Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / TUNED HOUDINI-report this post astroboy=stonehenge=Houdart
1 2 3 Previous Next  
- - By Barnard (Bronze) Date 2012-05-27 22:16 Edited 2012-05-28 02:15
here the values of Houdini 1.5a:

5D717974912A20202020202020202020202020202020202020202020202020202020202020

here are the values of Houdini 2:

5D717974702A2020202020202020202020202020202020202020202020201E2328161C1E20

values of queen is 995 at Houdini 1.5,and 962 at Houdini 2

to allow people to edit that internal values of Houdini,download a software called HoudiniConfig

i will tell to everybody interested,in PM (and only in PM),how and where download HoudiniConfig and how to use it

NOW LETS SEE IF ME,NOT MAKING ANY SINGLE ILLEGAL 'WORK' (SINCE I NEVER PUBLICED SOURCE CODE) ANOTHER MOD IS SO 'SMART' TO SERVE TO THAT ***** (yes,i edited myself :smile: ) AND EDIT THE POST TO DELETE DATA THAT ISNT ILLEGAL TO MAKE PUBLIC

the freedom of this forum is at risk with that ''guy'' working/annoying us here,and the worst part,mods continue protecting him!!none data was deleted when me/others atacked vas,none code was deleted when someone made public rybka code,and now,you are serving to that ''guy''????what happens with you mods???

appart from that,my latest tuning with Houdini...as soon as it will be tested,i will make public the values;so people will be able to have an improved Houdini WITHOUT BUYING HOUDINI 3

here we are Houdart...next time,report my post and you will be able to see how much stubborn i am when someone annoy me
Parent - - By Barnard (Bronze) Date 2012-05-28 03:16
end of the 101 games test

the final result shows a +59 elo points for the tuned version,and a probability of 95% of being the rersults between +3 and +122 elo points
Parent - - By NATIONAL12 (Gold) Date 2012-05-28 06:00 Edited 2012-05-28 06:02
Hi Salva,all you are doing is fine tuning H 1.5,this is all Robert is doing between 1.5 and 2. and when he announces H3 further work will have been done over maybe 100,000 games.None of these results will stand up to 40/40 testing never mind corr chess.He will sell many 1000 units and make lots of $ conning everyone with a derivative engine with a fast exe.It is not possible for him to improve Houdini as he is only a computer expert and not a top chess player.

I firmly believe this,top chess programers do not appear out of noware.Vas whatever you think of him is a OTB IM,Larry is a GM and even Vas wife is an IM.I call it the Harvey syndrome whereby a not so strong player OTB can get to final in World Corr Championship with a computer to help him.:smile:

As you know i tested Houdini at 40/40 on 8 core at 4.2 and found only +31 ELO over Rybka and i believe this was solely down to faster exe.There were certain games where Houdine lost with both black and white as it did not understand position.

All this testing of mine is available on computer chess thread,like my testing on one core at 40/40 of K4-H2 where Komodo came out slightly on top.

regards Paul.
Parent - - By NATIONAL12 (Gold) Date 2012-05-28 06:11
further to above this testing of mine took approx 10-11 days with computer running flat out to detriment of my electric bill.:smile:
Parent - By Barnard (Bronze) Date 2012-05-29 23:25
10-11 days with the big computer 24 h. full time?how much (appro) in electricity bill it cost?
Parent - By NATIONAL12 (Gold) Date 2012-05-28 06:22
I asked Vas a while back if exe would be faster in R5,he said no.There must be technical reasons for this as i am not a programmer nor a computer expert i cannot comment.
Parent - By Barnard (Bronze) Date 2012-05-29 23:23
Hi Paul

i agree with you,Houdini 2 is only a Houdini 1.5 well tuned,and that improvement will only reflect on blitz games

and i agree that he isnt a good player,and he wont be able to improve the engine soo much;from that point of view,and as you stated,have too much future Larry or vas

and i know that with long time controls,the elo difference tends to dissapear
Parent - By tano-urayoan (****) Date 2012-05-30 03:42

> It is not possible for him to improve Houdini as he is only a computer expert and not a top chess player.


By this logic Critter or Stockfish will never improve.
Parent - - By Dr.Wael Deeb (***) Date 2012-05-28 08:53
Great work Bernard and I fully support anyone who is opposite to one of the most obnoxious programmers known to the computer chess history,namely Robert H.

I was too optimistic when I predicted that he'll lose his number one spot at the rating list at the end of this year....

It seems that this will happen much much earlier :grin:
Happy to see Houdini sinks in the rating lists regards,
Dr.D
Parent - By Barnard (Bronze) Date 2012-05-29 23:26
thanks :smile:

i honestly think that when Larry and Don will make the next version,it will surpass Houdini
Parent - - By Rebel (****) Date 2012-05-30 12:23

>end of the 101 games test


> the final result shows a +59 elo points for the tuned version,and a probability of 95% of being the rersults between +3 and +122 elo points


View the picture at: http://www.top-5000.nl/eval_change.jpg

It's extracted from an eng-eng pgn match of 7000 games and its progress is divided in steps of 100 games. As you can see after 100 games the eval change performs bad (46%) then it starts to climb, 49.25% after 200 games, after 3500 games the 2 engines seems to settle on 52%, the whole table:

46.00 49.25 51.50 50.75 50.40 51.33 51.21 50.37 50.56 50.25
50.45 50.37 50.77 51.39 50.87 51.00 51.35 51.50 51.63 51.58
51.93 51.91 52.07 51.54 51.80 52.12 52.17 52.27 52.36 52.38
52.77 52.63 52.77 52.62 52.37 52.57 52.55 52.33 52.36 52.44
52.52 52.20 52.17 52.08 52.18 52.17 52.11 51.96 52.12 52.01
52.01 51.97 52.04 52.02 51.97 52.00 52.06 51.90 51.94 52.01
52.02 52.08 52.07 52.19 52.22 52.22 52.22 52.32 52.40 52.39

What may start as an improvement in the end isn't necessarily an improvement after all and vice versa as in this specific example.
Parent - By Barnard (Bronze) Date 2012-05-30 14:31
Hi Ed

you are totally right,but i cant run longer match engine-engine;i have just one computer,and if im playing the WBCCC2,for example,i must stop making experiments...

i hold my LSS games now because the next round still isnt playing,but when next round will start,i will must stop the currently match;now im playing a 1000 games match

if i only have 2 computers,i can use one computer to play and the other to do the match,being able to play,lets say,5000 games,but i havent that computer
Parent - - By Vempele (Silver) Date 2012-05-30 12:45

> end of the 101 games test


Invalid. Unless you were using BayesElo, you'd expect to get a 33 point difference between identical engines just by defining Engine A to be always White.
Parent - - By Barnard (Bronze) Date 2012-05-30 14:34
Hi vempele

im playing a match between that 2 engines,and each round,engines changes the colour (one with white pieces,and the other game with black pieces...and apart from that,each opening is played by both engines with both colours,white and black,to avoid if it is a bad line opening,that the same engine will play that opening with white and black pieces
Parent - - By Vempele (Silver) Date 2012-05-30 15:11
So how can the test end at 101 games? The number of games should be even.
Parent - By Barnard (Bronze) Date 2012-05-30 18:55
because i wanted to take a pic about all the information relative to the test,running it under chessbase GUI (meaning elo gain,% of chances to be elo in an interval,etc...),and when chessbase GUI ends a match,it stop sending that information

so i decided run a 101 games match,so i will be able to take a pic of that at the game 100,one game before the match ends,but i had a call,and i never was able to do that
Parent - - By RFK (Gold) Date 2012-05-28 06:04 Edited 2012-05-28 06:10
Sal...I think this can be regarded as Cloning! In effect you are developing different strains of Houdini's. Not unlike the Cloning of...say Cannabis! It's there in what you are doing

Here look at this-

Cloning Cannabis Plants

Growing from marijuana clones means that you will harvest more bud sooner. Many people who start a garden these days know someone who will give them a few marijuana clones for the simple pleasure of sharing a good thing. Some people sell marijuana clones, often growing an exclusively vegetative garden without ever flowering.

These clonemeisters keep a variety of select mothers from high-quality strains, and grow them on a continuous eighteen hour light cycle. They can make good money selling trays of cuttings, either rooted or simply fresh cut. Rooted marijuana clones are more expensive than fresh cuttings.


Robert Houdart is like a Clonemeister of sorts. :grin:
Parent - - By Labyrinth (*****) Date 2012-05-28 09:07
Rybka G-13!
Parent - - By Chaotic Chess (****) Date 2012-05-28 09:25
Houdini developer should start hiring with the likes of Larry K to optimize the evals function of Houdini. This will further boost the performance of this engine aka H3.
Parent - - By Labyrinth (*****) Date 2012-05-28 10:48
He doesn't get along with Larry :-/
Parent - - By Dr.Wael Deeb (***) Date 2012-05-28 11:26
Actually no one gets along with him :cool:
Dr.D
Parent - - By Labyrinth (*****) Date 2012-05-28 11:57
Who? Houdart or Larry?
Parent - - By Dr.Wael Deeb (***) Date 2012-05-28 13:34
Houdart of course :grin:
Parent - By RFK (Gold) Date 2012-05-28 14:34 Edited 2012-05-28 14:43
I still hypothesis that RH could possibly be one of the original developers, who after having a falling out- decided upon a " Ippolilt sanction".  The kill weapon being the development and implementation of the deadliest Assassin Chess engine-Houdini and no one still knows who done it.  ( Whatayou!  Think? Nobody's buying into my idea here?! WTF!  I think its a seller! I think I'll publish under the pseudonym Chesster A. Benchmark! )

Robert Houdart the 007 of computer chess- in

The Ippolit Sanction!
Parent - By Barnard (Bronze) Date 2012-05-29 23:27
it isnt cloning an engine,i just changed 1 letter to make able the GUI to load the engine,and just for that cause
Parent - - By cyberbrain (**) Date 2012-05-28 17:04
and you only found houdiconfig now. I have been using it for months now.
Downloaded it here: http://www.chessgui.com/chess/HoudiConfig.zip
Parent - - By RFK (Gold) Date 2012-05-28 17:10
He knows about that link!
Parent - By Barnard (Bronze) Date 2012-05-29 23:34
:lol:
Parent - By Barnard (Bronze) Date 2012-05-29 23:33
yes,i only noticed about that program just 1-2 weeks ago...

but tell me,where are your values?can you post pics about the match? :wink:
Parent - By Barnard (Bronze) Date 2012-06-03 15:45
since thw next round of WBCCC 2 is delayed until saaturday,im gonna start a 1000 games match between the tuned version of Houdini,and critter 1.4

p.s.if the next round of WBCCC will start before saturday,and the match wont be finished,i will must stop the match.i will upload screen captures of the results at about 250,500 and 750 games (meaning an aproximation about that numbers,i dont know when i will be at home or not)
Parent - By Barnard (Bronze) Date 2012-06-06 01:56 Edited 2012-06-06 21:08
FOR EVERY PEOPLE THAT READS THAT THREAD:

THE VALUES I GAVE WITH THE NAME HOUDINI 2,ARE THE ''OFICIAL'' VALUES OF THE ENGINE HOUDINI 2.0c,NOT THE VALUES OF THE TUNED VERSION HOUDINI 2.0z

im telling that because i saw in a lot of forums,that people are copy/pasting those values (Houdini 2),and they are saying are the values of the tuned version Houdini2.0z,and that is false

so,for the people that read that post and are posting those values in other forums:

THOSE ARENT THE VALUES OF HOUDINI 2.0Z

EDITADO:i will translate it inot Spanish down

como veo que en foros de habla hispana tambien estan dando estos valores,y solo pueden haberlos sacado de este thread,y por si algunos de ellos no saben Ingles,voy a explicar lo mismo que en Ingles,pero en Español:

PARA TODA LE GENTE QUE LEA ESTE THREAD:

LOS VALORES QUE DI CON EL NOMBRE HOUDINI 2,SON LOS VALORES ''OFICIALES'' DEL ENGINE HOUDINI 2.0c,NO LOS VALORES DE LA VERSION TUNEADA HOUDINI 2.0z

estoy diciendo esto porque vi en muchos foros,que la gente esta copiando esto valores (Houdini 2),y estan diciendo que son los valores de Houdini 2.0z,y esto es falso

asi,para la gente que lea este thread y postee los valores en otros foros:

ESTOS NO SON LOS VALORES DE HOUDINI 2.0z
Parent - - By Barnard (Bronze) Date 2012-06-08 21:11 Edited 2012-06-09 20:27
Hi

here a few more results

a test Houdini2.0z vs Komodo4 (both running on 1 thread),and Houdini 2.0z vs Stockfish

p.s.i will post the result of the match against Stockfish as soon as the match will reach the 250 games (now is 234 games),and at 500,750 and the final result

p.s.2 i cant provide the games about the match houdini2.0z against Komodo4,if the person that is running the match,doesnt give it to me

here the partial result of the 1000 games match Houdini2.0z vs Stokfish 2.2.2,played 540 games:
Parent - By Barnard (Bronze) Date 2012-06-10 22:57
end of the match,1000 games,Houdini2.0z vs Komodo 4 (both playing in 1 core)

a + 84 elo points for Houdini2.0z
Parent - - By Barnard (Bronze) Date 2012-06-12 22:29
final match Houdini2.0z-Stockfish2.2.2 1000 games
Parent - - By Banned for Life (Gold) Date 2012-06-12 23:59
9-9-9? Is this a subliminal plug for Herman Cain? Why not play an even 1000? :lol:
Parent - By Barnard (Bronze) Date 2012-06-13 00:00
hahahaha :smile:

with the GUI i was only able to set 999 games,not 1000...that is why the test end at 999 games
Parent - - By Barnard (Bronze) Date 2012-06-13 23:28
really funny;people asking at Houdart's facebook page what is Houdini2.0z... :lol:

but im pretty sure he wont answer to that person :lol:
Parent - By Barnard (Bronze) Date 2012-06-13 23:46
and here my answer at Houdart facebook page...i was very polite :smile:

p.s.people trying to know my name based on that account,will fail:is a second account that i never use and havent my name or any personal data
- - By rocket (****) Date 2012-05-28 22:48
Well i'ts old news that you increase elo by lowering the queen value. I am curious as to wheter mr Houdart has indeed taken up this advice with H3 or will he be stubbornly persistent on this point?

The rybka material value is actually contrary to popular belief probably more optimal than houdini .

There is a classic sacrifice of a Bishop and knight for two pawns and a rook in the sicilian positions were rybka gets  strong passed pawns on the queenside- in combination with the material  and is winning in the long run, while Houdini does'nt get it. I suspect Houdini dismissed such sacrifices because of restrictive and too rigid material value.

However these strong spots for rybka are overshadowed by superior king safety and tactical search in Houdini, the latter being outstanding while Rybka 4 looks incredibly slow tactically in my tests, in fact inferior to R3.
Parent - - By free Willy (*) Date 2012-05-29 14:15
Pohl4 setting.As there is nothing better!
Parent - - By Barnard (Bronze) Date 2012-05-29 23:19
im not gonna to give you the values;you are registered 2 days ago,and just made one post,and just for that thread...maybe im an ignorant like Lukas told,but not as much that you are trying to make me

you can perfectly be Houdart using a server from Italy...so you will have the values when all the guys on the forum will have those values,not before
Parent - - By free Willy (*) Date 2012-05-30 14:11
Yes, I'm from Italy. What I want to say, you can change the style of play, but you've no increase in Elo. However test your settings in a few hundred games in long time controls, perhaps in five,ten minutes games. I'm just so interested with your settings because maybe there is a different style. Certainly not  of increase in Elo, because this is a misconception and we have already a number one and from September it will be even better. Too bad for all the envious people who believe with parameter changes you could make something incredible without proving it.
Bernard,with best wishes and show us your skills in parameter changes, perhaps with queen value 860
Parent - - By Barnard (Bronze) Date 2012-05-30 14:38
thanks for yopur answer

i think that the tuned version plays more agressive than the not tuned version,even if it running a 2000 match games (or less) the engine will show not any improvement in elo

apart from me running the match engine-engine,i gave the values to other people,and they are also running games,but privately,to avoid people judge them

the first atempt show a +21 in the 100 games match,and not improvement in longer games,but the engine plays more agressive

p.s.im not using a value for the queen of 860,im user a higher value
Parent - - By free Willy (*) Date 2012-05-30 15:37
Hello, yes I realize it was. Already Pohl4 setting has another very attractive game behavior. I just wanted to say, that can be achieved with pure fine-grained tuning parameters no large increase in Elos. I think Houdini2 is far more than a well tuned Houdini 1.5! Test Houdini 15a with Houdini2 parameters and you will see that this does not bring improvement. From September we will all know more once Houdini3 is available.Queen value 860 was a joke to cheer best regards
Parent - By Barnard (Bronze) Date 2012-05-30 18:57
maybe im wrong,and only a minimal elo can be achieved tuning Houdini,but if it only plays at the same elo,but you changed his style (for example,doing the program more agressive),it will perform better in some positions,and will 'understand' better that positions
Parent - - By Gaмßito (****) Date 2012-05-30 14:40

> Too bad for all the envious people who believe with parameter changes you could make something incredible without proving it.


I know that it's possible to make some relative small improvements by tuning the parameters but it's not easy at all. In the past per example, some people did it with Chessmaster. They found a way to make Chessmaster a bit stronger than default engine.

But I am quite pessimistic regarding the good improvement he allegedly did (+50 Elo better than Houdini 2.0c!) only by tuning the parameters. I really hope to be wrong here, but I doubt it. :smile:

Regards,
Gaмßito.
Parent - - By Barnard (Bronze) Date 2012-05-30 14:55
Hi

i never claimed that i made a +50 elo improvement on the engine;i just made a screen capture of the end of the test,showing the elo advantage that the engine had

apart from that,i stated that there is a 95 % that the engine is in the interval between +3 and 122 elo points;that means that even if on the test the tuned version shows +50,the tuned version can only be +3 elo points improved,and even worst,there is a 5 % that the tuned version play worst than the 3 elo points,making that tuned version equal at the strength of the non-tuner version

i perfectly know that i well optimized engine cant be improved by +50 elo points in just 5 days of work(it need to occur a miracle to allow that imrpovement),it was only a screen capture of the end of the macth

now im running a 1000 games match,and at the game 142,the tuned version shows an improvement of +22 elo points,and maybe later it will decrease,and just show a +5 elo points at 1000 games(or even lower),i dont know,what is why im running the match
Parent - - By Gaмßito (****) Date 2012-05-31 12:07
Well, keep us informed about the results of your 1000 games match.

Keep in mind that Houdini 2.0c is about +25 Elo stronger than Houdini 1.5a in a match head to head, so I can say that even if you gain +20 Elo over Houdini 2.0c, that really will be very impressive. Robert will have to make note of your values to produce a bigger Elo gain in Houdini 3.

Regards,
Gaмßito.
Parent - By Barnard (Bronze) Date 2012-05-31 20:39
Im at middle of the match;played 500 games,it shows an improvement of + 19 ello points
Parent - By Barnard (Bronze) Date 2012-05-31 21:23
i must correct it;was only a ''provisional elo'' that i tought it can be,i recently arrived home,and the result is here (only 3 games remaining to 500 games):
Up Topic The Rybka Lounge / Computer Chess / TUNED HOUDINI-report this post astroboy=stonehenge=Houdart
1 2 3 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill