Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / Strange Aquarium issue
- - By NeoSteinitz [us] Date 2017-05-15 20:18
Hi all!

I'm a long time lurker, but recently ran into an odd issue with Aquarium that I thought I'd post here.

I have a somewhat large IDEA tree that I've been constructing over the last few months. I really just want it to spend a similar amount of compute on each position, so I've been using 300 seconds AND 1 ply (I don't really care how deep the search is), and I don't want it to just stop at 300 seconds if it's in the middle of an iteration, so I check the "Wait for next depth" option.

I also don't use automatic tree expansion. I use IDEA basically to persist analysis results and explore variations I hand-pick.

This has been working well for months, and I occasionally spot-check the project stats to make sure they make sense (basically, average time per task >300 seconds), and that's been the case since the beginning.

However, starting a few days ago, while manually exploring some lines and adding some tasks, I noticed the tasks were finishing very quickly, usually by 90 seconds, and sometimes as quickly as 20-30 seconds.

I've combed through all the settings and documentation I can find to see why a task might finish before meeting the time and depth requirement, but haven't found anything that explains it.

Has anybody run into this before?

Thanks in advance!
Parent - - By dickie (**) [gb] Date 2017-05-16 07:05
Perhaps you have not adjusted your time tuning parameters? See http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=26250
Parent - By NeoSteinitz [us] Date 2017-05-16 14:49
Ah, thanks!

I'll have to wait to get back to that machine before I can check, but that is almost certainly the case.

I didn't really start noticing until I began expanding analysis of an interesting and complex line that doesn't have a particularly good eval, so that lines up perfectly.

Thanks again!
Parent - - By pawnslinger (****) [us] Date 2017-05-16 22:11
Yes, you can control IDeA time tuning... but quality result is somewhat uneven.  I prefer to go for a constant depth instead of a fixed time... so I give more than ample time limits to IDeA and a fixed, obtainable, depth limit.  A common setting I use would be time of 300+ seconds and depth of 26+.  In this way the engine quits analysis when depth is reached, or when engine determines it cannot reach the depth and time expires.  Sometimes the engine will give up early when it sees that there is no chance to get to the next ply in time... but I try to give enough time so that doesn't happen often.
Parent - By NeoSteinitz [us] Date 2017-05-17 14:58
Pawnslinger,

Thanks!

Yeah, that was what I ended up doing as well when I initially discovered the "problem" (or as I now know, feature).

I just changed the ply requirement to 30 to make sure I at least got that, since I noticed that the ply requirement was always honored.

I still haven't had time to access that machine, but when I do I imagine I'll just effectively disable time tuning by setting first_interval extremely high.

Thanks again to both for your help!
Up Topic Rybka Support & Discussion / Aquarium / Strange Aquarium issue

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill