hgm wrote:Gian-Carlo Pascutto wrote:Participating with multiple engines increases the chance of winning. So it's a no-brainer to participate with as many engines as possible. But I only had 1 wife to help me
I doubt that very much. You think you could come up with 3 or 4 engines that share no code, and each stand a chance of winning?
This isn't about whether you stand a good chance, it's about increasing your chances period. If there were 100 entries and I had the weakest program, my chances are slim. But if I had the 5 weakest programs my chances have enormously increased - although they still remain slim.
It's like smoking a cigarette, smoking 1 is not likely to kill you. Smoking 10 is also not likely to kill you, but it's a lot more likely to kill you than smoking 1. No single cigarette is likely to cause any real harm.
Come on, I know you are far more skilled in mathematics than I am, I don't really understand what it is you are trying to refute here. You know this as well as I do.
Or even 3 or 4 engines that share code, but are so differently tuned that they would pass Don's similarity test as unrelated, and do not take a major hit in strength?
Why are you putting all those constraints on this? Why do they have to pass any tests? The argument is that we don't care about authorship. I want to enter 3 program with 3 different books for instance to see which book does best - and they are clearly 3 different entities. Or perhaps I want to try 3 different versions with slightly different mobility values for the bishop.
If the programs are far weaker then your odds do not go up as much, but it still goes up and that's the point.
A simple model to understand this is that all programs have equal chances with the same strength and the tournament is constructed such that only 1 program can win.
Let's assume there are 24 programs and yours is one of them. you have probability 1/24 to win our about 0.0417 if 2 of those 24 programs was yours your odds go to 2/24 or 0.0833. the more programs you have, the better your odds.
I challence you to try that, because it is really interesting: if many points on a multi-dimensional function have an approximately equal value, they are usually in the wing of a peak, and a much higher maximum lies in between.
At the Hong Kong tournament the Deep Blue team (Campbell) confided to me that they estimated their chances of winning to be about 50%. This was based on their belief that Deep Blue was far stronger than everyone else, but that they had to outperform many other programs. Assuming that their calculation was correct, what if ALL the other program belonged to a single author? You would be in a position where "your" odds of winning the tournament were as good as the Deep Blue authors chances.