Author: Vincent Diepeveen
Date: 03:01:17 06/28/98
Go up one level in this thread
On June 27, 1998 at 10:41:24, Robert Hyatt wrote: >On June 27, 1998 at 09:53:00, Vincent Diepeveen wrote: >>I doubt that Hsu has ever heart of the word 'pawn majority'. >>I dropped some day this word among some chessprogrammers, >>and they all didn't even know what the word means. Because being >>the programmer he's the one who needs to implement so he needs to >>be the one that must exactly understand what it is and what it is not. > > >"Has never heard of pawn majority" after having multiple GM players >work with them to make it better? You really believe that? Then I have >this Bridge in New York that I'd like to get rid of. It's been in my >family for years, but I'm willing to sell it to you cheap. Very cheap. >I'm sure you'll buy that too... I'll tell you a true story about a strong draughts player and a stupid draughts player. The stupid draughtsplayer is an excellent programmer, the strong draughtsplayer doesn't know what the 2 word 'programming languages' means. The programmer is me myself and i at your service. Our draughtsprogram is one of the best in the world. We search deeper than anyone else in history with our draughtsprogram, simply because i know more about move ordering, search algorithms and searching than all other *active* draughts programmers. Yet i lack draughts knowledge. So every tournament we join we hugely outsearch our opponent. Our draughtsprogram searches fullwidth btw, not a single node gets pruned. We don't have those laughable extensions other have in order to find some stupid tricks. We search in some endgames up to 40 ply (!). 40 ply fullwidth. So winning all stones in 30 moves is a microsecond then. That's about 20 ply more than the competition. They know shit from how to use hashtables. Yet we always lost from the same program. the strongest draughstprogram there is (no not Truus that's history): Flits: just searching 11 ply in middlegame. We get 12 ply out of our hashtables when we start to search, so we begin with 12 quickly passing on to the next iteration, and the next and the next. Lately we drew it a few times. Cool! That was AFTER we implemented knowledge. So i know all about how implementing knowledge goes when a strong player is helping you. I was very happy. Our evaluation became from 200 lines C code last time to around 600 lines C code. That's more than a factor 3 increase. Yet we started to play and suddenly it did according to my partner (the strong draughtsplayer) some horrible positional moves. And the weird thing he said: we have this knowledge into our program! So it's doing things wrong which it is supposed to do right. Very weird. Very strange. From an equal stupid program, called Dios we lost this time therefore. In chess we would call it a pin. A long term pin which lost the game. So we had to analyze what went wrong, and during the game i already knew what went wrong. Someone who's not programming himselve doesn't have an IDEA what happens when it must be put into binary format. This means that if the pattern is not true, that it's false, so nothing happens. This means pain, suffering, trouble. External positional help only makes it worse. The programmer needs to know HIMSELVE exactly what's going on. He understands the program How it works. A grandmaster knows shit about programming. It's great to communicate with such a guy. Wonderful. Even better is when he makes a book for your program. Wonderful. But evaluation. HANDS OFF GM. Explain the pattern to the programmer, otherwise it goes wrong. What's logical for a GM/strong player, goes wrong when added to a horrible game player. That happened in our case. That's what i'm seeing in a lot of chessprograms. Chessheuristics that's in it which totally ruin it. A good example is Chess system tal, which wants to sacrafice a rook just for a few checks (Patzer sieht schach, patzer gibt schach). That leads to a cool game. You either die quickly, or you win quickly. Objective it's no good of course. So that GM help is strongly over estimated. If Campbell/Hsu implements the chessknowledge than he needs to be the guy to know the chessknowledge. Not the GM, because the GM isn't implementing, and just giving Campbell or Hsu the pattern is like dropping a 4 years old in a cockpit, telling them the names of the instruments and then assume that that's enough for them to fly the plane. So bob don't tell fairy tales. I know way *more* about getting advice than you might think. I know for example that it sucks. Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.