Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / question about delayed analysis
- - By Kapaun (****) [de] Date 2007-10-03 13:11
I just went to the Convekta site and downloaded a manual of an older CA version in order to get an idea how the delayed analysis (which is supposed to base on the background analysis of CA) might work.

It's still not quite clear to me. The main question is: If you have a position you want to be analyzed - can you tell the GUI: Take Rybka, analyze the position for a certain amount of time (or to a certain depth), then take the five best moves. Start with the very best, analyze it again to a certain depth, then take the three best answers, go on analyzing the best answer, take the two best answers for that one, and so on. If you finally analyzed all moves of one line, go back to the first move not yet analyzed and start the process again.

So you really would get an automated analysis of the position split up in several lines according to a predefined system. Actually more or less what the DPA does. Could the delayed analysis do this? If it can do more that's fine, but it should certainly be able to do this.
Parent - - By Felix Kling (Gold) [de] Date 2007-10-03 16:26
I guess it's much better to do that with the randomizer, since it should produce better evaluations, because it gives also good evaluations for positions, where the engine gives no good evaluations normally (e.g. gambits).
Parent - - By Kapaun (****) [de] Date 2007-10-03 17:51
Actually, no. I've tried the randomizer when it first came out - but it's not what I am interested in. Which is: Getting a branched analysis of possible lines which is a) of high quality (using high depths and using only a few best moves, not all or at least many moves) and b) going forward only a number of moves, like 5 to 10 or so, and not playing a whole game).

The randomizer might be useful at high depths, too, but it is taking way too much time then, because it's doing complete games. And it is way too complicated to use, too.

So, the DPA feature of the Fritz GUI is more or less exactly what I'm looking for. Not that it couldn't be improved (maybe more flexibility, possibilities to interfere manually etc.)... But I would appreciate a new GUI which has this feature or a similar, maybe even better one. The delayed analysis looks a bit like that, but I'm not sure yet.
Parent - - By buffos (Silver) [gr] Date 2007-10-04 19:24
as an old time CA user i would say the background analysis is the best analysis tool one could have.
It actually helps you to quickly build the analysis tree of a position.

For example you have a variation. You send it to background analysis, checks the variation move for move (based on parameters you define).
You can have the variation analysed backwards (extremely usefull for deep variations that the engine does not see a combination and you need to find a possible defence)
autoplay a certain position and much more. A very powerfull tool, not very well documented i would say, but TOO powerfull.
Parent - By Kapaun (****) [de] Date 2007-10-04 22:07
Hm, I see. Can I do the things I mentioned above in the thread opening with background analysis?
Parent - - By Zruty (*****) [ru] Date 2007-10-05 09:27
Well, the delayed analysis in the upcoming GUI will not be based on CA approach. It is just intended to work in a similar way. Actually, it offers less features than BGA in Chess Assistant.

The delayed analysis doesn't work the way you described. It needs more human interference: you specify manually what positions do you want to analyse.

There are now 3 options of a delayed analysis:
- Analyze current position
- Auto-play N plies from current position
- Find N best moves from current position

So, typically you need more than one iteration "human analysis-delayed analysis" to get a deep insight into a position.
Parent - - By Kapaun (****) [de] Date 2007-10-05 11:28
Hm... I see. Sort of bad news. Speaking of CA: Would the background analysis of CA be able to do that?
Parent - By Zruty (*****) [ru] Date 2007-10-05 11:49
I don't know exactly, but I suspect it isn't. Maybe buffos knows more
Up Topic Rybka Support & Discussion / Aquarium / question about delayed analysis

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill