Author: Thorsten Czub
Date: 04:36:47 04/01/02
for all fans of CSTal 1 DOS, you can now get a special version. this is an easter version !! together with a special style of me, it gets arround 50% or even more, on autoplayers (400 and 1200 mhz versions in time control 40/120 measured), against latest fritz7.0.0.6 and junior7. The version number is 276_h. this executable has a history i will try to explain: Moritz Berger once claimed that CSTAL would lose 9-1 against Fritz. Or was it 10-0 ? i don't remember exact. We had a talk about this. it was an email-discussion with chris whittington, torsten tschoop, moritz berger, andreas mader, matthias wüllenweber, peter schreiner, dieter steinwender, dirk frickenschmidt, karsten bauermeister, berthold seifriz, enrique irazoqui. I have copied here this old email stuff. it says enough about 276 ... All you need to do is to download the zip-file using the link from http://www.thorstenczub.de or directly download it from: http://hometown.aol.de/Mclanecxantia/C276h.zip and copy the files out of the zip.file into the CSTal directory. The style should be loaded in the GUI of CSTAL1 or you can rename it in chess.sty and copy it in the CSTAL/DATAFILE directory, this loads it automatically each session. Have fun and let us know about the results... happy easter ! Thorsten Czub --------------------------------------------------------------------------------------------------- > Subject: The really official match: Fritz5 vs. CSTal (tc 40/120) > Date: 18 November 1997 12:30 > > The story so far, in a few words: > Mr. Moritz Berger, a friendly force of the ChessBase-Beta-Tester-team, > came to me with a bet. > He wanted to offer me a six-pack of beer when Chess System Tal would win > in a tournament of 10 games (40/120) on his machines against his > Fritz5.01 engine. > Because ChessSystemTal ranked in Paris (for some unknown reasons) a > little higher than Fritz5, we have come into a discussion before this > bet, how strong each program is. > This tournament should, at leas, show some more evidence. < loads of amusing Czub-ism and amusing Matthias-isms snipped for sake of my sanity > > > I answered to him: > > Wait until you see the results of MY match. Maybe you won't laugh or be > happy afterwards. > I don't see a reason is Moritz, testing a broken version, however this > makes him very confident that Fritz5 is stronger than CStal. > My games 276h against Fritz5 don't look like Fritz5 could win the match. > This fits into my pattern from Paris. > So wait and see. > > He replied: > > I am not laughing at you. Lets accept that Fritz5 is weaker than CST. I > am > nevertheless interested in the question why you are so unhappy about the > fact that two programming philosophies as radically different as CST and > Fritz can both lead to the capability of playing reasonable chess with > sometimes even comparable style. Why don't you share my view that the > higher the variety of paradigms the more fascinating computer chess > becomes? > > Matthias > > Matthias is mainly right: > It is not important how the result of a match is. It is an enrichement > to the whole computerchess market when two programs of a different > paradigm fight against each other, or have almost the same playing > strength. > I would have nothing against this statement. It was always my point of > view that we need MANY different programs from different programmers and > different companies. It is new to me that CHESS BASE has come to this > opinion too. But nevertheless, when a 180.000 NPS program like Fritz5 > plays a 4000 NPS program like CSTal, two different paradigms fight > against each other. I would say it is to early to subscribe Mr.Bergers > point of view, that the fast-search paradigm has won over the > slow-artificial-intelligence approach (how else could one name all the > bugs the programmer Mr.Chris Whittington names FEATURES or INVENTIONS in > his program-versions). > This discussion has not come to an end by now. > > > In the moment, it looks (and Paris - seen, completely ignoring Cstals > participation, but the rest of the games played there - has given me > right) like the knowledge approach leads. > Shredder won Jakarta championship. Junior Paris. Virtual-chess2, an > intelligent program plays knowledged too, follows second. Shredder2 in a > top 3rd position in paris. Mchess7, as fourth also very near. Rebel and > Hiarcs both in the top of the ssdf-list. This all indicates that the > fast-searchers have NO CHANCE at all in the moment to influence the > top-rankings anyway. The massive hardware increase (up to DEC Alpha 767 > Mhz !) that was forced by some fast-search- companies/programs in Paris > has NOT lead to the wished TOP-ranking. > The whole Paris championship was (from the very first beginning to the > end) dominated by knowledged programs. And if Rebel or Hiarcs would have > participated, it would not have developed any different. Maybe even > worse for the fast-searchers. > > I will continue to play more games. And as long as Mr.Berger cannot > provide any data that my data is wrong, we have data against data. > Please understand: I don't doubt Mr.Bergers data. I only doubt that it > represents the real playing strength of both programs. There must be a > system-immanent bug in his test-lab. Wether this is in his hardware or > came at least with installing the Whittington software downloaded from > Oxford. He will find out, I am sure. > > Results so far: > > CStal 2.5/4 > 1 wins > 3 draws > 0 losses > Aaaarrrrggghhhh :) So now we have a new N93 version, the 276_h version which defeats everything no matter how fast, which shows the way, truth and light of the non-materialist paradigm and casts the forces of the Imperium into the pit of Beelzebub. Great. Only I never saved the source code. And I don't understand why or how it is 'better'. Moritz, if Czubix is so right about this (and the reason I never saved the source to 276_h was because (a) it was broken, and (b) I thought it was a bad version) - you have a 276_h, why not play again the tournament with this version ? Place a large lead screen between yourself and the programs, and try to think of something boring, like the skyline of Bonn, or German humour, or something, so as not to influence the micro-processors. We will see. If 276_h wins, then I will pursue its source code recreation as the holy grail :) If it loses then, of course, because you are willing it, I will agree with comrade Czubix :) Chris --------------------------------------------------------------------------------------------------- Subject: Re: The really official match: Fritz5 vs. CSTal (tc 40/120) Date: 19 November 1997 21:22 On Tue, 18 Nov 1997 04:30:42 -0800, you wrote: >The story so far, in a few words: >Mr. Moritz Berger, a friendly force of the ChessBase-Beta-Tester-team, >came to me with a bet. <snip> >Mr.Moritz Berger gave CSTal his p5/166 Mhz machine and used his p5/133 >machine for Fritz5.01. He used 128 MB RAM to get at least 58 MB >Hash-Table size under Fritz5. My Toshiba notebook has 80 MB RAM at the moment (58 MB HT for Fritz). I wrote you this, of course, but please feel free to send me some more RAM ;-) CST got my Dual P166 machine with 128 MB RAM (48 MB HT for CST). CST therefore had a 25-30% speed advantage. You make it sound like it was the other way round... < snip> >I said ok. So Mr.Moritz Berger continued his official match with the >before downloaded "Paris"-version 284a (from the Oxford server) and >after 10 rounds the result was a clearly 9 from 10 for Fritz. <snip> >Please take the text that follows with a SMILEY. It is meant with many >:-) arround ! <snip> >Is the program wrongly styled. Is the guy a fool ? ** YOU ** insisted on using "active" style like you did in Paris. I recommended "manic", but you insisted on "active". >Are his machines from east-germany ? I thought it would suit your ideology to buy machines from your comrades, you know those red tower cases with Industrial Strength Marxist Operation System (ISMOS) for Personal Data Storage. <snip> >Papperlapapp. Schnickschnack. >>From my point of view Mr.Moritz Berger has used a broken version (by >will or without any willingly target) to show that a broken version >plays broken chess against an average opponent program. I asked you if I should take 276H (best version here so far, got 38% against Fritz in ~20 games at 60/60 time control). You told me to use the "official Paris" version :-) ***************** hello from cw ..... Oh gawd ..... Yes, Paris is broken. The count_bits routine (counts the pieces attacking a square, amongst other things) was adding 16 to the piece count whenever a king was one of the attackers. Stupid, but true. I'm surprised it plays at all, but it seems too ! There's now a version with this fixed. Czub and Moritz can get it tomorrow. Yes, manic is better than active. No, I don't believe in 276_h version. So we all disagree totally about everything ......... :) This is what I think of CSTal. It does quite well against the medium speed searchers, like Rebel and Mchess, especially if they have speculative evaluations. Its most difficult opponent IMO is Genius, strange that a dull and accurate medium speed searcher shoudl have the least problems. Fritz, I stopped testing against for unknown reasons, some time ago. But I find 1-9 a little impossible to happen, really :( But, long runs in one direction or another are frequent. We wins loads, then we lose loads, against the same opponent. CStal seems worse with black. More worse than you'ld expect from the standard w/b statistics. Moral seems to be play totally conservative and dull against it is the best policy. Introduce speculative stuff, or speculative pruning against it, and you start to get into difficulties. Just a thought. Chris Whittington < snip > >Here the situation is nearly similar. >Mr.Moritz Berger wanted to do a bet with me, and would pay me a six-pack >of beer or 3 currywurst if CSTal would only get 50 %, he said. >So the case is clear: from his point of view, CSTal had to lose, Rumpelstilzchen. >otherwise he would not only lose his bet, but also his reputation known >as Mr.Moritz Massive-Chess-Hash-Base-Table-Berger. I was out of beer at that moment, that's all I have to say about this. > >Mr.Moritz Berger has the version 276h and knows about how to set the >style (deep search and capture-search to manic) yes, finally you came to the same conclusion - in Paris you used "active", right? > and could reproduce the >games IF he would have these fast machines. Also any other guy in the >world would be able to reproduce these games with having a commercial >Fritz5 and the version 276h I used in Paris. You agreed on the conditions, I even offered you to repeat the match any time with stronger (read: non-broken) CST versions. But you needed your 2.5 KB hash table Fritz to compare a 4000 NPS CST with a 4000 NPS Fritz ;-) < snip, thanks for the games > >It does not look like Fritz5 is the superior program. you chose to ignore again the 60+% results of Fritz 5.01 vs. Rebel 9 and M-Chess 6. >Also my results, although I used the same opening-positions, do in no >way fit to Mr.Bergers opposite results. you know why ... >I can only believe in a minor advertising clue. This was a private match until you cc:ed games to the Dalai Lama and his friends ... I'm missing comrade Gysi and the late Erich Honecker on your cc: list ... >"I have never won a chess game against a healthy opponent." >Alexander Alekhine. ... so you reduce hash table size ... >Wait until you see the results of MY match. Maybe you won't laugh or be >happy afterwards. why don't you experiment with different voltage for the Fritz machine? >I don't see a reason is Moritz, testing a broken version, however this >makes him very confident that Fritz5 is stronger than CStal. There's nothing wrong with CST in general ... Don't take this match too serious, calm down ... You have always the excuse of the broken "Paris" version ... ;-))) <snip> >I would say it is to early to subscribe Mr.Bergers >point of view, that the fast-search paradigm has won over the >slow-artificial-intelligence approach (how else could one name all the >bugs the programmer Mr.Chris Whittington names FEATURES or INVENTIONS in >his program-versions). >This discussion has not come to an end by now. I *never* said something like what you suggested above. *Never ever*. In fact I plan to write a slow _and_ stupid program myself as an alternative to the "fast or intelligent" dilemma ;-) >In the moment, it looks (and Paris - seen, completely ignoring Cstals >participation, but the rest of the games played there - has given me >right) like the knowledge approach leads. >Shredder won Jakarta championship. Junior Paris. Virtual-chess2, an >intelligent program plays knowledged too, follows second. Shredder2 in a >top 3rd position in paris. Mchess7, as fourth also very near. Rebel and >Hiarcs both in the top of the ssdf-list. ... all of which are basically "fast searchers". On P166, Crafty gets less NPS than Rebel and just slightly more than M-Chess, less than 50% more than Hiarcs ... If NPS is the term you believe in ... >I will continue to play more games. And as long as Mr.Berger cannot >provide any data that my data is wrong, we have data against data. and then what? This is not a court, it's just for fun ... unless you want to make a holy war out of it. PLEASE no more ideology in this innocent little experiment. Go ahead, prove that Fritz needs more than 8 MB hash to play decent chess. I have maintained that view since 2 years now, nice to have you agree to my conclusion :-) Moritz P.S.: I offer you to buy a K6 200 (after all, it's gotten dirt cheap now) and a suitable mainboard, then let CST (a version you chose, a style you send me) play against poor Fritz on half as fast P133 hardware. I would play a 10 games match again with Currywurst for you if CST gets 50% or more and beer for me if it loses. Maybe you will also stop Fritz-bashing if it prevails? --------------------------------------------------------------------------------------------------- > Subject: Re: The really official match: Fritz5 vs. CSTal (tc 40/120) > Date: 20 November 1997 10:55 > > Ahhhh ! The answer !! > Please Moritz, > accept that my answers have nothing to do with you in person but with > your abstract point of view your mention in your email: > > Moritz Berger wrote: > > > > On Tue, 18 Nov 1997 04:30:42 -0800, you wrote: > > > My Toshiba notebook has 80 MB RAM at the moment (58 MB HT for Fritz). > > I wrote you this, of course, but please feel free to send me some more > > RAM ;-) > > I do not like having much ram nor having much money. Only materialist > have both. > Quality instead of quantity, Moritz ! > Not Hamburger but good food. Not the amount of food is important Moritz, > the quality of your food ... > > > > > CST got my Dual P166 machine with 128 MB RAM (48 MB HT for CST). > > > > CST therefore had a 25-30% speed advantage. You make it sound like it > > was the other way round... > > Pah - dinosaurs ! > Cstal is very slow on a p1 with only 16 KByte CPU-Cache ! > It plays much better on a MMX machine due to the bigger 32 Kbyte > internal CPU-Cache. The K6 with its 64 KB cache is - of course - even > better. Small programs fit with small internal cache, KNOWLEDGED > programs play better with a big cache. > I do not accept your 25-30 % announcement. EACH program runs different > on each machine. > How do you want to measure that one machine is xyz % when each program > has a different speed-relation ?? > You can only relate ONE program on 2 machines. Another program has > another speed-increase. > > > ** YOU ** insisted on using "active" style like you did in Paris. I > > recommended "manic", but you insisted on "active". > > Right. My testings at 60/60 controls on k6/200 shows: MANIC extensions > has sometimes problems (often in a won position) to find a way out of > the search-tree. > On slow machines I would assume using active is better. > On k6/200 40/120 I do use manic, not because I am your opinion which > style is better, but because different hardware/or time-control requires > different settings. > I am still the opinion that on slow hardware or fast games active is > better. > > > I am no marxist. > > > I asked you if I should take 276H (best version here so far, got 38% > > against Fritz in ~20 games at 60/60 time control). You told me to use > > the "official Paris" version :-) > > Your results of maximal 38% are a reason of your experimental-setting. I > got, as I have shown you in games and scores 60-78%. > May I remember you on the fact that CSTal 276h got 37.5 % on your p133 > against Fritz5 on your p166 with 100 MB hash-table. This was mailed by > YOU not by me. Don't try to fog your own data with opposite machine > usage. > > The reason you get weaker results with 276h has something to do with > your settings. Maybe active on slow machines on 60/60 gets (AS I HAVE > SAID) weaker results. Maybe your hardware is too slow. Maybe you > influence your machines with telepathical forces. I don't know. > I had: > > win draw losses enemy - cstal > > 5 1 3 wchess- 276h > 3 2 0 fritz3- 276h > 3 3 0 fritz3- 276h more games > 1 2 1 genius5-276h > 1 2 0 crafty- 276h > --------------------------------overall > 13 10 4 9-18 66.66 % > > This was all with BLACK !! > My white results are even better. > My results fit to PARIS results. YOUR results do NOT fit. > > > >Mr.Moritz Berger has the version 276h and knows about how to set the > > >style (deep search and capture-search to manic) > > > > yes, finally you came to the same conclusion - in Paris you used > > "active", right? > > As I said, the setting is a very sensible thing. If you have a PORSCHE > but start with driving in the 4th gear, your Porsche has major problems. > The setting is nothing that can be done general or without looking to > hardware, time-control, opponent, weather and the emotional-feeling of > the operator. > Of course for fritz this is not important. My pocket-calculator does > sqr(2) when the sun shines and also when it rains. So does Fritz5. His > calculations are not influenced by anything, neither the weather nor the > position on the chess-board! > It is not playing chess, a calculation is been done. > > > You agreed on the conditions, I even offered you to repeat the match > > any time with stronger (read: non-broken) CST versions. But you needed > > your 2.5 KB hash table Fritz to compare a 4000 NPS CST with a 4000 NPS > > Fritz ;-) > > My Fritz got 8192 Kbyte Hashtables. I am not sure if I can get more out > of 16 MB RAM without swapping. I have not decided to misuse RAM, > ChessBase decided to write a chess program that needs massive Ram and > WINDOWS, an OS that misuses memory in a very pervert way. Not my > problem. > > > >It does not look like Fritz5 is the superior program. > > > > you chose to ignore again the 60+% results of Fritz 5.01 vs. Rebel 9 > > and M-Chess 6. > > Pah. I do not believe in these scores above. Fritz5 is a class weaker. I > do not know you produce your "results", but Mchess6 and Rebel9 can play > chess. Fritz5 does something else. If you want I can show you > masterpieces of your program (I had to operate CSTAL : Fritz5 because > ChessBase is no nice to torture us testers not to publish their > autoplayer. This gave me great insight in some major features of this > program: it seems it evaluates the 0-ply position and the search tree is > only for some hash-table filling. It really plays as if it would only > evaluate the 0-ply and than do some loops, or control the sound device. > Reminds me from the tree approach on Diogenes. > Mchess6 and Rebel8 can play chess. They evaluate not only the root, they > plan a senseful main-line, they feel the trend in the game. The SMELL > what will happen. > You could give me a fritz that shows only 1 NPS and I would not like it, > because the main-lines are senseless and the evaluation is unbelievably > naive. Of course it plays chess. That has something to do with the fact > that it knows almost all rules of chess. > But you cannot convince me in NO way that Fritz5 is stronger than > Mchess6 or Fritz5. Only idiots or Grandmasters can use this engine as a > tool. I am not sure for what, maybe they need a stupid engine to beat, > but their usage seems completely different than any other mail-chess > player I know. > When I write this, Fritz5, with a search depth of 11/31, now 12/12 says > it is -0.22 weaker. CSTal says +6.96 with fail-high. Now Fritz5 says > after 12/34 -0.50 !! > Thats of course the program you claim it would beat mchess or rebel. How > shall this happen ? > Now - in the 13th ply fritz evaluates 0.00 !!! How shall I evaluate this > ? Is the game draw or lost ? > When I don't know about evaluating the position how shall I ever build a > search-tree that fits ?? > No - when other programs have +6 pawns fritz, with the 12-13 plies or > searches thinks it is draw. > > How do you plan in forward when you don't know how to evaluate the 0-ply > good enough. How can you ever find a working plan ?? > > > >Also my results, although I used the same opening-positions, do in no > > >way fit to Mr.Bergers opposite results. > > > > you know why ... > > Yes - because CSTal is stronger than you claim. > > > >I can only believe in a minor advertising clue. > > > > This was a private match until you cc:ed games to the Dalai Lama and > > his friends ... I'm missing comrade Gysi and the late Erich Honecker > > on your cc: list ... > > > > I have sent my friends and also some "enemies" copies. > Gregor has no computer and Erich was a fascist. I don't have fascists in > my friends-circle. > This would not work very good when we meet us somewhere in ONE ROOM. > > > >"I have never won a chess game against a healthy opponent." > > >Alexander Alekhine. > > > > ... so you reduce hash table size ... > > Fritz gets 16 MB like all the other programs. Also they all get k6/200 > Mhz. > If they can only use x/2 hash because of their unbelievable ineffective > user-interface - > that is not my problem. If they need 10 times more to play normal chess, > thats again not my problem. > > Shall I give Fritz more RAM than the other machine ?? Next Ed wants a > special SCREEN to be connected that Rebel plays stronger (e.g. like > Rebel10 runs only strong on 20" screens in resolution 1800x 2500 or > whatever) > and Mchess wants a DVD drive next to put all the Tablebases onto it ? > > > > >Wait until you see the results of MY match. Maybe you won't laugh or be > > >happy afterwards. > > > > why don't you experiment with different voltage for the Fritz machine? > > > > When I would reduce the voltage, fritz would play much weaker than now. > I would strike to play out more silly games. There comes a level of > bullshit I am not willed to operate myself. > I am used to operate CHECK-CHECK and Diogenes. I have tested PANDIX and > Goliath. > But what Fritz would produce on slower machines than my k6... > > > > >I don't see a reason is Moritz, testing a broken version, however this > > >makes him very confident that Fritz5 is stronger than CStal. > > > > There's nothing wrong with CST in general ... Don't take this match > > too serious, calm down ... You have always the excuse of the broken > > "Paris" version ... ;-))) > > Again: > Testers and programmers are in a discussion: > how strong is the engine. > Often chris thinks opposite than me. > My strength is that chris gets mainly the same results after 3 weeks or > 3 months later. > So I play 4 games, he has to play 400 to get the same idea about his > program. > Often then he has no source codes because HE thought this version was > shit. > I do not believe in Mr.Hyatt's nor Mr.Maders ideas about statistics. > I do not need 100 games to find out about a version. > Nor do I need 50 games. > Or 25. > It seems for some humans statistics are more relevant than for others. > Maybe this is a wrath of God. > He tortures the guys who do not believe in GOD. They have to play 10 > times more games to find out about reality. If you believe, you are much > faster. My generals tell me why we cannot do this or cannot do that, but I intuitively understand through years of struggle that I know and that I am right. There will be no retreat and no surrender ...... Now who said this Thorsten ? And, in not less than 300 words, what happened to him ? Just one clue: in his statistical system 1000 turned out to be 12 :))))))))))))))) Chris --------------------------------------------------------------------------------------------------- Subject: Re: The really official match: Fritz5 vs. CSTal (tc 40/120) Date: 21 November 1997 00:05 Chris Whittington wrote: > My generals tell me why we cannot do this or cannot do that, but I > intuitively understand through years of struggle that I know and that I am > right. There will be no retreat and no surrender ...... > > Now who said this Thorsten ? And, in not less than 300 words, what happened > to him ? Please , not always these cheap nazi-images. I criticized Reno and Rolf for using it. now you come with them again. Can't we stop these images ? ****** Sorry, I thought I was being frightfully clever and amusing. Obviously not :( Chris > > Just one clue: in his statistical system 1000 turned out to be 12 > :))))))))))))))) Again I claim that I don't believe in 1000 or 100 games. It is not important to have so much data. You can come to accurate analysis after 1 or 3 games. If you don't believe it, thats your problem. I do believe in my analysis. I had 287 versions of CSTal to proof my analysis and before CSTAL I had some dedicated machines and programs too. Of course only with getting the massive data/experience over the years you get a feeling of the programs. Only with this feeling you can come to judgements that are working although you have only seen a few samples. I don't think that i do never do mistakes. But I on the other hand do not believe that I have to watch 100 games to evaluate a program's strength. > > Chris -- "Was heute noch wie ein Märchen klingt, kann morgen Wirklichkeit sein. Hier ist ein Märchen von übermorgen: Es gibt keine Nationalstaaten mehr, es gibt nur noch die Menschheit und ihre Kolonien im Weltraum. Man siedelt auf fremden Sternen, der Meeresboden ist als Wohnraum erschlossen. Mit heute noch unvorstellbaren Geschwindigkeiten durcheilen Raumschiffe unser Milchstraßensystem. Eines dieser Raumschiffe ist die Orion, winziger Teil eines gigantischen Sicherheitssystems das die Erde vor Bedrohungen aus dem All schützt. Begleiten wir die Orion und ihre Besatzung auf ihrem Patrouillendienst am Rande der Unendlichkeit." Raumpatrouille Orion / Space Patrol Orion ----------
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.