2009 ACCA World Computer Rapid Chess Championships

[hgm]

If someone writes a completely different engine, and it is also very strong, it is only fair that he should have a better chance of winning. Why do people with only one engine think they are entitled to having the same chance?

Of course a serious question that should be asked, is: what is considered "completely different"? If I would translate my C code into Pascal, it might give me completely different code, but it would be the same engine. If someone replaces the move generator of a mailbox engine by bitboard, without changing search or eval, it is the same engine. If there is a different extensions / reduction scheme, but the same evaluation, it is a different tuning of the same engine.

[Gian-Carlo Pascutto]

If someone writes a completely different engine, and it is also very strong, it is only fair that he should have a better chance of winning.

Why? The argument is that the more similar in strength engines you enter, the better the odds are of finishing on top. This is simple statistics. If you allow this, then the contest turns from making the strongest possible engine into a contest of entering as many engines as you can at an as high level as you can.

If this is what you want, then your argument makes sense. But if it isn't, then it does not.

The tournament should be about who makes the strongest engine. Not about who can make the most.

Why do people with only one engine think they are entitled to having the same chance?

Why is making multiple engines an arbitrary factor that's allowed to increase your chances, but obtaining better hardware or better parallelism not (you argued against it)?

I could see how such a rule would benefit authors that have written multiple, non-parallel engines, but I don't think it's a good idea for anyone else or computer chess in general.

[Allard Siemelink]

2 entrants by same author? Ahem...

Ahem?
Well, the rules clearly allow this providing that the engines are substantially different.
Given the rules, you're welcome to do the same.

[Allard Siemelink]

What I can tell you about Spark is that I started developing it about a year ago (after completing bright 0.4a).
Since then, most of my efforts have been spent developing Spark.

The goal for Spark is to explore different ideas in search and evaluation.
For that, I needed a clean coding base, so I started to build a bitboard perft (Bright is based on a 0x88/mailbox perft) and went on from there.
Currently, I do not think Sparkt's eval is upto Bright's level. Although it would be easy to translate Bright's eval to bitboard code, this would
do me no good as I would end up with the same evaluation and learn nothing.
Eventually, my hope is to combine the best of Bright and Spark resulting in a better engine than both.

Although I would not call the two engines completely different (e.g. UCI protocol code is shared, and both use common ideas (such as null move and LMR), but in a different way) I would say it is clear that they pass the "substantially different" requirement.

If someone writes a completely different engine, and it is also very strong, it is only fair that he should have a better chance of winning. Why do people with only one engine think they are entitled to having the same chance?

Of course a serious question that should be asked, is: what is considered "completely different"? If I would translate my C code into Pascal, it might give me completely different code, but it would be the same engine. If someone replaces the move generator of a mailbox engine by bitboard, without changing search or eval, it is the same engine. If there is a different extensions / reduction scheme, but the same evaluation, it is a different tuning of the same engine.

[swami]

Why? The argument is that the more similar in strength engines you enter, the better the odds are of finishing on top.

In Openwar Tournament, Spark is not even close to Bright in chess strength. Spark plays at 2400-2500 while Bright plays at 2800. So I don't think Spark qualifies as one of the top contenders, Imo. For it to finish on top, it has to rely on Kenny's extremely faster hardware, that's assuming it's mp.

[sje]

The oldtimers here will remember a microcomputer chess event of some decades ago that included a number of commercial entries where, in some rounds, two different machines from the same company were matched.

In one such instance, the nominally weaker (and less profitable) machine had a winning position against a supposedly stronger (and more profitable) machine from the same company. Instead of continuing the game in a normal manner, the company operators simply stopped relaying moves from the weaker machine and so the more expensive model "won" by time forfeit. It was not one of the shining moments of computer chess, and since then there has been a natural aversion to allowing multiple entries by the same author or company.

[CRoberson]

Why? The argument is that the more similar in strength engines you enter, the better the odds are of finishing on top.

In Openwar Tournament, Spark is not even close to Bright in chess strength. Spark plays at 2400-2500 while Bright plays at 2800. So I don't think Spark qualifies as one of the top contenders, Imo. For it to finish on top, it has to rely on Kenny's extremely faster hardware, that's assuming it's mp.

That is a concern. You say that there is a 300 to 400 Elo difference and it runs single processor. I believe Allard has finished the mp code for Spark. On Kenny's 8 processor machine, a mp solution could produce a 6x speed up which could equate to a 250 Elo gain. That makes the difference only 50 to 150 Elo.

[Spock]

[

That is a concern. You say that there is a 300 to 400 Elo difference and it runs single processor. I believe Allard has finished the mp code for Spark. On Kenny's 8 processor machine, a mp solution could produce a 6x speed up which could equate to a 250 Elo gain. That makes the difference only 50 to 150 Elo.

I'm not so sure about that - from 1CPU to 4CPU seems to get most engines approx +100 ELO according to the ratings lists. So from 4CPU to 8CPU maybe another +50 ELO ? That is guesswork, but I'm pretty certain 1CPU to 8CPU would not give anywhere near +250 ELO

[Harvey Williamson]

Why? The argument is that the more similar in strength engines you enter, the better the odds are of finishing on top.

In Openwar Tournament, Spark is not even close to Bright in chess strength. Spark plays at 2400-2500 while Bright plays at 2800. So I don't think Spark qualifies as one of the top contenders, Imo. For it to finish on top, it has to rely on Kenny's extremely faster hardware, that's assuming it's mp.

That is a concern. You say that there is a 300 to 400 Elo difference and it runs single processor. I believe Allard has finished the mp code for Spark. On Kenny's 8 processor machine, a mp solution could produce a 6x speed up which could equate to a 250 Elo gain. That makes the difference only 50 to 150 Elo.

I am not sure the difference in ELO is the correct criteria to judge this by. An author is either allowed 2 entries or he is not. Engines can be 'substantially different' be by the same author and be of similar strength.

[Spock]

I am not sure the difference in ELO is the correct criteria to judge this by. An author is either allowed 2 entries or he is not. Engines can be 'substantially different' be by the same author and be of similar strength.

Yes, good point