a Telltale position

[SuneF]

No, Komodo goes to some trouble to make sure the values of the pieces are centered. We have a mobility adjustment factor so that low mobility is actually negative and the pieces square tables have negative as well as positive values in them too. It should not be the case for us that we have to use low base values for pieces to compensate for all the bonuses we might give.

For mobility and piece square tables I've done it also, or at least tried to do it. I've not actually bothered to collect the full required statistics about the patterns frequency, which I think would be needed to figure out the correct centering.
In many cases I've deliberately decided not to center the values, as it tends to create more complex code.

For instance the code

Code: Select all

score += 105; // Base material
if (somePattern)
  score += 5;
else
  score -= 5

is nicely centered around 105 (assuming 50% for somePattern), but it seems more complicated and possibly slower, than the equivalent but uncentered version:

Code: Select all

score += 100; // Base material
if (somePattern)
  score += 10;

Though maybe the first approach is easier to manually tune for a GM.

[Tom Barrister]

Yes, thousands of positions have been run and shown to be considerably different. This is expected, since, whatever has happened, Rybka uses a material table that gives a very specific flavor to the evaluation. R1 is not that close to fruit in terms of move selection.

Miguel

Be that as it may, there's just as much hard irrefutable evidence that Rybka 1.0 beta is or isn't based on Fruit as there is that Robbolito is or isn't based on Rybka.

I'm willing to accept that this is true. Fruit (some version) >> Rybka (some version) >> Robbolito (initially).

[michiguel]

Yes, thousands of positions have been run and shown to be considerably different. This is expected, since, whatever has happened, Rybka uses a material table that gives a very specific flavor to the evaluation. R1 is not that close to fruit in terms of move selection.

Miguel

Be that as it may, there's just as much hard irrefutable evidence that Rybka 1.0 beta is or isn't based on Fruit as there is that Robbolito is or isn't based on Rybka.

This has nothing to do with your statement. Whether R1 is derived or not from fruit, the move selection does not indicate that. R1 does not look like fruit. Someone bothered to look at this (as opposed to what you said) and the results are not what you imply.

Miguel

I'm willing to accept that this is true. Fruit (some version) >> Rybka (some version) >> Robbolito (initially).

[Tom Barrister]

Then it's possible that Rybka 1.0 beta could actually be (or not be) a clone/derivative of Fruit 2.2 (or whatever version), rewritten for bitboards and whatever else, yet not return the same evaluations. If that's the case, running positions through them would be useless.

Which brings us back to the "Is Rybka a derivative of Fruit x.x?" and "Is (insert your favorite engine) a clone/derivative of Rykba x.x)?" debates.

Sad as it is to say, I wouldn't put it past any of the commercial engine authors to help themselves to the Ippolit/Robbolito/whatever sources for ideas to improve their products. It's certainly in the realm of possibility.

It isn't as though anybody could tell. There isn't even full agreement whether Rybka 1.0 was Fruit, and there are still people arguing that the Ippolit series isn't Rybka. The proponents of each say that Rybka 1.0 was Fruit rewritten for bitboards, and that the Ippolit series began exactly as reverse-engineered (or decompiled or whatever technical term fits) Rybka. If the programming geniuses can't come to a consensus about alleged wholesale copying, who's to know if the authors of Naum, Shredder, Zappa, Fritz, Hiarcs, etc. take a few things here and there and put them into the next versions of their programs (if they haven't done so already)?

Understand that it doesn't matter to me. I've already bought and paid for Naum, Rybka, and Shredder. A while ago, I liked to run engine tournaments on Arena, and those were some of the top engines. If Houdini, Fire, and Ivanhoe are based on Rybka, as is claimed, I can use these without feeling guilty, since I've already paid Mr. Rajlich for his engine. If Xann wants, I'll send him a few dollars for using Rybka, on the premise that it's based on Fruit.

In fact, I might as well cut out all the middlemen and send some dough to John McCarthy. Or better yet, to John Wood!

"Greetings Professor Falken."

"Hello Joshua."

"A strange game. The only winning move is not to play.

How about a nice game of Global Thermonuclear War?"

[UncombedCoconut]

Hello Chris et. al.,
Do you have a download site and checksum for the historical Houdini executable you used?
Being an MP version, hence nondeterministic, how often does it produce the PVs you posted?

[Graham Banks]

Have you noticed how the argument evolves over time? It goes something like this:

1. Vehement denial and bogus evidence that it cannot possibly be a clone.

Then as the evidence that it IS a clone becomes stronger ....

2. Assertions that Rybka is a clone of Fruit (presumably this is some kind of excuse)

Then as the evidence becomes overwhelming

3. Justifications and excuses for why it doesn't matter to them anyway.


You can see we have progressed to phase 3 now.

Couldn't agree with you more and subsequent posts just confirm what you've said.
Strength is all that matters to some, no matter what or how.

[Tom Barrister]

Have you noticed how the argument evolves over time? It goes something like this:

1. Vehement denial and bogus evidence that it cannot possibly be a clone.

Then as the evidence that it IS a clone becomes stronger ....

2. Assertions that Rybka is a clone of Fruit (presumably this is some kind of excuse)

Then as the evidence becomes overwhelming

3. Justifications and excuses for why it doesn't matter to them anyway.


You can see we have progressed to phase 3 now.

Couldn't agree with you more and subsequent posts just confirm what you've said.
Strength is all that matters to some, no matter what or how.

You can make the same argument for both sides of this debate, and proponents of both sides have used the same tactic.

[yanquis1972]

As long as he is not charging any money for his improved Robbo open source based engine I am not complaining, as well as all others who are using it, and there are many...

I don't have a problem with it's existence but I don't believe it should be getting recognition as if were an original program.

i don't think this is a huge problem (ie, i don't think there is a lot of recognition in that regard). like ippolit & rybka 3, it seemed clear to me from the start that houdini 1.03 was just an ippolit with values changed to be more conservative (& some other changed that were pointed out about basic endgame knowledge). strength was identical, etc. now i still have trouble stating 'houdart started with ippolit as a base' because i'm relying on my own end user impressions & hearsay (if from a lot of respected & knowledgeable people). but if he did, & i know i've talked to don about this, i don't really see it as a crime or 'crime'. maybe it's unfortunate or worse in the eyes of other programmers who started from scratch, but for whatever reason i haven't developed the empathy to sympathize with this POV.

all that said i too wish, if he did so, that robert would simply admit it. i'm not sure what it would change. i'd like to see on his website, say, that he used ippolit & what major ideas & knowledge he added or changed (from a layman's POV) from that source. i'm also not sure about the argument that he's obligated to make his program open source. i don't really know where that comes from.

[edit -- one thing it would certainly change, is that we could stop having so many threads like this]

[Carotino]

Perhaps the whole thing is not as easy as they say. Vitruvius IS a Hippo-derived, as I always said, and here is its output (on a small laptop dual core):

1 +9.13 11.... Dxd4 12.Cxd4 (0.00)
1 +0.76 11.... Th8 (0.00)
1 +0.72 11.... Tg8 (0.00)
1 +0.60 11.... Dd7 (0.00)
1 +0.50 11.... Dd6 (0.00)
1 +0.28 11.... Cc6 (0.00)
2 +0.14 11.... Cc6 12.d5 (0.00)
3 +0.20 11.... Cc6 12.d5 Cb4 (0.00)
4 +0.11 11.... Cc6 12.d5 Cb4 13.O-O Dd6 (0.00)
5 +0.24 11.... Cc6 12.Af4 f6 13.d5 Cb4 (0.00)
6 +0.36 11.... Cc6 12.Af4 f6 13.d5 Cb4 14.O-O Dd7 (0.01)
7 +0.28 11.... Cc6 12.Af4 f6 13.d5 Cb4 14.O-O Dd7 15.Tfe1 (0.01)
.....

14 +0.29 11.... Cc6 12.Af4 e6 13.O-O Df6 14.Ce2 Df5 15.Axc7 Tac8 16.Ad6 Tfd8 17.Aa6 Txd6 18.Axc8 Db5 19.Tfe1 Dxb2 20.Ab7 (0.41)
15 +0.37 11.... Cc6 12.Af4 e6 13.O-O Df6 14.Ce2 Df5 15.Axc7 Tac8 16.Ad6 Tfd8 17.Aa6 Txd6 18.Axc8 Db5 19.Tfe1 Dxb2 20.Ab7 (2.13)
15 +0.28 11.... Cc6 12.Ae3 Cb4 13.O-O Cc2 14.Tad1 Cxe3 15.fxe3 e6 16.d5 exd5 17.Axd5 Tb8 18.Ce4 f5 19.Cfg5 (3.36)
16 +0.36 11.... Cc6 12.Ae3 Cb4 13.O-O Cc2 14.Tad1 Cxe3 15.fxe3 e6 16.d5 exd5 17.Axd5 Tb8 18.Ce4 f5 19.Cfg5 (5.32)
16 +0.44 11.... Cc6 12.Ae3 Cb4 13.O-O Cc2 14.Tad1 Cxe3 15.fxe3 e6 16.d5 exd5 17.Axd5 Tb8 18.Ce4 f5 19.Cfg5 (10.00)
16 +0.39 11.... Cc6 12.d5 Cb4 13.O-O Dd7 14.Ce5 Df5 15.Te1 e6 16.dxe6 fxe6 17.Cf3 Tae8 18.Ae3 e5 19.Tad1 Dg4 20.Ae2 (10.07)
17 +0.45 11.... Dd6 12.O-O Cc6 13.Ae3 Tad8 14.Tac1 Db4 15.Cb5 Ca5 16.Ad3 c6 17.Cxa7 c5 18.b3 cxd4 19.Af4 Da3 20.Cxd4 Dxa2 21.Ae5+ Rg8 (13.37)
18 +0.37 11.... Dd6 12.O-O Cc6 13.Ae3 Tad8 14.Tac1 Db4 15.Cb5 Ca5 16.Ad3 c6 17.Cxa7 c5 18.b3 cxd4 19.Af4 Da3 20.Cxd4 Dxa2 21.Ae5+ Rg8 22.Tc2 (24.08)
18 +0.42 11.... Dd6 12.O-O Cd7 13.Ae3 Cf6 14.Ce5 a6 15.Ae2 b5 16.Af3 Tae8 17.a4 c5 18.Tad1 b4 19.Ce4 Cxe4 20.Axe4 f6 (25.34)
18 +0.42 11.... Dd6 12.O-O Cd7 13.Ae3 Cf6 14.Ce5 a6 15.Ae2 b5 16.Af3 Tac8 17.Tac1 c5 18.Tfd1 Tfd8 19.h3 b4 20.Cc4 De6 (37.82)
...

As seen, the first move is equal to all the Hippo (even if the evaluation is different) but the final score is completely different from zero. After 60 seconds it stabilizes at about +0.48 / +0.50...
According to the arguments I've read, Vitruvius should be considered "virgin" ... But instead he is a "sailed woman"!

[rodolfoleoni]

........................

Vitruvius IS a Hippo-derived

As seen, the first move is equal to all the Hippo (even if the evaluation is different) but the final score is completely different from zero. After 60 seconds it stabilizes at about +0.48 / +0.50...
According to the arguments I've read, Vitruvius should be considered "virgin" ... But instead he is a "sailed woman"!

The move gen is the same, it seems clear. What version of Ippo/Robbo/Ivan is Vitruvius derived from?