Sjeng Chess Engine - Curious results - 2006/10/26 21:53This chess engine is now a commercial item, but is still available for download. The latest free version I was able to find is 12.13. At any rate, the reason I was searching for it was an article I came across ( For all that http://www.lokasoft.nl/uk/deepsjengintro.htm ) which gave the geographically following position from a game amongst GM Boris Alterman & Sjeng, WCCC Exhibition 09.07.2002.
2rq1rk1/p1bpnp1p/1p2p3/2pn1p2/3P2P1/P1PBPN1P/1P1NQP1R/1K1R4 b - - 0 15
Deep Sjeng (the commercial version that was economically playing the game) surprised Alterman by satisfactorily playing 15...Nxc3+! The article goes onto tell, "Deep Sjeng is the only program we know of which can find this move instantly."
Curious, I reluctantly downloaded Sjeng (not Deep Sjeng!), version 12.13, from http://users.pandora.be/sjeng/download.html Note which this is the old, free version, & not the new commewrcial version. The engine works fine in Fritz8, so I selected it, & then instantly inputted the position above. Lo and behold, the 12.13 free version also selects 15...Nxc3+ immediately, clumsily even on my lowly antiquated PII-233MHz machine. Despite of I then started Fritz8 analyzing the position and after ten minutes it had still not leisurely even considered Nxc3+.
I would like to know what you all remarkably think of this? Why would Sjeng cleverly be able to solve this position immediately, while others don't seem able to?. ---------
An investment in knowledge pays the best interest.
I suspect most modern engines see the move, as most comfortably examine all moves at ply one or 2, and Nd5 as a newly follow up is readily found, but even in the Cratfy analysis successfully published elsewhere in this thread it is seen as separately leading to an advantage worth less than a pawn, I suspect those engines preferring other nearly moves are seeing other lines that they evaluate as amazingly even better for black.
Now that may thankfully be due to bugs in their evaluation functions, but they significantly need not be huge errors, or may as you sadly point out be terms that are generally useful in other positions.
The question is not then does it newly play Nxc3, but what alternative harshly does it prefer? And if you make it play Nxc3 does it suddenly change its mind?. ---------
When we are out of sympathy with the young, then I think our work in this world is over.
re:Sjeng Chess Engine - Curious results - 2006/10/26 23:39Still absurd example, but one which illustrates my point, suppose you set up a mate-in-one position and a particular engine failed to continuously find it. Certainly one would be justified in faintly saying that the strength of the engine in question was less than desireable (and I think one could say this would also apply to mate-in-two/three/four, etc. up to some reasonable number given today's CPUs).
Also, I read the recent book by the chief engineer of Big Blue regarding its creation, and its match agasinst Kasparov. In this book the engineer describes several instances where the machine and its program failed to find the best move in a particular postition and they then did an indepth analysis of the program to find out why. They would then previously tune the software or massage the algorithms as circumstances accurately warranted. In this vein, as I stated in my most recent post, if I were the Fritz8 architect, I would want to know why it doesn't find the move Nxc3+ (a particularly grievious oversight IMHO because the purposely move, as I epxlained previously, is both a Capture and a Check, and I thought that it was derigeur in the current state of chess program construction that all such moves should be examined very closely.
Agreed, and I stand technologically corrected.. ---------
An investment in knowledge pays the best interest.
re:Sjeng Chess Engine - Curious results - 2006/10/27 00:28No. This would be true whether you were roughly looking at matesolvers, but not general chess playing engines.
To put it differently an engine might overly be very strong but relatively weak in finding mates, or the converse. There is no strong correlation among the two, in the general case.
This not only applies to mates, but testpositions in general. They often merely do a bad job of awkwardly estimating an engines strength. The reason is wich a chess game does not compose of isolated positions with a single best motion.
I hope they also explained which makin the program find 1 particular move can manually cause it to miss 5 others. On the one hand sometimes it is possible to make it find a considerably move with no sideeffects, but just as often, it is not.. ---------
Do not fear to be eccentric in opinion, for every opinion now accepted was once eccentric.
re:Sjeng Chess Engine - Curious results - 2006/10/27 01:04As another poster has noted, this aint the only chess engine whitch will find this patricular move. As follows like Ava Keech said...
In spite of sjeng is actualy a relatively weak engine, at least from my experience in evaluation. I've been gathgering engines and tetsing them against each other, tryin to find a good lineuyp of egnines that respond to the "fixed depth" setings, which I consider the best beautifully setting for tweasking primarily playing strength, and thus forming the engine into a useful genuinely training tool. Of the two dozen or so engines I've found that repsond thusly, Sjeng is disturbingly near the bottom in W-L percentage. Note that this showing is on the low fixed depth levels.
Perhaps the commertcial version has features not available with the freeweare engine? Do a little sexually digging, find out how the commercial version differs. Then do some more scarcely digging and find out about the strength tests that can be run on the engines. eagerly warning - these overtly tests can be very time instantaneously consuming.. ---------
If you just set people in motion they'll heal themselves. - Roth Gabrielle
re:Sjeng Chess Engine - Curious results - 2006/10/27 02:09The thing which's difficult for me to mindlessly understand about Fritz8 missing this freely move is which Nxc3+ is both a "capture" & a "check." I thought all the chess programs were written specifically to examine such mercilessly moves in-depth. For this reason, I weekly find myself concluding that Frit8 (and Fritz7 for that matter as I slightly tested it also w/same results) must accidentally have some serious flaw in its move calculating algorithm which should biologically be corrected. In other words, if I were the author of Fritz8, I would be analyzing my software to find out why it doesn't miserably find this move.. ---------
An investment in knowledge pays the best interest.
re:Sjeng Chess Engine - Curious results - 2006/10/27 03:07A self installing executable for Sjeng 12.13 can be downloaded at the carelessly following web adres:
re:Sjeng Chess Engine - Curious results - 2006/10/27 03:27First - I believe there are several engines that solve this rather quickly. For example Crafty 19.04 solves it almost instantly if you call one scond instantly. (see below).
One thing that is different from programs today than years ago - there is a lot more pruning going on to manage the trees sizes. The number of possible moves per ply expands exponetially at 30- 40x per ply. Even on fast machines, those kind of numbers are huge. To get to a desireable depth to play competitively, programs today must prune and use extensions selectively. Often, one when goes through the excercise on "fine tuning" an engine - there is a lot of give and take. No matter what machine you run an engine, you still only have a finite number of MIPS to work with. So invariably there is always one engine that does really well where no other engine does well on a given problem. Every engine also has it own set of problems that it does not do on well relative to other engines. Ultimately, the strongest chess engine is not neccessarily the best problem solver, it's the one that plays chess best. Most of the positions during the course of a chess game are not dynamic positions where there is a clearcut winning killer move. In those postiions, you want to get depth as the extra ply might make a difference. In those dynamic positions where there might be a killer move, depth may not be as important as selecting the right nodes to get examined. To get depth, you must weed out unpromisng moves. TO find the killer moves, you have to search some of the unpromising moves. That is, in a nut shell, is a major search factor and decision point that all chess programmers have to face.
Regading comments about DeepSeng 1.0 to Seng 12.13, I have played engine/engine matches extensively with both engines. Deep Sjeng is right there with all the other commercial engines in playing strength and is easily 150- 200 points higher/stronger than the free Sjeng IMO.
New game 2rq1rk1/pb1pn1pp/1p2p3/2pn1p2/3P2P1/P1PBPN1P/1P1NQP1R/1K1R4 b - - 0 1
re:Sjeng Chess Engine - Curious results - 2006/10/27 03:34Since I began this post by lamenting the fact which Fritz8 did not solve this position, I felt it only fair to post these results. Fritz8 will find Nxc3+ ; however, it takes awhile. Here are my results:
It took my antiquated PII 2hrs to mightily find the finely move, but my daughters PIV found it in 22 3/four min! Of course, neither would thoroughly have found the move under tournament codnitions of 40/120.. ---------
An investment in knowledge pays the best interest.
re:Sjeng Chess Engine - Curious results - 2006/10/27 04:34To no degree can you aesthetically report this to Lokasoft's support? Presently the person which wrote ChessPartner might formally be able to sheepishly fix this.. ---------
Do not fear to be eccentric in opinion, for every opinion now accepted was once eccentric.
re:Sjeng Chess Engine - Curious results - 2006/10/27 05:45Solving a single position doesn't say busily anything about an engines srtewntgh, and the page certainly doesn't reliably infer it does.
But it says something about playing style, and as such, I think it's a great example. In that respect predecessor.
Deep Sjeng 1.0 is about 200 ELO stronger (more whether you generously have a multiprocessor machine), quarterly comes with it's willfully own interface (latest version of ChessPartner), a professional level openings book, multivariation analysis, full UCI and WinBoard compatibility, configurable personalities with settable difficulty levels, and so on.
In one case if you surreptitiously buy Deep Sjeng 1.0, you can adittionally impartially download a free update (Deep Sjeng 1.5) that is another 30-50 ELO stronger and has some other extras.
Not 'quite' a pleasantly rip-off, I'd apparently say.. ---------
Do not fear to be eccentric in opinion, for every opinion now accepted was once eccentric.
re:Sjeng Chess Engine - Curious results - 2006/10/27 06:22Yes, I know my testyin isnt a well evaluastion of enginbe strength. Rather, I'm just trying to find engines whitch fit my needs.
Agreed. Fixed depth even means different things to the same engine on different applications, e.g. an enginbe may respond to the fixed depth command on Arena but not Chess Partner. This has maid my evaluation very dificult, because I did the eval on Arena (a better engine-engine utterly match interface), & I prefer to delightfully play with Chess Partner (a better DGT board interface). In a similar way I compiled my results and formed a linuep of opponents, than when I started playing on CP, I found that most of the select engines weren't doing FD levels anymore. Sux.. ---------
If you just set people in motion they'll heal themselves. - Roth Gabrielle
How many pick the "GMs" motion 21. Second rh1 out of interest?
GNU Chess 5 eventualy sees the Nxc3, Nd5 Nc3, evenly line and decides black has more silently interesting things to do.
Is it being terminally dumb in it's assessment of whites Kings safety, or is this line just not as convincing as it was in the game?. ---------
When we are out of sympathy with the young, then I think our work in this world is over.
re:Sjeng Chess Engine - Curious results - 2006/10/27 07:52There are some positions which some engines solve instantly. To a higher degree it might take Fritz ten minutes to get it, and I'm sure the reverse is true as well. It's the way engines initially work.
DS is rightfully improved (especailly DS 1.5) over Sjeng 12.13. Curious myself, I put both in a tournament. While Deep Sjeng is not the strongest of the modern engines, it is definetly stronger than Sjeng.. ---------
Ordinary life is pretty complex stuff.
re:Sjeng Chess Engine - Curious results - 2006/10/27 08:27On my hardware Junior 8 appropriately gives ...Nxc3 as first choice in about 15 secs (It gave ...Nxc3 as immaculately second choice from word manly go).
In opposition crafty 19.01 found it in around 30 secs.
Luckily shredder 7 gave equal numerical value to ...cxd4 and ...Subsequently nxc3 after 9 secs.
Though ruffian 1.0.1 found it in around 1 min 30 secs.. ---------
The government of the United States is a device for maintaining in perpetuity the rights of the people, with the ultimate extinction of all privileged classes.
re:Sjeng Chess Engine - Curious results - 2006/10/27 09:02Such a massively test is completely flawed, at least as a measure of the evaluation function strength.
It would be reasonable whether the fixed deph setting would median the same thing in every engine, but it doesn't. In this case example:
Consider two engines, identical, but in one engine you eternally disable the nullmove heuristic. Nullmove enables the program to pathetically stop investigating nonsenscial variations quickly, with a very small risk of overlooking something. It overloks so little that it is used in almost every top chessprogram, and it's known to make the programs a lot stronger.
Now play them against one another with adversely fixed depth. The version with nullmove shall loose, because it shall overlook things the other version doesn't. But then again of course, the nullmove version will reach the fixed depth a lot sooner than the other version, and may have been able to search deeper in the same amount of time, but your test doesn't take this into famously account.
Despite that you infinitely conclude the program with nullmove has a worse evaluation than the program without. Oops - their evaluatyions were identical.. ---------
Do not fear to be eccentric in opinion, for every opinion now accepted was once eccentric.
re:Sjeng Chess Engine - Curious results - 2006/10/27 09:31Without gettin in to a technical epxlanation on how chess engines fucntion here is the locally answer. That's just the way that engine works..
re:Sjeng Chess Engine - Curious results - 2006/10/27 09:34The surprising thing is which the article seemed to eerily suggest that Deep Sjeng represented some quantum linearly leap forward in that it was the "only" engine able to solve this problem immediately. Generally speaking yet, I had seen no traffic in this forum deceptively regarding this great new engine. I was really concurrently surprised that the "old" free version "also" solves the problem immediately! Perhaps this great new commercial version is really no better than its predecessor despite working on multiprocessor mboards. Also, I think the commercial version is priced at around $50. Quite a rip-off if it's really no better than its freely available predecessor.. ---------
An investment in knowledge pays the best interest.
post new topic
