Computer Chess Biggest Liar

General discussion about computer chess...
kranium
Posts: 55
Joined: Mon Aug 02, 2010 10:49 pm
Real Name: Norman Schmidt

Re: Computer Chess Biggest Liar

Post by kranium » Sat Nov 05, 2011 10:35 pm

http://www.talkchess.com/forum/viewtopic.php?t=34394
Osipov Jury wrote: Yes, the PST weights in Houdini are multiplied by 2 as compared to Ippo/Robbo.
It is very easy to check that every weight is identical with the PST values of Robbolito with factor 2.
Osipov Jury wrote:
lkaufman wrote: Could it be that that the piece square table numbers are doubled (as you say), but later the entire eval is divided by 2? This would at least come close to explaining much of the eval, though it doesn't seem to me to be a complete explanation.
Houdini use more complex formula for recalculation of eval for output:

Code: Select all

  if ((true_eval + 899) > 1798 ) { output_eval = 16 * true_eval; }
  else {
    if ( true_eval > -300 ) {
      if ( true_eval > -150 ) {
        if ( true_eval > 150 ) {
          if ( true_eval > 300 ) output_eval = 2 * (9 * true_eval - 900);
          else output_eval = 14 * true_eval - 600;
        }
        else { output_eval = 10 * true_eval; }
      }
      else { output_eval = 14 * true_eval + 600; }
    }
    else { output_eval = 2 * (9 * true_eval + 900); }
  }
  sprintf(output_eval_string, "cp %d", output_eval / 32);
For simple case (true_eval in range [-150;+150]) we can use:

output_eval = (true_eval * 10) / 32;
I have verified with Hex Rays, Jury's post is accurate...
all Houdini's eval values are doubled, and score output is obfuscated with a cascading/scaling 'if else' statement as folows:

PS-
in Robbo 0.09,
28976 = VALUE_MATE - depth_maximum

Here is Robbolito 0.09:

Code: Select all

//----- (00493D20) --------------------------------------------------------
char *__cdecl cp_mate(int value, char *s)
{
  char v3; // [sp+Ch] [bp-C0h]@1

  memset(&v3, -858993460, 0xC0u);
  if ( value <= 28976 )
  {
    if ( value >= -28976 )
      j__sprintf(s, "cp %d");
    else
      j__sprintf(s, "mate %d");
  }
  else
  {
    j__sprintf(s, "mate %d");
  }
  return (char *)j___RTC_CheckEsp();
}
Here is cp_mate from Houdini_w32_1CPU (Houdini 1.0):

Code: Select all

int __fastcall sub_41E7F0(signed int a1)
{
  int result; // eax@2
  signed int v2; // eax@7

  if ( a1 <= 28976 )
  {
    if ( a1 >= -28976 )
    {
      if ( (unsigned int)(a1 + 899) > 0x706 )
      {
        v2 = 16 * a1;
      }
      else
      {
        if ( a1 > -300 )
        {
          if ( a1 > -150 )
          {
            if ( a1 > 150 )
            {
              if ( a1 > 300 )
                v2 = 2 * (9 * a1 - 900);
              else
                v2 = 14 * a1 - 600;
            }
            else
            {
              v2 = 10 * a1;
            }
          }
          else
          {
            v2 = 14 * a1 + 600;
          }
        }
        else
        {
          v2 = 2 * (9 * a1 + 900);
        }
      }
      sprintf(byte_585340, "cp %d", v2 / 32);
      result = (int)byte_585340;
    }
    else
    {
      sprintf(byte_585340, "mate %d", (-30000 - a1) / 2);
      result = (int)byte_585340;
    }
  }
  else
  {
    sprintf(byte_585340, "mate %d", (30001 - a1) / 2);
    result = (int)byte_585340;
  }
  return result;
}

kranium
Posts: 55
Joined: Mon Aug 02, 2010 10:49 pm
Real Name: Norman Schmidt

Re: Computer Chess Biggest Liar

Post by kranium » Sat Nov 05, 2011 10:59 pm

Some thoughts...

IMO, this clearly explains the difference between Strelka 5 output and Houdini...
Jury O's Strelka 5 simply doesn't obfuscate the PV output scores as is done in Houdini, therefore resulting in Stelka's large eval scores.

It seems plausible to conclude that the overall doubling of the eval scores in Houdini (resulting in a kind of 200 cp eval scheme),
and consequent obfuscation of output,
could be a viable (albeit crude) method/attempt to make Houdin's evals appear significantly different to Robbolito's,
and may have been the 1st (accidental?) step towards an strength improvement...

User avatar
Uly
Posts: 838
Joined: Thu Jun 10, 2010 5:33 am

Re: Computer Chess Biggest Liar

Post by Uly » Wed Nov 16, 2011 7:40 am

kingliveson wrote:On pages 2, 3, & 6 of this thread, it has been demonstrated that Houdini chess is RobboLito (IPPOLIT) modified.
Hi kingliveson, at Rybka Forum it has been argued that this is irrelevant because the engine that is being sold is Houdini 2.0 and not the Houdini that you examined. I recall that at some point you claimed that at the code level, Houdini 1.5 and 1.0 were mostly the same, making all versions of Houdini RobboLito (IPPOLIT) modified.

Please, could you show this connection between Hoidini 1.0 and Houdini 1.5?

Prima
Posts: 328
Joined: Tue Dec 14, 2010 6:12 am

Re: Computer Chess Biggest Liar

Post by Prima » Wed Nov 16, 2011 9:11 am

@ Kingliveson

This request is the same nonsensical argument that the Rybka 2.3.2 & its source code that competed at Leiden(?) wasn't the same engine/source as the commercial Rybka 2.3.2a made publicly available. However, that turned out to be a blatant lie. As always. And/or the implication that succeeding versions are usually completely rewritten from scratch, therefore, non-germane to its predecessor. Not likely, in the course of version-control.

I wouldn't waste my time, energy, effort on ANY requests on such. But your call...

User avatar
Uly
Posts: 838
Joined: Thu Jun 10, 2010 5:33 am

Re: Computer Chess Biggest Liar

Post by Uly » Wed Nov 16, 2011 9:41 am

Prima wrote:This request is the same nonsensical argument that the Rybka 2.3.2 & its source code that competed at Leiden(?) wasn't the same engine/source as the commercial Rybka 2.3.2a made publicly available.
I don't think that's a corresponding argument.

The corresponding argument is that Rybka 3 was never checked, and Vas was disqualified without any idea of whether the latest versions of the engines (the ones that would compete) were clean or not.

Is there a doubt that Houdini 1.5 is mostly Houdini 1.0? Otherwise showing it would be easier than showing any Robbolito to Houdini 1.5 connection, hopefully (this is of course if kingliveson cares, he already proved that Houdini's roots are unoriginal).

User avatar
kingliveson
Posts: 1388
Joined: Thu Jun 10, 2010 1:22 am
Real Name: Franklin Titus
Location: 28°32'1"N 81°22'33"W

Re: Computer Chess Biggest Liar

Post by kingliveson » Thu Nov 17, 2011 3:51 am

Uly, it is true what you wrote on Rybka forum that I said Houdini 1.5 is 95% RobboLito. That comment was posted on this thread, then deleted. The reason, well, because I foresaw exactly this request to produce more disassembled code.

Now, assuming I indulged, someone will come back and say, "but, but, that's not Houdini 2.0." I have examined Houdini 1.5a and it is not much different from Houdini 1.0 which is 99.9% RobboLito (IPPOLIT) -- just ask Jury Osipov.

Of course Houdini 1.5's evaluation is improved compared to 1.0, in addition to tablebases support; this is where the 95% figure arose, and it was just being generous. I do not have Houdini 2.0 to examine, but I am quite confident it was not re-written, and from what we know, it is just 1.5a fine-tuned with additional features, hence it is still improved RobboLito.

See, at first, I was a little disappointed in the community for encouraging Robert Houdart's behavior and not calling him out on an obvious and deliberate in-your-face lie, but then it occurred to me that most in the community are not software engineers/developers. Mr. Houdart knows this, and he is able to use silly vagaries and cunning language to take advantage of the situation.
PAWN : Knight >> Bishop >> Rook >>Queen

Webkikr
Posts: 16
Joined: Sat Feb 05, 2011 9:18 pm
Real Name: Peter Skinner
Location: Canada
Contact:

Re: Computer Chess Biggest Liar

Post by Webkikr » Wed Nov 23, 2011 10:33 pm

thorstenczub wrote: i could make a post and call conkie or skinner blackmailers.

it would be fact.
I'm a blackmailer? Since when? I admit to a few hostage takings, but never have I used materials to extort anything from anyone.

This is "skinner" btw. Did you forget "czub" that I have dirty little pictures of you in a dress groping "Witty"? For shame...

Jeremy Bernstein
Site Admin
Posts: 1226
Joined: Wed Jun 09, 2010 7:49 am
Real Name: Jeremy Bernstein
Location: Berlin, Germany
Contact:

Re: Computer Chess Biggest Liar

Post by Jeremy Bernstein » Thu Nov 24, 2011 9:15 am

Houdini wrote:
Ron Langeveld wrote:Well said Don!

You have my full support in this ethical debate, and it won't be just words. For the very same reason I might use dirty derivatives like Houdini but I will never be buying it. Komodo on the other hand I will buy as soon as the MP version is available.

Ron Langeveld
ICCF GM
Please be consistent with yourself, and stop using the free versions of dirty derivatives like Houdini.

Robert
Ah, finally a bit of honesty. Good on you, Robert Houdart. ;)

http://talkchess.com/forum/viewtopic.ph ... 302#434302

jb

User avatar
thorstenczub
Posts: 593
Joined: Wed Jun 09, 2010 12:51 pm
Real Name: Thorsten Czub
Location: United States of Europe, germany, NRW, Lünen
Contact:

Re: Computer Chess Biggest Liar

Post by thorstenczub » Sat Nov 26, 2011 1:06 am

Webkikr wrote:
thorstenczub wrote: i could make a post and call conkie or skinner blackmailers.

it would be fact.
I'm a blackmailer? Since when? I admit to a few hostage takings, but never have I used materials to extort anything from anyone.

This is "skinner" btw. Did you forget "czub" that I have dirty little pictures of you in a dress groping "Witty"? For shame...
http://dejure.org/gesetze/StGB/253.html

blackmail people is a crime in germany.
§ 253
Erpressung

(1) Wer einen Menschen rechtswidrig mit Gewalt oder durch Drohung mit einem empfindlichen Übel zu einer Handlung, Duldung oder Unterlassung nötigt und dadurch dem Vermögen des Genötigten oder eines anderen Nachteil zufügt, um sich oder einen Dritten zu Unrecht zu bereichern, wird mit Freiheitsstrafe bis zu fünf Jahren oder mit Geldstrafe bestraft.

(2) Rechtswidrig ist die Tat, wenn die Anwendung der Gewalt oder die Androhung des Übels zu dem angestrebten Zweck als verwerflich anzusehen ist.

(3) Der Versuch ist strafbar.

(4) In besonders schweren Fällen ist die Strafe Freiheitsstrafe nicht unter einem Jahr. Ein besonders schwerer Fall liegt in der Regel vor, wenn der Täter gewerbsmäßig oder als Mitglied einer Bande handelt, die sich zur fortgesetzten Begehung einer Erpressung verbunden hat.
it is judged in germany with punishment up to 5 years or money-payments.
the EFFORT of doing it alone will be punished.

you tried to blackmail me, conkie also tried.

you said if i am publishing this or that, you will ban me.
conkie said he would publish private email if i am not doing this or that.

thats blackmailing somebody. as i said, this is a crime in germany.

User avatar
kingliveson
Posts: 1388
Joined: Thu Jun 10, 2010 1:22 am
Real Name: Franklin Titus
Location: 28°32'1"N 81°22'33"W

Re: Computer Chess Biggest Liar

Post by kingliveson » Mon Dec 05, 2011 7:48 pm

It appears another software developer has joined the "Witch-Hunt" described by Mr. Houdart. It was surprising reading the comments as it seemed people were afraid to call things as they saw them -- or come out and say 'you are not telling the truth' when one is indeed being dishonest.

One thing that makes me laugh though is the idea that Houdini is written in C++. :lol: Perhaps it's time for correction to Houdini chess' Wikipedia page to reflect facts, that is, it is RobboLito (IPPOLIT) modified.

Richard Vida wrote:
Robert Houdart wrote:Strelka 5 is Houdini 1.5 RE, not Houdini 2.
I share your fears, any #1 engine will be RE'd in a matter of months. For Houdini 1.5 it's taken about 6 months.

Robert
You mean 6 months until Strelka has been released to public. You do not know how much time did Yuri Osipov spent on RE, neither when he did start his RE project. For me it took barely one week. As Milos S. pointed out, you made RE very very easy because of starting your project from Robbolito codebase... Personally I see no problem with that (other than you not telling the truth).
http://talkchess.com/forum/viewtopic.ph ... 498#436498
PAWN : Knight >> Bishop >> Rook >>Queen

Post Reply