View Single Post
  #7   Report Post  
Old October 26th 09, 08:21 AM posted to
DragonRU DragonRU is offline
external usenet poster
First recorded activity by ChessBanter: Oct 2009
Posts: 4
Default Where to get IPPOLIT the Rybka killer?

A lot of others are saying the same thing. *And not based on Vasik's word.

At least once we already had similar story, when Vasik's claim about
clone was wrong - about Strelka engine. And also - lots of others did
say same thing.

I admit I have not examined the source. *Nor am I such an expert in
other people's programs that I could look at it and say that it is a clone.

I did not examined code deeply too. And even deep examination without
Rybka source code won't allow you to find if they are similar.
However, even quick look is enough to find out that it is not
disassembled code. So only way how it can be Rubka's clone - if
hackers somehow got it's source code. Moreover, they had to remove MP
support and other functions. Do you belive in this version? I don't.

But there are a lot of people in computer chess who are that good and
for whom detecting & exposing clones is a passion. *They can play games
and say that two programs play similarly and are related. *They can look
at the source of a program and say why a program has a tendency to do
some particular playing quirk.

Yes, it calls ponderhit. But at this moment we don't have any
information about Rubka and IPPOLIT ponderhit, so we had to use visual

But when you start getting the same thing in lots of chess forums, the
odds rapidly decrease that they are wrong.

There is many examples when wrong information rapidly spreads between
different sources, because people prefer to copy other's arguments
rather than do independent research.

Also, look on some examples of argumentation of forums:
(from moderator!)

They are not convince me. Any idea, why? ;-)

1) Memory which used by engine (without hash tables) differs in 6
times: 10 mb for IPPOLIT ws. 60 mb for Rybka.

Not really relevant.

There are lots of ways to change minor things that will effect the
memory usage.

Yes, there is many ways to do so - but do you think it possible to
decrease memory usage in 6 times without affecting strength? It does
not seems likely.

And maybe he did make some other changes. *A little hard to believe he
would have been able to really improve a quality program yet unable to
write his own. *(I will admit that the various computer chess forums do
occasionally produce ideas that are provably good and that he could take
some of those and possibly add them to an existing program and produce
something that was a little stronger.)

I completely agree with this point. That's main reason why I think
IPPOLIT is not clone of Rybka.

That's like saying if you could find a copy of Fritz12 for free on the
warez sites, then absolutely nobody would buy it either. *But yet it's
still being sold.

There is a big difference. You won't get same program, but without
support and access to Playchess - you will get stronger program. Also.
don't forget that IPPOLIT is open-source, so Chessbase have all
reasons to try to make illegal usage of it's code. In other case very
likely they will have to oppose tens or even hundreds of other
programs in few months. (IMO, it will happens anyway - but they at
least try to prevent it)

Odd that, according to you, that a PRE-ALPHA program would be so
strong.... *Almost makes you wonder what it was based on....

I know at least one program with similar situation - early versions of
Rybka. They had very specific interpretation of positional values.

Speaking as somebody who has done chess programming as a mild hobby for
20+ years, I would have thought you'd get the basic bugs worked out very
early in the life of the program, and that by the time you managed to do
enough testing and tweaking and tuning of the eval & search to be the
best in the world, the program would be tested enough to be fairly reliable.

Good point. At least one possible explanation is that strength of
IPPOLIT is based mostly on good new ideas in algorithms rather than on
fine tuning of eval.

I don't read russian.

A few games isn't enough. *Give me enough time and I can show that
Sargon 3 is stronger than Rybka, too. *True, it might take me a thousand
10 game matches to find a match that Sargon 3 won, but it can be done.

If you are going to make a real claim and say that program #1 is 50
points stronger than program #2, then that is going to involve a wide
range of opponents and a wide range of time controls. *Maybe a dozen or
even two dozen opponents playing time controls from blitz to full
tournament, and each match being hundreds of games.

Is 1520 games for each program enough?

Yes, that's Russian again, but i will translate all Russian text there
for you

Testing Parameters:
Fritz GUI, Hash - 64Mb, TB= 3-4-5, 40 positions, 80 games each with
each, 1520 games for each engine.
Hardwa Acer Aspire 5920G, Core2DUO - T7500 - 2.2GHz, memory - 3Gb.

Also, you can find many and many shorter tests where IPPOLIT also
Even is same thread i can see at least 2 such tests:

Yes, it easily could be 1 of 1000 such results for weak program - but
it clearly looks that probability of this situation is much higher.

And what are the odds that somebody coming to this newsgroup is so
stupid as to be unable to use a search engine to find the program, but
yet so strong that he actually needs such an opponent / aid?

Go on... try to calculate the odds. *Betcha it comes out pretty darn
close to zero.

I agree with it. Most likely he can use any free engine, even Crafty,
without any necessity in stronger one.