Murka 3.0 released

[Dr.Wael Deeb]

Though the similarity test seemingly has not produced any false positives,

Hi Adam,

That is a bold statement. "We do not know of false positives" would be better!

Anyhow, I don't like the test at all as it is used for a witch hunt. Just have a look at this thread. Even if the source is available anyone not knowing anything about programming can use it to throw stones. But that is something you already mentioned.

I personally do not think that this tool is good to prove anything. If there IS something like "perfect play" it might give a 100% similarity. Similarity is true but not saying that the two programs are identical. So, the better the programs get the more similar they might be ...

Bye
Ingo

Hi Ingo,

That is why I used the word "seemingly". As far as anyone knows (= seemingly), the similarity test has not produced a false positive.

I do not think that this tool can prove anything either. But I am certain it is very good at showing where to look.

Adam

The English language can be unkind to those for whom it is not their native tongue.

Indeed Graham,but still you can find someone who navigates his mind properly in the ocean of the English language....
Ahem regards ,
Dr.D

[jshriver]

Here is the rating change as it's been playing over the last 24 hours.

http://olympuschess.com/tourney/engine.php?name=murkajs

Adding it to core list of engines for OICS.

[lucasart]

it seems to be a derivate

look here: http://computer-chess.org/forum/index.p ... ad&id=2507

This is what happens when you give complicated tools to ignorant people. It's absolutely pathetic to read that.

So what do these numbers show ? All of them are around 50-55. So what ? Is it similar to all of the engines ? How can that be possible (since they are not all clones of the same thing) ?

Now think of this as drawings of a random variable. Under the hypothesis H0: not a clone, you can assume that this similarity nonsense gives a drawing of a variable with some mean around 53 or so (apparently) with a certain standard deviation. If you draw lots of values of that variable, you will eventually see a value above 55. So what ?

Seriously the level of ignorance in this forum is shocking. And what is more shocking is to see ignorants start a smearing campaign based on this similarity test nonsense. Frankly, if you don't understand anything about the similarity test, do not use it. Use instead Astrology or Tarot card reading. I'm sure it's more scientific that what is shown in this link.

I do feel sorry for Igor who must be quite disappointed, after all the effort he put into it, to see a bunch of ignorant accuse him of cheating. By this "measure" all engines can be proven to be clones (sufficiently large sample size will give you > 55 value).

As for me, I would much rather trust the judgment of a competent chess programmer like Jon Dart's (who *did* look into the code), rather than this hocus pocus similarity analysis in the hands of ignorants.

=> not a clone (until proven otherwise).

[geots]

it seems to be a derivate

look here: http://computer-chess.org/forum/index.p ... ad&id=2507

This is what happens when you give complicated tools to ignorant people. It's absolutely pathetic to read that.

So what do these numbers show ? All of them are around 50-55. So what ? Is it similar to all of the engines ? How can that be possible (since they are not all clones of the same thing) ?

Now think of this as drawings of a random variable. Under the hypothesis H0: not a clone, you can assume that this similarity nonsense gives a drawing of a variable with some mean around 53 or so (apparently) with a certain standard deviation. If you draw lots of values of that variable, you will eventually see a value above 55. So what ?

Seriously the level of ignorance in this forum is shocking. And what is more shocking is to see ignorants start a smearing campaign based on this similarity test nonsense. Frankly, if you don't understand anything about the similarity test, do not use it. Use instead Astrology or Tarot card reading. I'm sure it's more scientific that what is shown in this link.

I do feel sorry for Igor who must be quite disappointed, after all the effort he put into it, to see a bunch of ignorant accuse him of cheating. By this "measure" all engines can be proven to be clones (sufficiently large sample size will give you > 55 value).

As for me, I would much rather trust the judgment of a competent chess programmer like Jon Dart's (who *did* look into the code), rather than this hocus pocus similarity analysis in the hands of ignorants.

=> not a clone (until proven otherwise).

I would have to call your thread a wonderful breath of some fresh air. Thank you, Lucas.



All the best,

gts

[Adam Hair]

it seems to be a derivate

look here: http://computer-chess.org/forum/index.p ... ad&id=2507

This is what happens when you give complicated tools to ignorant people. It's absolutely pathetic to read that.

So what do these numbers show ? All of them are around 50-55. So what ? Is it similar to all of the engines ? How can that be possible (since they are not all clones of the same thing) ?

Now think of this as drawings of a random variable. Under the hypothesis H0: not a clone, you can assume that this similarity nonsense gives a drawing of a variable with some mean around 53 or so (apparently) with a certain standard deviation. If you draw lots of values of that variable, you will eventually see a value above 55. So what ?

Seriously the level of ignorance in this forum is shocking. And what is more shocking is to see ignorants start a smearing campaign based on this similarity test nonsense. Frankly, if you don't understand anything about the similarity test, do not use it. Use instead Astrology or Tarot card reading. I'm sure it's more scientific that what is shown in this link.

I do feel sorry for Igor who must be quite disappointed, after all the effort he put into it, to see a bunch of ignorant accuse him of cheating. By this "measure" all engines can be proven to be clones (sufficiently large sample size will give you > 55 value).

As for me, I would much rather trust the judgment of a competent chess programmer like Jon Dart's (who *did* look into the code), rather than this hocus pocus similarity analysis in the hands of ignorants.

=> not a clone (until proven otherwise).

And what is just as pitiful is to read a rant where the ranter has not done a bit of investigation. In other posts at the Computer Chess Wiki forum, Gerhard has posted data for other engines. For Murka 3, he collected the sim data and compared it to all of the other engines he has data for. Then he simply reported the highest matches. I do think it is rash exclude it from his F.A.C.E. tournament, but it is his tournament.

As for Jon, he noted that Murka 3 did not look similar to Stockfish or Ippolit. However, if someone wanted to compare Murka 3 to an engine to look for possible similarities, Strelka is the engine to look at.

Also, I am at a loss at what constitutes a smear campaign. The only two posts that could conceivably come close are these:

it seems to be a derivate

look here: http://computer-chess.org/forum/index.p ... ad&id=2507

When I take a look at the Murka 3.0's code, I have a strange feeling : a sense of already seen.

Some parts of the code, some namings, ... make me thinking about Fruit / Toga family.

But nothing as a clone, just as "in the family".

I can not find the "bunch of ignorant(s)" accusing Igor of cheating. Just one guy in a different forum who thinks, rightly or wrongly, Murka 3 is too much like Rybka 2.3.2a/Strelka 2.0B for his tastes. Personally, I would include Murka 3 in my testing, for I know of several closed source engines that match other open source engines to a much higher degree.

[mvk]

Positions are chosen such as to not have a single best move.

I thought they were hundreds of random positions.

No, they are well filtered neutral positions, which have no clear best move, and unrelated engines often disagree on. There are positions on which even perfect engines can choose different moves, as shown by TBs.

Kai, do you have a source for that?
From discusion with Adam and results from Ed I thought Don's set was randomly sampled.

On the other hand, the 10k-position csvn simtest was indeed filtered, with tactical moves and positions where a large amount of engines play the same move were dropped.

[Laskos]

Positions are chosen such as to not have a single best move.

I thought they were hundreds of random positions.

No, they are well filtered neutral positions, which have no clear best move, and unrelated engines often disagree on. There are positions on which even perfect engines can choose different moves, as shown by TBs.

Kai, do you have a source for that?
From discusion with Adam and results from Ed I thought Don's set was randomly sampled.

On the other hand, the 10k-position csvn simtest was indeed filtered, with tactical moves and positions where a large amount of engines play the same move were dropped.

Don posted here, and I followed the development of Sim to Sim03. First Sim had 2,000+ neutral positions, mildly filtered, and the similarity was up to 75% for related engines at 100ms, with an error bar of up to 1-2% 2SD. In Sim03 Don had 8,000+ positions more heavily filtered. The hit for related engines dropped to around 60%, and for totally unrelated to 35-40%, and the error margin to 0.3-0.7%. This last Sim03, having better filtered positions and more of them, had not only smaller error margins, but a better sensitivity, the distance related-unrelated increased because of better filtering. 10k filtered positions of CSVN seem perfectly adequate.

[Gusev]

Murka 3



ПаРамеТРЫ КОмаНдНОй сТРОКи
Parameters of the command line
--------------------------

-h N (размер табЛицы перестаНовок в мегабайтах, по умоЛчаНие 128)
size of the transposition table in megabytes, 128 by default
-l (вкЛючить Лог) turn on the log


ПРОТОКОЛЫ Protocols
---------

- UCI
- XBoard


сТРуКТуРЫ Structures
---------

- магические битборды magic bitboards


геНеРаТОРЫ хОдОв move generators
----------------

- все ходы all moves
- взятия и превращеНия в ферзя captures and promotions to Queen
- защиты от шаха check defenses
- тихие ходы quiet moves
- шахи (с маской дЛя искЛечеНия уже сгеНерироваННых взятий)
checks (with a mask for exclusion of the captures already generated)


ПеРебОР Search
-------

- fail soft PVS
- сортировка: хеш, хорошие взятия (SEE>=0) по MVV/LVA, 2 киЛЛера, история, пЛохие взятия
- sorting: hash, good captures (SEE>=0) by MVV/LVA, 2 killers, history, bad captures
- табЛица перестаНовок (стратегия простой замеНы) transposition table (strategy of simple replacement)
- пустой ход null move
- продЛеНие шахов, ответНых хороших взятий в ПЛ
- extension of checks, opposite side's good captures in PL
- коНтроЛь повтореНия позиций position repetition control
- иНкремеНтаЛьНая геНерация ходов incremental move generation
- отсечеНия по дистаНции до мата pruning by distance to mate
- особо обрабатывется фв под шахом special handling of FV when in check
- отсечеНия На маЛой гЛубиНе (в сЛишком хороших и в сЛишком пЛохих позициях) small depth pruning (in positions that are either too good, or too bad)
- сокращеНие посЛедНих (по сортировке) ходов с верификацией
- reduction of the last (by sorting) moves with verification
- поиск первого хода перебором На меНьшуе гЛубиНу в ПЛ (IID)
seeking the first move by search to a lower depth in PL (IID)


РасПРедеЛеНие вРемеНи time management
---------------------

- очевидНый ход obvious move
- есЛи Не быЛо существеННого падеНия оцеНки На посЛедНей итерации if there were no significant drop of evaluation on the last iteration
- есЛи Лучший ход Не меНяЛся if the best move didn't change
- есЛи Лучший ход Не помеНяЛся На посЛедНей итерации if the best move didn't change on the last iteration


фв FV
__

- рассматриваем тоЛько поЛезНые взятия и превращеНия в ферзя
- consider only useful captures and promotions to Queen
- шахи На входе в фв checks at the entrance to the FV
- аЛьфа-бета отсечеНия alpha-beta pruning
- сортировка MVV/LVA MVV/LVA sorting


Оф Of
--

- материаЛ с коррекцией material with correction
- цеННость поЛей value of squares
- НичейНость материаЛьНого соотНошеНия draw by insufficient material
- Ладьи На открытых/поЛуоткрытых ЛиНиях rooks on open/semi-open files
- мобиЛьНость mobility
- атака На короЛя attack on the King
- пешечНый щит pawn shield
- возможНость рокировки castling opportunity
- открытые/закрытые пешки open/blocked pawns
- изоЛироваННые пешки isolated pawns
- отстаЛые пешки pawns left behind
- проходНые пешки passing pawns
- Нет своих На пути проходНой no own pieces in the way of a passer
- Нет чужих На пути проходНой no enemy pieces in the way of a passer
- На пути проходНой Нет поЛей атаковаННых тоЛько соперНиком
- in the way of a passer, there are no squares attacked only by opponent
- дистаНция от проходНой до короЛей distance from passer to the kings
- каНдидаты в проходНые passer candidates
- поймаННый сЛоН caught bishop
- запертый сЛоН blocked bishop
- запертая Ладья blocked rook
- разНоцветНый сЛоНовый эНдшпиЛь ending with bishops on opposite color squares
- очередНость хода whose turn it is


КОНТРОЛи Controls
--------

- гЛубиНа depth
- время На ход time per move
- время На НескоЛько ходов time per several moves
- время На партие time per game
- иНкремеНт increment
- бескоНечНый аНаЛиз infinite analysis
- позиций На ход (дЛя теНиНга) positions per move (for tuning)

[OliverBr]

IIRC, there are three regions:

Code: Select all

Similarity > 60%:         guilty/clone/derivative region.
55% =< similarity =< 60%: dubious region.
Similarity < 55%:         safe region.

I like this.