Houdini 3.0 --> Tuning

[Dr.Wael Deeb]

...

Btw,
There is no doubt that R.Houdart is one of the most talented chess programmers
My congratulations again to the author of Houdini - for his superior work !!
...

Please, don't forget that Houdart went through Robolitto source code forwards and backwards and took everything.

Didn't know that

Thanks for sharing this important fact...........
Dr.D

[Uri Blass]

The SCCT list gives 27.3 elo advantage for Houdini 2.0Kayra2 relative to houdini 2c but considering the number of houdini personalities that they test I believe that the real difference is smaller and with small differences the best personality is the best also because of some luck.
http://www.sedatcanbaz.com/chess/?page_id=127


They have poor results for the weak programs relative to other lists
so I think that their 27.3 elo is worth less than 25 elo in other lists even if I ignore the fact that you can expect the best houdini personality to be lucky.

Please don't mix apple with pear !

Once more i'd like to mention that the current SCCT's conditions are completely different than all available rating lists

See the full details about SCCT diffrences (advantages/disadvantages):
http://www.sedatcanbaz.com/chess/?page_id=147


Btw,
There is no doubt that R.Houdart is one of the most talented chess programmers
My congratulations again to the author of Houdini - for his superior work !!
But anyway, we should not forget to say BIG THANKS to the creators of Houdini Setting Makers - they have done great job too !!!

But however, the reality is reality...
The great performance (by the H2 tuned settings) cannot be taken as due to luck !

Chess is not a Lottery, especially we can't say ''luck'', if the amount of the played results are based on thousands of games
And my feelings say that:
-Without many ideas from the Houdini Setting Makers too, Houdini 3 would not nearly be as strong as it is nowadays !!

One thing more (e.g calculation by Elostat 1.3)
Some Top Houdini 2 settings performed approx.30 Elo better than default:

SCCT Rating List - All Versions

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Houdini 2.0Kayra2 Pro x64 6c   : 3368   17  16  1100    73.9 %   3187   38.4 %
  2 Houdini 2.0t3* Pro x64 6c      : 3363   18  18  1000    74.6 %   3176   37.2 %
  3 Houdini 2.0t3 Pro x64 6c       : 3357   13  13  1800    69.6 %   3213   39.1 %
  4 Houdini 2.0s2 Pro x64 6c       : 3353   18  18  1000    74.2 %   3170   34.4 %
  5 Houdini 2.0Bar2 Pro x64 6c     : 3350   15  15  1100    71.6 %   3190   44.6 %
  6 Houdini 2.0z Pro x64 6c        : 3349   14  14  1600    70.9 %   3195   36.3 %
  7 Houdini 2.0Higgs Pro x64 6c    : 3340   16  16  1050    70.4 %   3189   41.6 %
  8 Houdini 1.5a x64 6c            : 3338   16  16  1100    67.5 %   3211   41.0 %
  9 Houdini 2.0c Pro x64 6c        : 3336   14  14  1500    70.5 %   3185   39.5 %

Best,
Sedat

I do not see the more than 30 elo improvement in the full list

http://www.sedatcanbaz.com/chess/?page_id=127

1 Houdini 2.0Kayra2 Pro x64 6c : 3367.6 813.0 1100 73.9%
8 Houdini 2.0c Pro x64 6c : 3340.3 1058.0 1500 70.5%

I do not understand how houdini2c lost 4 elo in the rating that you post and got only 3336

Note that 3+2 time control that gives 11 minutes per game means that the average game is something like 80 moves and practically the engines use something like 2.75 minutes/40 moves

2.75 minutes/40 moves for 6 cpu with 4 ghz processors is probably slower time control than the 4 minutes/40 moves for 4 cpu with AMD64 X2 4600+ (2.4GHz).

Inspite of it I see bigger differences in the SCCT rating list relative to the
CCRL rating list and my conclusion is that probably 27 elo difference in the list that I give is equivalent to less than 25 elo in other lists.

[Uri Blass]

Note also that I did not claim that there is no improvement from tuning but when I see many personalities that are 10-27 elo better than houdini2c then probably the best of them is simply a lucky one and I do not buy the biggest number as real improvement espacially when I do not see that the best personality played the biggest number of games and errors of less than 10 elo are clearly possible when all the versions have real SCCT rating 3358-3360.

1 Houdini 2.0Kayra2 Pro x64 6c : 3367.6 813.0 1100 73.9%
2 Houdini 2.0t3* Pro x64 6c : 3363.2 746.0 1000 74.6%
3 Houdini 2.0t3 Pro x64 6c : 3361.2 1252.0 1800 69.6%
4 Houdini 2.0z Pro x64 6c : 3355.9 1134.5 1600 70.9%
5 Houdini 2.0s2 Pro x64 6c : 3353.5 742.0 1000 74.2%
6 Houdini 2.0Bar2 Pro x64 6c : 3350.7 787.5 1100 71.6%
8 Houdini 2.0c Pro x64 6c : 3340.3 1058.0 1500 70.5%

[Sedat Canbaz]

I do not see the more than 30 elo improvement in the full list

http://www.sedatcanbaz.com/chess/?page_id=127

1 Houdini 2.0Kayra2 Pro x64 6c : 3367.6 813.0 1100 73.9%
8 Houdini 2.0c Pro x64 6c : 3340.3 1058.0 1500 70.5%

I do not understand how houdini2c lost 4 elo in the rating that you post and got only 3336

Note that 3+2 time control that gives 11 minutes per game means that the average game is something like 80 moves and practically the engines use something like 2.75 minutes/40 moves

2.75 minutes/40 moves for 6 cpu with 4 ghz processors is probably slower time control than the 4 minutes/40 moves for 4 cpu with AMD64 X2 4600+ (2.4GHz).

Inspite of it I see bigger differences in the SCCT rating list relative to the
CCRL rating list and my conclusion is that probably 27 elo difference in the list that I give is equivalent to less than 25 elo in other lists.

I think you missed to check the below link...
And as we see, there is at least 30 Elo difference

SCCT Rating List - All Versions (Calculation by Elostat 1.3)
http://www.sedatcanbaz.com/chess/files/ ... lostat.txt

I repeat again:
-SCCT time control is not 2.75 minutes/40 moves,it is:
-3 Minutes/Game + 2 Seconds/Move

Note also that overall move game length of the total SCCT database equals to approx.70 moves (not 80 moves)



Plus,depending on chess positions (with the current time control),the engines play differently time per move:


For more details:
http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=25629
http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=25590

I do not understand how houdini2c lost 4 elo in the rating that you post and got only 3336

I think the best answer you can get from the authors of Houdini/Calculation tools

Btw,
I have much more interest and i would prefer to relay on game results,which are based on:
-Fastest decent hardwares (testing the engines with maximum cores performance)
-Strongest modern openings(testing the engines with openings up to 8-12 moves)
-Games (available for downloading with annotations) played with popular Blitz/Slow time controls
vs...


Regards,
Sedat

[Sedat Canbaz]

Note also that I did not claim that there is no improvement from tuning but when I see many personalities that are 10-27 elo better than houdini2c then probably the best of them is simply a lucky one and I do not buy the biggest number as real improvement espacially when I do not see that the best personality played the biggest number of games and errors of less than 10 elo are clearly possible when all the versions have real SCCT rating 3358-3360.

1 Houdini 2.0Kayra2 Pro x64 6c : 3367.6 813.0 1100 73.9%
2 Houdini 2.0t3* Pro x64 6c : 3363.2 746.0 1000 74.6%
3 Houdini 2.0t3 Pro x64 6c : 3361.2 1252.0 1800 69.6%
4 Houdini 2.0z Pro x64 6c : 3355.9 1134.5 1600 70.9%
5 Houdini 2.0s2 Pro x64 6c : 3353.5 742.0 1000 74.2%
6 Houdini 2.0Bar2 Pro x64 6c : 3350.7 787.5 1100 71.6%
8 Houdini 2.0c Pro x64 6c : 3340.3 1058.0 1500 70.5%

One thing more (for those,who has some suspicious about the SCCT Elo error bar margins)

Both tested t3/t3* settings are completely same,exception t3* is tested with slightly different shorter book (Perfect 2012c)

And the current SCCT results (1000 games per player) is quite enough data about proving that the tuned Top settings are performed better than default :its not due to Elo bar error margins

Code: Select all

RANK PARTICIPANT                     : RATING    POINTS   GAMES    (%)
   2 Houdini 2.0t3* Pro x64 6c       : 3363.2     746.0    1000   74.6%
   3 Houdini 2.0t3 Pro x64 6c        : 3361.2    1252.0    1800   69.6%

Best,
Sedat

[Tennison]

Norman,

I'm just banned from chess2U forum just because I inform the users that Houdiconfig doesn't work with Houdini 3.0 !

It's true.

I test it with absolutely idiot params and it's equal results ... not logical !!

I'm very disappointed but R.H. is right with this : houdiconfig works correctly with 1.5 and 2.0 but not with 3.0

I'll ask why you've been banned. Can't surely be just because of this?

Your message on Chess2U has just been removed !!!

I've gained permisssion to post Robert's message about the configuration file not working with Houdini 3, so have done so.

Ben - you must been banned for a reason other than this. You'd need to ask why if you've not already been told.

Graham.

Hi Graham,

It seems that the R.H.'s account on chess2U is banned too :

http://www.chess2u.com/t6426p45-houdini-3-tuning

And it's why I think maybe "Victor" is thinking I'm R.H. when I posted the warning about Houdiconfig.

But I can confirm I'm not R.H. ! lol

What I disliked is that "Victor" don't even respond to me to explain why he banned me !

[Houdini]

I can confirm that I've been banned as of today.
ROFL.

[Graham Banks]

I can confirm that I've been banned as of today.
ROFL.

Why? Message me if you don't want to say in here.

[ZirconiumX]

I can confirm that I've been banned as of today.
ROFL.

Why? Message me if you don't want to say in here.

He's probably been banned because he's Robert Houdart.

I have a feeling that if Houdini was open-source Robert would be hugely admired, rather than hugely hated.

Matthew:out

[Graham Banks]

I can confirm that I've been banned as of today.
ROFL.

Why? Message me if you don't want to say in here.

He's probably been banned because he's Robert Houdart.

I have a feeling that if Houdini was open-source Robert would be hugely admired, rather than hugely hated.

Matthew:out

Seems to be that whoever is top dog is always going to be the subject of derision/hate/criticism, etc from many quarters. History repeating.
In NZ, we call it the tall poppy syndrome.