bob wrote:
Take the following in context:
(1) you mentioned "tactics first strategy second" as a way of teaching;
(2) I have never seen your source code;
(3) you claim your engine is 1600 or so.
In light of all of those, it would be reasonable to guess "minimum search, minimum evaluation". If you have more evaluation than I suspected, and less search, OK. But evaluation work is essential (IMHO) to reach any decent level of play. Tactics alone won't do it. I'm going to try to set up a temp account on ICC to test this theory using Crafty, just to get some data. Knowing that ICC ratings are pretty badly inflated.
Yes, it would reasonable in context given only those three bits of information. I was sure that I had given more information than that, but perhaps for some reason or other these bits got lost in the mountains of discussion.
Let me clarify:
1) The approach is "teach the basic strategy, such as: Develop your pieces, control the center, castle quickly, avoid isolated pawns, knight on the rim is dim, king opposition in pawn endgames etc.". Beyond that basic strategy, only above about 2000 Elo (international, not USCF) will a human normally have the tactical capabilities and things like concentration to be able to avoid simple blunders and, in particular, have the ability to hold their own in a coffee-house full-scale attack.
More specifically what this means is that one should avoid advanced concepts such as Reti and fianchetto. It's like teaching curveballs to kids: Yes, they will win more games with it in the short run, but they will be short-changing themselves because it will be much harder for them to develop a good fast-ball.
This approach is not unique to the German chess federation, incidentally, but comes from the Russian school of chess. and I believe, but'm not sure, the USCF also advocates that approach. Note that the approach is meant for promising children good enough to be future members of the national team, and these will progress very quickly beyond 2000.
2) And it's better that you haven't, because it is violating many of the solid software engineering principles that I use in my daily work.
You wouldn't need to see it, its approach can be seen directly in games, examples of those can be found on Olivier Deville's ChessWar site, Eden is still deservedly in the lowest division, or on the UCI engines ligue site. There's no need to look there, and I'm not taking it personally that you haven't decided to adopt my engine as your new pet
Looking at PGNs of Eden's games against similar opponents, it can be seen that:
-Eden's moves look much more natural (just look at Apilchess for an extreme counterexample)
-The other engines usually go at least two plies deeper than Eden
-Eden usually likes disregarding individual pawns and prefers stopping the opponent from castling, and really likes to open lines against the king. Usually in positions where you'd normally think there should be some way to use this sort of dynamic compensation, Eden doesn't know how to finish the job, the attack fizzles out and the material usually starts to get felt. Mainly because I haven't put any eval in that would entice it to throw everything at the opposite king...
Not sure if it can be seen by looking at the games, but I but in some of Larry Kaufman's ideas, long before everybody talked about Rybka probably having done so. Worked really well for me (at a very modest level, of course).
When watching a game, it can be seen that Eden still has a lot of potential regarding branching factor and nps (currently about 50,000 nps). I had a version with many of the deficiencies removed and got perfts of around 1.7 Mps, apparently still not much, but much better than now and acceptable to me considering it is written in Java.
3) I have not measured Elo in any meaningful way that would make it directly comparable to USCF or DWZ, the 1600 is a very rough estimate based on my own strength. Because my engine plays more human-like than most others at its level, perhaps this estimation is a little easier. As you mentioned, engine-engine Elos do not necessarily bear any relationship to Human numbers.
In the end, saying that tactics will get you a long way is an extrapolation of such subjects as Deep Blue, Fruit and most of the engines I "meet", which all seem to have a heavy emphasis on pure speed and almost no knowledge. They play horrible games (I mean the lower-level ones) but they still beat Eden's nicer moves.
Incidentally, the name "Eden" is a nod to David Wilkins' program "Paradise", that I discovered in 1989, when I was wondering why I couldn't find any papers that cited the article in all those 13 years. I looked recently, and it seems to have got a little more recognition since then.