please add a "tl;dr" at the bottom. although v good, no one will read long post like this
Perfect Chess
Sort:
Apr 26, 2016
0
#4
+ tldr
(The only thing that would be more impressive is that if it was in ALL CAPS.)
" Perfect Chess" Why is h4 better for black. figure it out. b/c theoretically with a worse move by white he can equalize quicker and draw the game out if he so desires. Store only the result, store that however many 3-pice egtbs are won by the side with the extra pice and his turn to move all the time. back up the hdd. keep updating it etc. all you need is Blue Gene. store that if a side has at least 7 pawns advantage static eval and the other side has none and/or just 1 pawn it's always a win seeping away terabytes of strage rquuired. see how easy that was? whatever positions the comp has perfect eval for can also be zapped out of the TBs. and have 1/10 of a 100 TB would be only 1 TB of TBs with a little thinking and a good imagination see what you can accomplhish? I do have a general outline how to prog this into a comp. Anytime a seeeping generalization can be made about a set of TBs in a way that a few TByte can be seeeped clean then by all means do it. always test the theory w/ the aid of the supercomp and ur on ur way. I'd like to really know how much space can be reclaimed for more pieces ( 8-iece?0 tbs. I know the bigger tbs are going to be x ( it's going to be extremely hard to generalize in a meaningful way. If it's a draw, it's a draw. It doesn't matter if it's a draw-in-487 or draw-in-459 with d4 and h4 =draw-in459.insufficient material. have the comp keep updating its knowledge sweeping away terabytes at a time. simultaneously go forward and backward in the game score. somebody smart enough can do this. the smartest people might be making computers faster, better, bigger hdds. andHz but I don't forget how complicated chess and checkers can be. optical, quantum, and dna comps. theoretical limit . if he's not sure at all . . . very quick look and long look, prog the comp even more extremely well and eventually weed out the tail ends of comp score sheets, deleting all the games where this occurs etc. eventually you'll have perfect games. We have perfect games I'm sure but nobody seems to be able to identify the one where this occurs etc. eventually you'll have perfect games. We have perfect games but nobody seems to recognize them for what they are. the question is which one is the perfect game when so many do look so excellently played. the comp might even say 1.h4 draw-in-487 insufficient material. or 1.d4 draw-in-457. At that point it doesn't matter, right? Capa wasn't sure but I'm am. It's the best attempt either d4 or e4. It's d4. almost exactly the same. ever so slightly better. the best chance any chance to win trying ur damnedest. At that point h4 draw-in-448 because it's a better fight. to the bitter end, down to skin and bones to quote b.f. play e4 or d4. the best way to play is that sneaky, a trap i know but if you're trying to win the quickest you'll never play . h4. d4 is 1/100 of a pawn better after letting houdini think on a moder puter for a few days. God only knows. It's for sure a draw in the i.s.p. it's extremely cerebral after the 10th move. i can tell via capa 1927 hoptacong ny. the best player we have is the only one close to approaching perfection. if i say for sure it's a draw you don't know who i am you'll have to take my word for it. if you saw my head size it would help you to dicide whether to bellieve it or not. pefect chess looks like a lot of old player's chess skill and modern computers too and wouldn't make a helluva big impression until something like a deep sac is played and you could check it for accuracy. iow all the world's computers couldn't store a databse if all possible perfect chess games were played unless we cut off the games where a good player would agreee it's a draw with best play. Figure out which positions are drawn by sampling with a chess masters help. even down to a mere master. I'm sure all then the games would fit on a supercomputers hdd memory. the best singular 10 games would just be beautiful! all the hyuman+ comp intelligence would playe pefect chess I'm sure. But who care?
Hoptacong and the French Exchange games aren't included in the Fred Reinfeld's best games of capablanca book I have but they're perfect according to the compprog until the 7th and 10th move each. nobody has got to store the whole 64 pice tb. a perfect eval is one approach. It depends on how fine-tuned is the prog and how good is ur comp. Fischer's best-by-test is total garbage. some comps do like the sicilian though vs the french which is better. i could prove to some ppl who could follow my rreasoning and train of thought. but at least it's an idea that might work. the positions that eval between -1.00 to +1.00 r if it's seen or proven to be drawn then again ZAP all that out the hell out of the TBs. the hdd will be full of gaps that are known to be almost always drawn with the positions that are not almost zero and then you will only need a GB andstead of a TB of space to store the extremely few positions that are the exception. any +7 static eval in the best written compprog on earth, even have the prog have separate way of just evaluating the tb and thoroughly and ZAP out every each and every one that is correctly evaluated and save even more space. and like the tb program that uses a 5 pice to crreated the 6 pice and the 6 pice to created the 7 pice ad infinitum. moscow state university etc lomonosov. if anyone can see a pattern and then have the comp check and double-check the pattern in detail completely, somebody can code this general knowledge again into the computer and save almost again another almost or more than 1 whole TB of hdd space !!Anybody with a brain knows KQQQQk is always won. it's not stalemate because whatever the last move was a move before would have been a win also. think like that. So a quick look by a person who "just knows" that can also be a good chess player and just know w/o being able to prove it to just anyone.if an extremely, if all extremely even endgames are drawn or always drawn (exactly equal material) then what about the opening where it's extremely equal material and stays that way for so long you could imagine why I believe it's a joke for ppl to think at all that it's not a draw w/ perfect play. Isn't there a draw in a formerly famous line in the Sicilian that Kasparov found a way to draw by black by 3-fold repetition at ~ move 18? That that's out another one billion trillionth of the database really easy. I believe it would be easy to sweep away most of the tbs with 3 and 4 pices then the bigger ones ( by way of generalization ) b/c the general rule of thumb would be easier, to follow the train of thought and reasoning. One thing I mentioned before is to use a kind of compprog to test every single tb and compar the comps eval to see if there ever if even anyway at all that a +5.50 eval is not winning and seep away more hdd. if not, for the good, very good, most greatest eval by any prog on earth. But which prog is it? My guess is a top 10 prog on earth. Only have a comp evaluate all the positions in a said database and to see if of the millions if all possible positions it is everr wrong and only keep the positions where it is wrong and sweep away the rest. I ust don't want to have to reinvent the wheel. In collaboration w/ another person familiar w/ the inner workings of tablebase probing and tablebases, I could start the process. If somebody could probea 7-ice TB and find one in which almost every position in it ( a particular one ) is won by the side to mmove with the extra 5.50 pawn advantage then store the positions that aren't won and decrease the size of that one db. That helps in a minor way to eliminate huge databases of tablebases in order to save hdd space. This is an approach that is helpful to solve chess for bigger tablebases. If it can be proven a certain TB positiion is evaluated at +3.17 by the prog another ad infinitum except for, idk, 7 positions ouf of 3 million, then you save 99.9999% of ur files size in that manner. gradually it becomes clear that it is a good approach. a really good supercomp can do it faster than a hobbyist like me can with a $1,000 setup. The way to prove d4 is better than h4 is to think that with d4 is the only move that wins with perfect play whereas with h4 black can Draw with any move HE plays. completely opposite than if white plays d4 and black is sure to lose no matter what b/c white is sneaky and black is lost whatever his defense. Best-by-test Fischer should have used a more nuanced approach to deciding which opening move is better by weeding out a 99.999%of moves by ppl who are weakcomparatively speaking. Only super-GM games would count and then when there are left only a few hundred games out of your database and your comp sees a huge mistake in the games that nullifies any possibililty that perfect or even close to perfect chess is being played then ur left with an empty database. Bobby fischer is a nothing compared to Houdini. rybka found g4!!! in fischer-spassky 1992 in 22 seconds when fischer sat on it for 50 years. So a database of comp vs. comp games should be used to better find out which is etter, d4 or e4, not extraordinarily weak human games. Only the top 3 human players have played so well that their games could be of any use in the attempt to find out if chess is a draw or not with perfect play. Even then that too doesn't work. I would recommend an enthusiast with an octal let houdini 4 think kfor one whole week and then come back to check whether it changed its mind within the past 6 1/2 days. If so what ws its earlier intention. was the eval with its original intention better by more than one hundredth of a pawn? If not then d4 is nearly identically a good move along with e4. If one looks at comp eval and its 0.15 at best in the opening and it only goes down within 10 moves every time, one has to get the message and see the writing on the wall. Even stipulate to ur comp to make only materially-only even for both sides to see how it is possible to way down the line it is impossible to win a completely even 4 pice vs 4 pice endgame. If anybody ever saw a -5.5 pawn evaluatiion be an actually winning position I'd be amazed. There is a point where a quadrillion out of a quadrillion positions that are evaluated +3 r always won or less than 1 pawn materially positiions, a quadrillion minus 1 out of a quadrillion are always drawn I feel confident that since the eval is 99.99999999 % correct of the time and the tbs confirm my suspicions that the eval hasn't been wrong ( hardly ever ) statistically proven by not regorously. Capa said study the ending. I believe that. He played two openings that were very well-played. Hoptacong NY and a French Exchange that Houdini agrees with until Black's Bg4 mirror instead of a better ...h6. If the TB probe fails to see the position it means it's a won position. The generalization made about a few pice tbs could be made about specific 7 pice tablebases and get a small percentage of these swiped away. So iow tbs will not be able to solve chess hardly no matter what. Somebody always thinking will be able to improve on my method but it will not be easy. If all and/or every game of chess is evaluated at its 10th move and last position and the results studied to see what happened, then perhaps it will be possible to see a pattern that shows it is, in fact, possible to forcibly win a game of chess. If only master level games are studied ( to be serious) then almost never has it even been possible to play accurately and win a game of chess through good skill. So a better, the best a person could do is high-level comp v comp Houdini and better, or critter level only and better chess games played on at least 4 CPU STC should be used as a baseline to solve chess weakly. if a cp always win with a + 2.73 - + 9.99 eval, if it always wins with a +2.73 and it always draw with a -2.4 to + 2.5 eval then u can ZAP a big chunk of the DB where this applies ( and is true ). Does anyone have the ability to prog a prog to accomplish this? I can only wonder how long it would take to go through the whole KXk databases and all the way through the 6's and a downloaded single 7-pice and re upload with the fixes. If a cp like Komodo evals all the poistions white v black where white has 3 pices to black's 2 pices correctly if ur talking about at least at 2.3 pawn advantage for the winning side then that's a few tens of % u can ZAP if every -1.7 ot + 1.4 eval is drawn ZAP those in a big way , or mark them in a way that's easiest to probe. Whatever is igger. If its if there are more obviously won positions according to both the prog and the tablebases and that's the way to save the most space, that's the one to delete. and make the TB smaller. Since u can only del either all the won positions via agreement between the comps eva and the TB eval won, draw, or loss then it means the obviously drawn positions, at least according to both eval and tb, then they will return a 0 for draw and a false if it's a lost position. check out Capablanca hoptacong NY 1927 QID ECO e16 and French exchange ECO C01. Nobody needs the KQk since its always won for white. KRk too. like that. u only need an eval prog to test every single position the the tablebase. the best eval ever written by the mind of man. the best one, test 5 or 6 really good ones if necessary. depending on how long it takes the1st 2 then maybe try 5 or 8 and when ur satisfied take a single 3 pice and see how my trick works. Hopefully if even a -1 to +1 eval is always drawn store these in a new separate db. act accordingly. always ZAP all u can. Really this can possibly work on the 6's and even the 7-pice tb. simple.