Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Is Monte Carlo, a great developement?
- - By staylor (***) Date 2009-10-20 01:40
Is Monte Carlo a new system of attempting long term planning, and
Will it be greatly developed in Rybka 4?
Parent - By Permanent Brain (*****) Date 2009-10-20 04:56
Monte Carlo is not a planning system. It's random based. Google to learn more about it.

P.S. Rybka 2.3.2a supported a similar concept already, the "Randomizer". I need to admit that I did not pay enough attention to it, but it seems very interesting. I think it can really help in analysis (preparation), to provide useful, good alternatives. Maybe for other purposes too.
Parent - - By M ANSARI (*****) Date 2009-10-20 05:25
It is very useful in finding obscure moves in dead positions that can turn the game around.  Also useful in finding "only" saving moves where everything looks totally lost.  At the moment it can only be used successfully for analysis as the time and hardware required are huge, but in the future it could be a useful thing to have in an engine where hardware is overly abundant.
Parent - - By staylor (***) Date 2009-10-20 13:45 Edited 2009-10-20 13:49
What you have said it is good for, is VERY good. Perhaps it is also good for long term strategy, whidh until now has eluded chess playing computers? If all this is true, then perhaps, if it might be incorporated in Rybka 5, at tournament timings, then it would be a shame that it is not already put into use for Rybka 4. (And who knows? maybe it is only on plan for Rybka 6? Perhaps Rybka 5 is meant for certain perfection of things which have been discussed, and it needs to wait for R6 for monte carlo, and in that case, after all the bug corrections and other improvements, it might wait for Rybka 7 to be the final Rybka? If so, we might be looking at year 2015 [I don't think i'll still be hanging around in chess forums till then!]).
Parent - - By M ANSARI (*****) Date 2009-10-20 14:22
It has not eluded chess at all and many are using it quite often in analysis.  I think there was a position on talkchess where I used it to prove that one position was not lost and that there was an obscure defense.  I used Monte Carlo analysis overnight to generate a large database of positions and to find the move that draws, which was impossible to find by normal chess engine analysis.  This usefulness is well known in case of a chess entity such as a cluster, where such hard to find moves could be used in one of the many nodes of the cluster.  This way all corners of chess search would be covered ... but this is still very far away.  Maybe in 10 years where the cluster would have say 10,000 cores, it could use 100 cores exclusively to generate in real time Monte Carlo analysis.
Parent - By staylor (***) Date 2009-10-20 16:34
From what i understand, it could even be possible to simulate the monte carlo idea with old programs like genius 3, as follows:  Genius 3 could be set to autoplay (playing against itself) on instant, and can play whole games, from any position, in a matter of seconds. The moves even at this speed, coming from genius 3 were not that stupid. And many variations can be tried out from a certain position, and tested on genius 3 playing like this, and each time you could make g3 play it more than once to see if the results are consistently one way or the other.
In this way, you COULD say that genius 3 looked 100 plies down the line in SOME variations.
So perhaps this would be like human assisted MC.
Parent - - By Kapaun (****) Date 2009-10-20 15:07
The hardware isn't quite there yet to make MC a really useful analysis method. The most powerful analysis feature at this time is probably Interactive Deep Analysis in the Aquarium GUI. Though you will need to learn how to use and to interprete it.
Parent - By staylor (***) Date 2009-10-20 22:00
I hope to meet it in the upcoming rybka 4, i didn't get rybka 3.
Interpreting how to use computer analysis, i'm quite good at (as long as I can figure out what it is saying about which position, and in which context).
Parent - By Dadi Jonsson (Silver) Date 2009-10-20 22:19
You might want to look at the experience of someone who actually tried to use it for choosing moves in a game he played: Using Monte Carlo Results to Choose Moves.
Up Topic Rybka Support & Discussion / Rybka Discussion / Is Monte Carlo, a great developement?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill