Author: Robin Smith
Date: 14:04:17 01/05/04
Go up one level in this thread
On January 05, 2004 at 11:34:57, Albert Silver wrote: >On January 04, 2004 at 22:10:34, Robin Smith wrote: > >>On January 04, 2004 at 09:44:39, Albert Silver wrote: >> >>>The interactive analysis mode in CA is completely under the user's control. >>>That's why it is referred to as interactive, so perhaps you are thinking of >>>something else. >> >>No I am not thinking of something else, and yes, CA does give you some nice >>controls for "interactive" analysis; but when I am talking about interactive >>analysis of positions I mean you are sitting in front of the monitor and >>controlling the analysis the whole time, not that you set some parameters and >>walk away. To me that is not really interactive any more than a pilot turning on >>an autopilot by plugging in some numbers, and then going to sleep, is >>"interactive". But even for this type of automated "interactive" analysis, the >>Fritz GUI's "deep position analysis" (the most similar thing to CA's interactive >>position analysis) has a nice feature too, which I keep mentioning: you can >>specify multiple strong engines to do the analysis, including Tiger, Shredder, >>Fritz, Hiarcs and Junior, all at one time. As with CA, Fritz also lets you set a >>bunch of parameters to control the search, although CA's are a little more >>flexible and interesting. But not enough to make up for the lack of engine >>support. And besides, for the _best_ interactive analysis you should be guiding >>it at every step, not setting it up and going to sleep, although admittedly most >>people don't have time for this. > >You are wrong about the interactive analysis. Once the interactive analysis has >started, according to the parameters you set, you retain full control though you >have the *option* of leaving it on automatic pilot. > >You can start and stop it in mid analysis of a line, deepen the analysis of >specific lines of your choice, jump to the next position of interest in the >analysis of a line or subvariation, see what each and every move in the lines is >on the main board, and even see what is in the hash tables by selecting the hash >dialogue option in the local menu. Yes, I know about this. And this is what people should do to get the best analysis, in most situtations. Nothing is better than being there directly controling everything, except in certain types of positions that computers understand very well. And since CB gives you more engines, it works better for this, in my opinion, even though you have to do more handholding with CB than with CA. >>>Much more so than other programs? If it is really so much more wrong than other >>>programs, how do you account for its superior results? >> >>Yes, _much_ more so than other programs. Its superior results have nothing to do >>with the magnitude of the evaluations. If you take a program, any program, and >>modify it by multiplying all the evaluations by 10, it will look very weird to >>human eyes, and be very confusing to someone not expecting this, but the program >>will play exactly the same moves, since all the evaluations were scaled by the >>same amount. Shredder is fairly consistantly optimistic, which does not hurt >>play at all, but "looks" very odd to the analyst and can mislead you if you >>don't know about it. > >Eh? So it scores better because it is continuously more optimistic than every >other program? That doesn't make any sense to me. Scaling everything upwards >should have *disastrous* effects on the results. Can you imagine if everytime it >chose a losing continuation, spurning a repetition, because it thought it was >+1.00 instead of -2.00? It may indeed have optimistic evaluations based on piece >activity, attck potential, etc. more so than others, but I would be shocked if >it was simply scaled upwards universally, and this would somehow have no >detrimental effects on the results. You misunderstand. Shredder's optimistic eval would make a -2.00 into a -4.00, not a +1.00. Shredder is optimistic for the winning side. The optimistic eval would never change the sign of an eval. Take every eval and multiply by 2, this is almost what Shredder seems to do. It has _zero_ effect on playing, or moves selected. But the evals can be very weird to human eyes. They also look very odd when compared to other engines. >>Yes Fritz has this a little too, many engines do, but I think Tigers is bigger. >>Fritz 5 used to be very bad, but I believe each new version of Fritz has gotten >>better. I haven't seen it in Fritz 8, although I haven't looked for it real hard >>either. > >I take it that you don't use it much then, since I presume the attention you >have devoted to the weaknesses of engines is universal and not limited to Tiger >and Shredder. I did not look for weaknesses of engines in a systematic way, it is merely based on my experience using them. It is possible, though I believe not likely, that Fritz 8 has eval discontinuities as great as Tiger's. But I have not bumped into one and I use Fritz a lot, probably more than Tiger. >>>>Engine tournaments can be a very good analysis tool in some circumstances. I >>>>have writen a book on chess analysis that will soon be published by Gambit that >>>>will talk about this, amoung other things. I also have correspondence IM and SIM >>>>titles, a correspondence rating in the world top 100 (and soon will have a >>>>correspondence GM title), so your above statement is no longer true. >>> >>>I was referring to OTB IMs and GMs. I know two correspondence players here in >>>Rio, one who is also one of the top players in the world, and another who is an >>>SIM, but I admit to not being a big follower of the modality, mostly because I >>>think it has been spoiled by excessive use of engines. >> >>Yes, correspondence has been spoiled for many people by chess engines. Some >>people have stopped playing altogether because of it, while others have adapted. >>But I can assure you that at the top level there is much more to correspondence >>play than just plugging a position into a computer. I have run into many players >>rated around ~2300-2350 doing (I believe) this, but not any players rated ~2600. > >You should talk with Uri Blass. I have read what Blass says. I don't believe Blass has played in as strong of events as the cat 13 section I am currently playing, where the average rating is 2563. I have several opponents who, if they had followed a computers advice, no matter which program, they would have lost the game. Instead they were able to salvage draws. Blass also did not just set up his favourite program and let it make all his moves. He used multiple engines and multiple analysis methods. He used engines as analysis tools, not all knowing oracles. >>By the way, OTB GM John Nunn is one of the editors of my book, and he has not >>had any negative comments regarding my section on engine tournaments. I think >>many people, both OTB players and correspondence players, have been keeping >>their program use tricks secret. > >I tend to disagree. I have instructed classes of masters and grandmasters in the >use of both CA and CB, and have found their knowledge of these programs to be >rudimentary for the most part. Naturally, there are true power-user players, but >I think they are a strict minority. I would think that in a class you would only get those GM students with rudimentary computer knowledge, based on what the class is for. I doesn't make sense to in one breath say GM's in general have rudimentary knowledge of these programs and then next only refer to those GM's taking a class from you. For example I seriously doubt GM Nunn, who is very experienced with computers, was in one of your classes. But neither has he been very forthcoming with information about how _he_ uses computers. GM's who are experienced computer users keep the knowlege to themselves as a competitive edge. >Furthermore, the fact that Nunn said nothing >bad means very little. I'm not saying he is secretly critical about it, but it >could also simply be that this idea of serious engine tournament use is novel to >him and thus he had nothing to say on the subject. It *is* your book after all, >and not his. Fair enough. But I believe that if what I said was nonesense he would be all over it. Engine tournaments can be powerfull, when used properly. >>>>>Actually, CA has a LOT more to it than that over CB with all due respect. >>>> >>>>I agree. And with all due respect, CB also has a lot more over CA. >>> >>>Such as? >> >>I have already mentioned the two biggest ones: many strong engines supported and >>the ability (in Fritz) to run tournaments. > >You can run tournaments in CA, though the support is certainly not as extensive >as in Fritz. However you claimed this severely hampered its use as a top-level >analysis tool, and with all due respect, I think that's bunk. And I know it isn't bunk. I didn't win the US corr championship, twice, by using bunk methods. I didn't win a world championship semi-finals by the highest score in any of the sections, 9/10, by using bunk methods. But also please remember I am not saying that CA can't be a top-level analysis tool. It can be. It just has some limitations, like all programs do. One of CA's biggest limitations is its more limited support of chess engines. This is just a simple fact. Can you run Fritz, Junior or Hiarcs in CA? No. This _is_ a limitation of CA. >That you found a way to use engine tournaments as an analysis tool is laudable, >but to claim that without identical support to Fritz, CA does have it after all, >it is no longer a top-level tool.... That's silly as you say. I don't say CA is not a top level tool. I say it is missing a top level feature. I _like_ CA. I _like_ CB. I _like_ the Fritz family GUI. Unlike some people I am not religious about chess programs and which is "best". They are all best... at what they do best. >You also went into detail on flaws in Tiger and Shredder that you perceive, Yes, you asked me to elaborate, so I did. >that make them unusable, Why do you exagerate what I said. I _never_ said Tiger and Shredder are unusable! I use them all the time!! But they have flaws. Just like _any_ complex program. Windows is notorious for bugs, but I still use it, very day. Same with chess programs. >but somehow the flaws in Fritz and Hiarcs, that you haven't >tried very hard to find in your own words, are less important. I never said anything about the relative importance or unimportance of various engine flaws. Please stop putting words in my mouth. The flaws I have found I found by using the programs, reading this message board, etc. I didn't go out trying to find flaws in only Shreddeer and Tiger, nor do I say they are the only programs with flaws. But they are the only programs whos flaws you asked me about. >CA also has incomparable tree support, which your lack of any mention other than >that they are better suggests you barely use, and better organizational and >repertoire tools all of which give it a serious productivity edge over CB and >Fritz (IMHO). How many times do I have to agree with you that CA trees are great? Productivity edge over Fritz? If you are working with trees, yes! But trees, in my opinion, are mostly more suitable for chess study, learning opening repertoirs and such, than chess analysis. My focus is and always has been on analysis, so I don't talk so much about trees, although they do of course have a place in analysis as well. It is just that trees have a relatively lesser importance for analysis, compared to the importance of trees for studying opening repetoirs. >For me these are the main ones, >>although I also think: >>1) The CB GUI is, as you mention below, more attractive. This is not a small >>thing, the CA chess boards are not nearly as comfortable to look at for long >>periods of time. Of course if you set it up and walk away this doesn't matter so >>much. >>2) The CB GUI is easier to learn. One example out of hundreds of easier to >>learn... cutting a position to the clipboard in CB it is just the windows >>standard, <ctrl> C, rather than the harder to remember <ctrl> \ of CA. >>3) The CB GUI is easier to navigate your way around, probably related to the >>easier to learn part. Much of CA's "extra functionality" is fluff stuff I don't care about and that only makes the program harder to use. > >Well, there is no doubt that what is important to some, is fluff to the next, so >that is rather subjective. Granted, it is subjective. >>4) I find that sometimes CA leaves engines running, even after you close the >>program, which then steals CPU time and memory resources until you notice. > >You mean the engine is running in CA even though you closed the engine window, >or that the engine is still analyzing, even though you closed CA? The engine is still running even though CA is no longer running. I have not found a way to reproduce this, although it has happened to me more than once. In either >case, if you can reproduce such a bug, you should report it. I'd also be certain >you have the latest patch installed, though I'm presuming you do. > >>5) If you do a search of a database of endgame positions, such as Comprehensive >>Chess Endings, looking for positions with a particular number of pieces, like >>7-8, it will show zero games, even though there are many. > >Perhaps you are making a mistake in the search setup, such as leaving other >options activated that filter the search undesirably. No, the _only_ search option selected is the number of pieces. >If not, please give the >exact search conditions as I'd like to see this for myself. Get "Comprehensive chess endings" (a database that contains _only_ endgames). Search for a specific number of pieces, and nothing else. Material of anything other than 2-32 pieces. You can try 3-32, or 8-8, doesn't matter. Anyway, it doesn't matter. You aren't meaning to say that CA is bug free, are you? I don't know of any complex but bug free program. >>and some other little things. Not a big deal. I like _both_ CB and CA. But I use >>CB more. >> >>>>They each >>>>have strong and weak points. You talk almost as though CA is better than CB in >>>>every way. This is silliness. >>> >>>Not every way. CB is more attractive, but it is less functional. I'd be curious >>>to know what aspects of its functionality you find to be superior. >> >>See above. >> >>>>Yes CA is better than CB here. But I think Bookup is even better than CA for >>>>this. >>> >>>I wouldn't know as I do not own BookUp. Could you tell me what aspects of >>>repertoire building you think are better? >> >>Bookup is a position database, instead of a games database, sort of like CA >>trees. But Bookup allows you to assign position evaluations, both numeric (such >>as 25 for 25 centipawns) and symbolic (such as +/=) and then you can backsolve >>these positions, somewhat like a program performing minimax, so that positions >>closer to the root have their evaluations determined by the later positions. >>Bookup also imports and exports EPD files correctly, including position >>assessments, and Bookup makes it very easy to find moves that can transpose from >>one branch of the tree into a different branch of the tree. Bookup also shows >>you at a glance how many _positions_ result from a particular move choice, so >>you have a better idea how big each branch is. > >I'd suggest a closer look at the CA manual. ALL of the above can be found in CA, >except for the last item. I don't believe CA can backsolve, which is what makes this powerful in bookup. Am I mistaken? >It will tell you how many games result from the >position, all the known moves that lead to it and from it, Yes, but number of games and number of positions are _very_ different pieces of information. >and it will also show >you at a glance every single known way to *transpose* to it. > >For example, if the same position from a Caro-Kann Panov Attack could be reached >from a Nimzo-Indian, QGA, or other, I can see all those transpositions at once >in a glance. I can compare the statistics of this tree with those of several >others by simply flipping back and forth from tree to tree in a click. Can you have CA add _all_ the posible transpositions to the tree _automatically_, with the subsequent change of statistics? If so, I have not found it and I should take your class. Can CA then do a backsolve of the evaluations, so you can see how the evaluations changed? If so, I haven't seen it and I should take your class. >- I can annotate or evaluate the move, the position, or the variation, and add >my own moves, positions, or variations to the tree (as symbol or numerical >evaluation) without having to rebuild the tree or have the move even exist in a >game. Though of course it will always give me an instant list of stats and all >the games to be found with said position. > >- I can add special markers in the tree to find my special analysis again, or >export this all into a separate tree. > >- I can import or export EPD moves and evaluations, and add them to the CAP >data. CAP data is really CRAP data, no offense meant to Dan, since the basic idea is good. But the implimentation of CAP is flawed. It does not tag which program did the analysis, nor for how long/what depth. It mixes analysis done by different programs. A BIG problem when mixing analysis from optimistic Shredder with asymetric Crafty and other programs. And since most CAP positions were analysed a while ago, on slower hardware with weaker programs than what is available now, you can get better results by just analysing the positions you are interested in at the time of your interest. But Bookup has CAP data access too, if you like CAP. >- I can add a tree to the tree or subtract a tree from a tree > >Within the tree, I can ask optimistic Shredder or buggy Tiger to: You make it sound like I don't like or use Shredder or Tiger. Nothing could be further from the truth. >a) Analyse the current position >b) Analyse the current position and then play against itself from there >presenting the results. >c) Automatically search for new moves. In other words, just tell the engine to >find an adequate opening novelty if possible. >d) Expand the best lines. >e) Check the variation in question with engine analysis. >f) Check the variation and try to find opening novelties in the entire >variation. Fritz can do those things too. >And heck, I'm just scratching the surface and am deliberately not mentioning >many other items. How they can be considered fluff or less interesting than >Engine Tournaments in terms of Top-Level Analysis is beyond me. I haven't even >mentioned the repertoire tools used in the classifiers. Albert, I can see you love CA. I do too. I am just not religious about it. Robin > > Albert
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.