Topic Rybka Support & Discussion / Rybka Discussion / Vas,This is how Rybka can become great on tactics!
![[de]](/mwf/v12/flags/de.png "Germany")
Yesterday I saw an amazing chess engine. It is called Rebel 12. Rebel 12 itself is not so good at tactics but it can be switched to the engine:Q3-tactical engine.
Q3 can find any tatcical shot in less than a second!!! Even the hardest one!!
So,maybe Vas can build an engine like Q3 or he can make an agreement with Rebel company and connect it with Rybka. Then in the match,Rebel will search for tactical shots for 1.5 seconds and then Rybka will start to think.What do you say?
I think it will increase Rybka overall strength in more than 50 elo.
Here are some examples of Q3 tcatical abillities:
Q3 can find any tatcical shot in less than a second!!! Even the hardest one!!
So,maybe Vas can build an engine like Q3 or he can make an agreement with Rebel company and connect it with Rybka. Then in the match,Rebel will search for tactical shots for 1.5 seconds and then Rybka will start to think.What do you say?
I think it will increase Rybka overall strength in more than 50 elo.
Here are some examples of Q3 tcatical abillities:
3R4/1kr2pp1/2p1p3/1pP1PnBP/pP3K2/8/P4P2/8 w - - 0 1
Bf6!! time to find:00.6 sec.4rrk1/pbq2ppp/1p1b4/n1pPN2Q/5B2/2P3P1/P4PBP/R3R1K1 w - - 0 1
Nxf7!! time to find:0.08 sec.r1bq1rk1/1p2b1pp/p1np4/8/2P1P1n1/N1N3B1/PP2BP1P/R2QK2R b KQ - 0 1
Nxf2!! time to find:1.1 sec.r1b2rk1/pp1n2pp/1qn1pb2/3p4/3P1P2/3B1NK1/PP2N1PP/R1BQR3 b - - 0 1
e5!! time to find:0.04 sec.
![[il]](/mwf/v12/flags/il.png "Israel")
I can only say that this is nonsense.
There is no program that can find every tactical shot.
It is correct that Q3 personality is relatively better at finding sacrifices that is different than tactics because tactics is not only sacrifices and it can be some quiet moves that simply wins material or force mate with no sacrifice.
As far as I know
Rebel personality is good at combinations mainly because it has lower values for material.
It can cause it also to "find" wrong sacrifices so using such engine in games is not going to help because in games rybka is not going to know if the sacrifice that is suggested by rebel is good or wrong.
Uri
There is no program that can find every tactical shot.
It is correct that Q3 personality is relatively better at finding sacrifices that is different than tactics because tactics is not only sacrifices and it can be some quiet moves that simply wins material or force mate with no sacrifice.
As far as I know
Rebel personality is good at combinations mainly because it has lower values for material.
It can cause it also to "find" wrong sacrifices so using such engine in games is not going to help because in games rybka is not going to know if the sacrifice that is suggested by rebel is good or wrong.
Uri
![[us]](/mwf/v12/flags/us.png "United States")
There is no need for this type of engine. Rybka winfinder already finds all the tactical shots in the position.
You can try rybkawinfinder in the position that were posted.
Rybkawinfinder does not find the winning moves in less than 1.5 seconds.
It does not mean that this type of engine can help rybka to play better because the main problem is to know if the move
that was suggested by rebel personality is a blunder or a good move.
Uri
Rybkawinfinder does not find the winning moves in less than 1.5 seconds.
It does not mean that this type of engine can help rybka to play better because the main problem is to know if the move
that was suggested by rebel personality is a blunder or a good move.
Uri
![[nl]](/mwf/v12/flags/nl.png "Netherlands")
Very nice idea imo. I already did some brainstorming bout teaming up chessengines with all their specific strong aspects.
I was more thinking of the idea to team up Rybka Winfinder (for tactics), Rybka 2.2n2 (for positional play) and Shredder 10 (for endgames) and make it one engine, but let them calculate all seperately after eachoter depending on the position. Not sure if such is possible to build, but the idea gives me a good feeling indeed :-)
I was more thinking of the idea to team up Rybka Winfinder (for tactics), Rybka 2.2n2 (for positional play) and Shredder 10 (for endgames) and make it one engine, but let them calculate all seperately after eachoter depending on the position. Not sure if such is possible to build, but the idea gives me a good feeling indeed :-)
Hey Maso,
Its certainly possible to build an interface to a GUI with a UCI interface on one side and multiple engines (running on multiple computers) on the other (this has already been done). The tricky part is figuring out how to fuse the results together in a meaningful way and then figuring out whether the combination is really much stronger than a single engine running alone.
By the way, do you know of any reasonable testing that has been done to evaluate the endgame performance of Rybka and DS10? I tend to oscillate back and forth based on my latest endgame dissaster . :-)
Its certainly possible to build an interface to a GUI with a UCI interface on one side and multiple engines (running on multiple computers) on the other (this has already been done). The tricky part is figuring out how to fuse the results together in a meaningful way and then figuring out whether the combination is really much stronger than a single engine running alone.
By the way, do you know of any reasonable testing that has been done to evaluate the endgame performance of Rybka and DS10? I tend to oscillate back and forth based on my latest endgame dissaster . :-)
![[kw]](/mwf/v12/flags/kw.png "Kuwait")
Rybka Winfinder is great for finding tactical shots. As more and more processors become available, it might be a nice implementation to have say an 8 core system use 2 cores to play Winfinder and the other 6 to play normal Rybka ... and somehow communicate. Or could be another separate independent remote computer running Winfinder and somehow communicating with Rybka to be aware of such tactical shots. I remember Vas thought about that and decided it would be easier to simply make normal Rybka as strong tactically as Winfinder on the same computer. Still it would be nice to have Rybka "consulting" with other remote computers on such matters ... for example I would have a Shredder engine with full 6 EGTB's watching remotely any EGTB shots ... and another remote computer running Winfinder full blast ... and normal Rybka running full blast without local EGTB support (supposedly would make it faster). Rybka would play and if some strong move is found by one of the remote computers ... that particular variation would immediately be sent for main Rybka to analyze and then decide if it is truly good or not.
I think the plan that Vas put forward was to integrate the WinFinder capabilities into the same engine, but I thought it was going to be a separate mode. It would be great if you could specify the number of cores you wanted each mode (normal Rybka and Winfinder) to run on and they could both use the same hash table (would this work?). As you indicate you wouldn't have to share the nodes equally to be able to take advantage of WinFinder's huge speed advantage in some positions.
![[hu]](/mwf/v12/flags/hu.png "Hungary")
Ideally, there would be one engine which is good at both tactics and positional play.
I really hesitate to go after some complex GUI-side solution for integrating normal Rybka and Rybka WinFinder results, when the proper way is to include both positional and tactical variations as appropriate into the same search tree.
If it turns out that the difference between tactical and positional play is really fundamental, and that human analysts are needed for a proper integration, then I think we really can consider some sort of a GUI-side solution.
In the meantime, it's quite easy to provide a completely separate engine with better tactical capability, and let users come up with their own ways to handle the situation. Personally, when we play in the Freestyle events, we never use the WinFinder. I am not a big fan of this constant "search for gold" when playing chess.
Vas
I really hesitate to go after some complex GUI-side solution for integrating normal Rybka and Rybka WinFinder results, when the proper way is to include both positional and tactical variations as appropriate into the same search tree.
If it turns out that the difference between tactical and positional play is really fundamental, and that human analysts are needed for a proper integration, then I think we really can consider some sort of a GUI-side solution.
In the meantime, it's quite easy to provide a completely separate engine with better tactical capability, and let users come up with their own ways to handle the situation. Personally, when we play in the Freestyle events, we never use the WinFinder. I am not a big fan of this constant "search for gold" when playing chess.
Vas
Hi Alan,
I think such multi-to-one engine creation would be great only at the longer time controls. Also it would be hard for the programmer to pick a dominant engine in each position I guess. Best would be to let the real tactical engine try to find a quick solution first, as the author of this thread already mentioned and then let the powerful positional engine do the work after that when there's no " killing" line to find. Question is how long to let the tactical engine give its opinion...Also the evaluation should be valued the same way, as shredder's evaluation is less secure in general.
I don't own DS 10, I only own SP versions unfortunately as I don't have a multi-core system yet. From my tests I did I find Shredder especially good at endgames, not that Rybka is bad at it, but when to point out some weaknesses of Rybka it would be the endgame and some tactical abilities at times. I really wish Vas has time enough to update Winfinder 2.2.
I think such multi-to-one engine creation would be great only at the longer time controls. Also it would be hard for the programmer to pick a dominant engine in each position I guess. Best would be to let the real tactical engine try to find a quick solution first, as the author of this thread already mentioned and then let the powerful positional engine do the work after that when there's no " killing" line to find. Question is how long to let the tactical engine give its opinion...Also the evaluation should be valued the same way, as shredder's evaluation is less secure in general.
I don't own DS 10, I only own SP versions unfortunately as I don't have a multi-core system yet. From my tests I did I find Shredder especially good at endgames, not that Rybka is bad at it, but when to point out some weaknesses of Rybka it would be the endgame and some tactical abilities at times. I really wish Vas has time enough to update Winfinder 2.2.
The winfinder integration idea sounds ridiculous to me.
The whole point about winfinder, and the reason it finds tactics faster, is that it's pruning and positional evaluation are significantly worse than the normal rybka. This is not in rybka simply because rybka plays a LOT stronger with its heavier positional evaluations and pruning.
No magic to it. There's no point in integrating it back into the normal Rybka.
As to integrating several engines: Interesting, though they will take a lot of resources away from each other, which means that to make it effective, you need to come up with a REALLY good way of putting together the results from both engines in a useful enough way to find a mistake/inaccuracy by one or the other (a really difficult problem, if at all possible, to solve effectively).
But then, it seems to me, that the engine that found the correct move would have found it even faster (and maybe found an even stronger move!) if given all the computational power available, so you're counting on the fact that the engines will "correct" each other enough times to make it worthwhile.
Of course, the ideal solution would be to take the strong suits of each engine and integrate them into one very strong engine, but that would require the cooperation of competing commercial programs / companies.
I think something like: Rybka's evaluation as a base, coupled with the endgame analysis from another engine (say shredder, if that's the best one at endgames today...if not, just replace with the best), maybe add Junior 10's compensation evaluation,
and for the icing on the cake, let the Zappa programmer add the unbelievable parallelization support to this fantastic dream engine, so that when run on a quad core, you get an unstoppable monster.
Nice dream, but unfortunately, it would require the cooperation of competing companies / products / engines, so I wouldn't hold my breath.
But then again, without all this rivalry, would chess engines be at the level they are today? would there be enough drive to improve them to the current level without this competition? I think not.
Alas, life is not perfect...
The whole point about winfinder, and the reason it finds tactics faster, is that it's pruning and positional evaluation are significantly worse than the normal rybka. This is not in rybka simply because rybka plays a LOT stronger with its heavier positional evaluations and pruning.
No magic to it. There's no point in integrating it back into the normal Rybka.
As to integrating several engines: Interesting, though they will take a lot of resources away from each other, which means that to make it effective, you need to come up with a REALLY good way of putting together the results from both engines in a useful enough way to find a mistake/inaccuracy by one or the other (a really difficult problem, if at all possible, to solve effectively).
But then, it seems to me, that the engine that found the correct move would have found it even faster (and maybe found an even stronger move!) if given all the computational power available, so you're counting on the fact that the engines will "correct" each other enough times to make it worthwhile.
Of course, the ideal solution would be to take the strong suits of each engine and integrate them into one very strong engine, but that would require the cooperation of competing commercial programs / companies.
I think something like: Rybka's evaluation as a base, coupled with the endgame analysis from another engine (say shredder, if that's the best one at endgames today...if not, just replace with the best), maybe add Junior 10's compensation evaluation,
and for the icing on the cake, let the Zappa programmer add the unbelievable parallelization support to this fantastic dream engine, so that when run on a quad core, you get an unstoppable monster.
Nice dream, but unfortunately, it would require the cooperation of competing companies / products / engines, so I wouldn't hold my breath.
But then again, without all this rivalry, would chess engines be at the level they are today? would there be enough drive to improve them to the current level without this competition? I think not.
Alas, life is not perfect...
Topic Rybka Support & Discussion / Rybka Discussion / Vas,This is how Rybka can become great on tactics!
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill