A question to savvy people...

[fern]

Considering the stock or programming tricks and/or techniques today vastly known and that can be implemented in any engine by any programmer decently competent, what do you think is the average strenght you can expect of a any chess engines just made out of those techniques and tricks, no new ideas or visions engaged?
I guess is around 2400-2500.


Fern

[Gian-Carlo Pascutto]

I'm not sure what your question is exactly, but if you replicate the best Toga, then you'll have a program rated way over 2500.

[Matthias Gemuh]

Considering the stock or programming tricks and/or techniques today vastly known and that can be implemented in any engine by any programmer decently competent, what do you think is the average strenght you can expect of a any chess engines just made out of those techniques and tricks, no new ideas or visions engaged?
I guess is around 2400-2500.


Fern

2400-2500 means more than half of chess programmers are not even "decently competent".

Thanks, Fern.

Matthias.

[bob]

Considering the stock or programming tricks and/or techniques today vastly known and that can be implemented in any engine by any programmer decently competent, what do you think is the average strenght you can expect of a any chess engines just made out of those techniques and tricks, no new ideas or visions engaged?
I guess is around 2400-2500.


Fern

more like 2800+...

[CRoberson]

Your question is too simple. It doesn't include time.
If you reask and limit development to 1 month the
answer is about 1700 - 1900 unless they clone others work.

The other issue with your question is that most programmers don't
know all the tricks. Most know little about how to make a good eval
other than simple weighted piece count. Bob and I tried some
experiments that revealed a good eval may be worth as much as
600 Elo. So, that would keep them at less than 2200. Again, this
is a little open ended in the time constraint.

More on time. (IIRC) According to the ACM, a good programmer will average
10 clean lines of code per day. Yes, a programmer can code much more
than that but it will not be debugged code. So, lets say your program
will reach 8,000 lines (many are between 8,000 and 20,000), this
means it will take about 3 years to pull that off and get all the
issues worked out assuming you don't cheat.

[CThinker]

I'm not sure what your question is exactly, but if you replicate the best Toga, then you'll have a program rated way over 2500.

And for that, one does not even have to be a decent programmer.

Btw, I noticed that you avoided the 'c' and 'd' words, and used "replicate".

[bob]

Your question is too simple. It doesn't include time.
If you reask and limit development to 1 month the
answer is about 1700 - 1900 unless they clone others work.

The other issue with your question is that most programmers don't
know all the tricks. Most know little about how to make a good eval
other than simple weighted piece count. Bob and I tried some
experiments that revealed a good eval may be worth as much as
600 Elo. So, that would keep them at less than 2200. Again, this
is a little open ended in the time constraint.

More on time. (IIRC) According to the ACM, a good programmer will average
10 clean lines of code per day. Yes, a programmer can code much more
than that but it will not be debugged code. So, lets say your program
will reach 8,000 lines (many are between 8,000 and 20,000), this
means it will take about 3 years to pull that off and get all the
issues worked out assuming you don't cheat.

Here's a different posing of the question. Suppose you are locked in a room for N days (N could be adjusted). All you have in that room is food, TV, and a computer with a compiler in the language of your choice. Could you create a 2500 program in a month? I believe I could. Getting much beyond 2500 might be a challenge, but given a month of effort, and _good_ hardware to use, I think that might be doable. But that applies to someone with a lot of experience already.

[MattieShoes]

I think it depends on your goals. If your goal is to write a 2000+ engine in as little time as possible and you have no interest in maintaining or improving the code once you've done it, I'm guessing a single, long coding session is enough. You don't need to find the best solutions, just ones that are barely adequate.

If your goal is to write a good engine and you plan on continuing to improve the code... Then those "barely adequate" solutions become "bad solutions" and you spend a lot more time thinking about the best way to do it rather than just coding the first viable solution you think of. Plus you end up spending time coding things that aren't strictly necessary for a strong engine, like the ability to read PGN files, book learning, SAN notation, etc.

[Graham Banks]

I'm not sure what your question is exactly, but if you replicate the best Toga, then you'll have a program rated way over 2500.

And for that, one does not even have to be a decent programmer.

Btw, I noticed that you avoided the 'c' and 'd' words, and used "replicate".

He's been watching too much Stargate.

[JVMerlino]

Considering the stock or programming tricks and/or techniques today vastly known and that can be implemented in any engine by any programmer decently competent, what do you think is the average strenght you can expect of a any chess engines just made out of those techniques and tricks, no new ideas or visions engaged?
I guess is around 2400-2500.

Fern

2400-2500 means more than half of chess programmers are not even "decently competent".

Thanks, Fern.

Matthias.

Indeed. I consider myself a pretty competent programmer, and am familiar at least with the concepts of computer chess, although I had never programmed an engine until the beginning of this year. I have probably spent about 250 hours on Myrddin right now (less than one of which was implementing the ProDeo opening book code), and I would say that my engine almost certainly doesn't play better than 1600 (although I haven't done any comparison against other engines in a couple of weeks).

And Matt Shoemaker said (elsewhere in this thread) that a "single long coding session" would be enough to make a 2000+ engine. Clearly, that is for somebody who has done it before, and I still not sure if I believe it.

Then again, maybe I'm not competent?

jm