Author: Dann Corbit
Date: 16:28:17 05/03/01
Go up one level in this thread
******************************************************************************* The Louguet Chess Test II (LCT II) FAQ Version 1.21, April 02, 1996 ******************************************************************************* This FAQ is intended to give readers informations on the LCT II Chess Computer Test, in 10 questions and answers (at least for this version). 1) What is the LCT II Chess Test ? It is a set of 35 chess positions designed to evaluate the strength of Chess Programs and Chess Computers. It gives a good approximation of the International Elo rating of a program. The test is available via a ZIP file, LCTII121.ZIP, which you can download from Mads Brevik chess page (http://knoll.kih.no:8001/stud/mads2/chess.htm) or I can send it directly to you, just drop me a word about it in my mail. LCTII121.ZIP contains the following files : * README.TXT : General information, history, instructions for PC chess programs. * RATINGS.TXT : Some results on Pentium 90 and Pentium Pro 200 * SUBSCRIB.TXT : How to subscribe to "La Puce Echiqueenne" * HOWTO.TXT : How to run the tests * DIAGRAMS.TXT : Solutions of the tests and comments * LCTII.XLS : Excel file with detailed results for PC programs (P90 & PP200). * LCT2LIST.TXT : The complete LCT II rating list on various hardware. * LCT2FAQ.TXT : Questions and answers about the LCT II (this file). The 35 positions are provided in their native formats for : - Chess Genius 1, 2, 3 and 4 - Hiarcs 2, 3 and 4 - Zarkov 2.5, 2.6, 3.0 - Socrates 3 - Fritz 1, 2, 3 and 4 - Chessmaster 4000 - M Chess Pro 3.5, 4 and 5 - W Chess 1.0 - Kallisto 1.83 - Gideon Pro 1.0 - Rebel 6 and 7 - Virtua Chess 1.0 and Virtual Chess for Windows 95 - Crafty 2) Who is the author of the LCT II ? Me. Ah, ah. Well, my name is Frederic Louguet, born in Boulogne, France, in 1963. I am a freelance journalist, writing articles and books about computer hardware and software, especially on Computer Graphics and 3D Image Synthesis. And, of course, I love Chess, and computers. That's why I created this test, which is published every three months by "La Puce Echiqueenne", Paris, France. 3) What is "La Puce Echiqueenne" ? This is the french magazine of reference on Computer Chess. Reviews, technical informations, games, analysis, databases, etc. It is published by ACTIV-ECHECS, Paris, France. If you want to subscribe, please send : France : 120 FF - CEE 160 FF - USA / Canada 200 FF by money order to : ACTIV-ECHECS 101, AVENUE DU GENERAL LECLERC 75014 PARIS, FRANCE TEL. AND FAX : 33-1-45.47.80.83 OR VIA INTERNET E-MAIL : ORELLANA@DIALUP.FRANCENET.FR 4) What is the main idea behind the LCT II ? Yes, there is an idea ! I strongly believe that, in a lot of well-known tests, too much importance is given to tactical positions. Tactics are important, but not THAT important. After all, combinations appear only when there are positional weaknesses in a position, allowing tactical moves to be effective. But if you look at the whole game, you can see that positional moves are much more frequent than tactical moves. So the idea behind the LCT II is to give positional play the place it deserves. There are 35 positions in the test : 14 of them are positional, 12 are tactics and 9 are endgames. So 40% of the overall rating can be obtained by doing a perfect score on positional tests. 5) What were my goals in creating the LCT II ? The test must : * Give a good approximation of the International Elo rating of a program, by correctly balancing relative importance of positional play, tactics and endgames. * Be available to everybody who wants to test his own chess computer or program. However, it must be strong, because the test is only valid for Elo greater than 1900. * Be simple to use. No coefficients, no complicated formulas. The program get points each time it solves a position in a certain amount of time (see question 7 for the details). You just add all the points to 1900, and you get an approximation of the International Elo rating. * Have a good number of positions, but not too much. It must be run in a few hours, so it is possible to test a program on different hardware, with different hash-table sizes, with different styles. It can provide a lot of useful information. 6) Was there an LCT I test ? Yes. The first version of this test was created in July, 1992, for the first issue of La Puce Echiqueenne. It was a bit more complicated than it is now, but it worked OK. Of course, the LCT II is much better, and was introduced in issue #8 (November 1994). 7) How do I run the test ? First, you must have a Chess computer or a Chess program on a PC or a Workstation. Then you have to follow this procedure : * The program tested must be configured to level : Infinite. * Permanent brain must be disabled. Some programs begin their analysis in setup mode, so it is very important to disable this option. * Learning tables must be disabled (or moved elsewhere on the disk during the tests). * The program must run each test during 10 minutes (600 seconds). At 10:01 (601th second), you must stop the analysis. * The time to keep is the time when the program finds the best move and does not change his mind until the ten minutes are over. For example : the best move (A) is found in 32 seconds, but the program change his mind at 1'56 and now prefers (B). Later, he again considers (A) as the best move after 7'16, and does not change anymore until 10:01. So 7'16 is the time to put in the results. * It is very important to let the program "think" 10 minutes, and particularly for positional tests, since some programs change their mind frequently. For tactical tests, it is of course not necessary. * Points attributed : 30 points if move found between 0 and 9 seconds (included) 25 points if move found between 10 and 29 seconds 20 points if move found between 30 and 89 seconds (30" - 1'29) 15 points if move found between 90 and 179 seconds (1'30 - 2'59) 10 points if move found between 180 and 389 seconds (3' - 6'29) 5 points if move found between 390 and 600 seconds (6'30 - 10') 0 points if move not found in 10 minutes The lower limit is 1900 Elo points. For example, if a program gets 105 points (positional), 200 (tactical) and 70 (endgame), its rating will be 1900 + 105 + 200 + 70 = 2275. The upper limit is 1900 + (30 points * 35 tests) = 2950 points. That is 40 % for positional moves, 34 % for combinations and 26 % for endgame. You can see from these figures that it is useless to test entry-level programs or Chess computers with the LCT II test. It is clearly intended for strong ones. 8) Can the LCT II Elo be compared with the Swedish Rating List Elo ? It was not intended to be compared with the SSDF List in the first place, but it is inevitable that such comparisons occur. In fact, I have found that the LCT II correlation with the Swedish Rating List is quite high. Here are a few examples, taken from the 1995/10/31 list. In these examples, all programs were tested on Pentium 90, but you can find similar figures on slower hardware : Program LCT II SSDF Genius 3.0 2425 2447 Rebel 6.0 2405 2430 Hiarcs 3.0 2440 2420 MChess Pro 4 2360 2377 Fritz 3 2380 2371 It seems good enough, doesn't it ? More interesting, in some cases the LCT II predicted these scores before the SSDF. For example, the rating of M Chess Pro 4.0 was 2396 in the 1995/09/08 version of the SSDF list, and it has now come down 19 points, to be only 17 points more than the LCT II rating. Of course, 35 positions can not replace thousands of games, but it can give a good indication of the strength of a new program, only days after it is available to the public, not months. Do not forget another thing : the SSDF list is based on Computer against Computer matches, and it is clear that some programs perform better against humans than against other computers. So I think that the SSDF List is a very useful tool, but that there is also a need to have a good general purpose test. 9) Where can I send my test results ? You can send your LCT II results by fax or E-Mail to ACTIV-ECHECS (see question 3) or directly to my E-Mail, which is : louguet@worldnet.net. Please don't forget to mention your configuration (PC or Workstation, CPU, RAM, cache, hash-tables). If you have comment or suggestions, I will be pleased to hear them. Suggestions for a future LCT III will also be considered. 10) An LCT III test ? Well, there will definitely be a LCT III test. Why ? because I have already a few ideas to make the test even better. When, I don't know, because I have a lot of other things to do before ! And I can assure you that the conception of the LCT II took a LARGE amount of time... So the answer is probably : not before 1997. Nevertheless, I will be collecting material in the meantime. So if you have good positions that are difficult and interesting enough, please send them to my E-Mail (see 9). Maybe they will be included in the LCT III. Thanks for reading this FAQ. ----------------------------- Frederic Louguet, 04/02, 1996 E-Mail : louguet@worldnet.net -----------------------------
This page took 0.01 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.