt=15.
M ANSARI wrote: Also I noticed a setting in Robbolito that is very effective against Rybka 3, which is to move immediately on ponder hits. This gets Rybka 3 way behind on time in fast time control games.
Robbolito is stronger even with ponder off.
M ANSARI wrote:There is no doubt IMHO that Robbolito's strength comes from having Rybka evaluation and search code. If you manage to incorporate these 2 things into ANY engine, you will have an extremely strong engine.
But how do you get a stronger one?
I can only guess, but having unencrypted and obfuscated code running protection free would probably make the executable run much quicker. I am not sure what protection Rybka 3 was using to protect its code, but depending on the methods used it would make a difference. This advantage would seem to be more prominent at faster time controls than longer time controls, and that seems to be the case in testing with the two engines. It could be that someone hacked into a computer that had one of the R4 beta's but I really doubt it as the difference between R3 and Ippolit are less than I would expect from a possible R4 clone.
I mean you could play even at -2 plys Rybka just shows -3 plys compared to other engines. Playing fixed plys with Rybka against other non Rybkas is not recommended.
.