Author: Enrique Irazoqui
Date: 06:44:23 10/11/98
Go up one level in this thread
On October 11, 1998 at 08:28:24, Moritz Berger wrote: >On October 11, 1998 at 07:12:58, blass uri wrote: > > >>>Finally, my educated guess (as promised ...): >>>Fritz 5 on P120 with 32 MB hash tables - about 2500 SSDF ELO >> >>This rating is based on learning. > >No, it isn't entirely based on learning. Of course not. Learners don't play chess... :) Seriously, Fritz 5 is a strong engine and in every way a wonderful program, no doubt about it. Only that I never found it stronger than other top chess engines. The SSDF rating list says otherwise, but on closer inspection what you find is that this rating is SSDF specific and not necessarily transferable to other kinds of competition. The purpose of the Elo rating is to predict the performance of a player in a given event, based on statistic calculations from games played in the past. This Elo rating in the SSDF list appears to be valid for the SSDF way to test and for no other events. On this list, Fritz 5 is between 30 and 70 points higher than all the other top engines, so you would expect Fritz to score about 55 to 60% against them, and this is not necessarily true. If Fritz 5 plays 20 games long matches, it will get this score. If it plays 10 games matches, it won't. If in tournaments it plays a different opponent every game, it won't get that score either. In fact, it doesn't. Then, the SSDF rating list is no indication of the score Fritz 5 will get in a future event, unless this event reproduces exactly the SSDF way to test. In other words, this Elo list defeats its own purpose of being able to predict performances. An example I already posted: in my tournament of 200 games at 40:2, Fritz 5 scored 44% in the first half and 59% in the last half. In the SSDF games, 64% in the first half and 72% in the second half. If you play a tournament of 20 games matches, you will get a very different performance if Fritz 5 plays these 20 games in a row or if you split these matches in two halves by exiting the program and restart it again for the last 10 games. What's going to be the predicted performance after the Elo rating? It depends of how you make Fritz 5 play. That's why I'm talking of an SSDF-specific rating. Again: I think Fritz 5 is very strong and a tactical wonder. I think the SSDF is not to blame for distortions in their rating list. But these distortions are real. Learners can be SSDF specific, meaning: much more efficient in the SSDF way to play matches than in any other case, and this influences greatly this rating list. Enrique > I played about a hundred games from >Dirk Frickenschmidt's "Play The Game" positions and some from positions of my >own choice, without any books at all. Plus a lot of Nunn-tests, but I don't know >if you will readily discount these results. Additionally, I played dozens of >games at faster time controls on chess.net and ICC. > >Finally trying to appease you, I should mention that I built a tree from about >1000 Anand games taken from CB Mega Database '98 and Fritz still scored >EXTREMELY well against all kind of opponents (yes, also right from the start of >each "match" :-)). Just try it yourself: Play an engine match with "Anand" book >for Fritz5 and "PowerBook" for Fritz5. "Anand" will not perform much worse... > >I think you're putting to much emphasis on books and learning. The Fritz 5 >engine itself is really strong and can compensate for almost every "book >deficit". Even using the old Fritz3 or Fritz4 .fbk books without any learning >(<200KB each) or converting books from Rebel, Genius or Chessmaster to Fritz >trees will produce decent results for Fritz. The tree offers more in terms of >education, statistics and learning, but if you're not willing to give it MBs on >your HD, that's also a reasonable choice. > > >Moritz > >P.S.: Since Fritz doesn't do "inclusive learning", i.e. it doesn't learn new >moves but just changes weights after each game, I recommend to "import" (not >merely learn) games into the book after you played against strong opponents at >significant time controls (i.e. at time controls where you will be using the >book in the future). This helps Fritz to build its own book, e.g. starting from >the aforementioned Anand (or Fischer or Kasparov or Capablanca...) repertoire. > >No, I didn't do this in my computer-computer matches. >Yes, I recommend doing this if you start with a very small book.
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.