Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / SUGGESTION FOR AQ 2019
- By ANUAR71 (**) [my] Date 2018-11-15 02:56 Edited 2018-11-15 03:31
may be late for any suggestions but I still want to write for knowledge.

1-In the future,  the idea that allow a certain amount of ideas (positions) to be blocked.
for the purpose of creating a book. (create book from idea analysis )
2 or can change or add at opening Book configuration in column weight
- - By pawnslinger (****) [us] Date 2018-11-22 04:05
I am running AQ 2017, so if this has been fixed in the latest release, please let me know.  Thanks.

I am currently analyzing a project with IDeA using +/- 1.50 as the score bounds.  I noticed that a particular task was taking a rather long time to complete, so out of curiosity I clicked on it in the event queue.  The evaluation being reported was 1.90 below the stated project score bounds... but AQ insisted on continuing the analysis until the requested depth or time limit had been reached.  A colossal waste of time, since the move under analysis was out of bounds, and AQ could "see" this as easily as I could.

I guess AQ felt that it must push on with the analysis, in the hopes that the score for the move would improve by the time the analysis completed.  A futile hope.

I think the method that AQ uses to enforce the project score bounds needs to be improved.  It is a huge waste to spend vast amounts of time on a move so obviously out of bounds.

Perhaps the time and depth parameters could be expanded to include another parameter, one that would allow analysis to be terminated once the engine reported score fell outside certain limits.  In effect have a 2nd score bounds, one for move generation (the current situation) and one for move evaluation.
Parent - - By dickie (**) [gb] Date 2018-11-22 08:06
Time tuning (time_tuning.xml in the IDeA folder) tapers the amount of time spent on diverging evaluations. I imagine you have turned it off, as I did, because the formula was too harsh. That would explain why your analysis continues to completetion irrespective of the score deviation. The problem was solved in AQ 2018 with a much improved time tuning formula. I now use time tuning again with only a minor modification to the formula to suit my personal taste. It works well now, and probably as originally intended.
Parent - By pawnslinger (****) [us] Date 2018-11-22 18:25
Thanks dickie.  I had forgotten about that... honestly I cannot recall how I have it set.  But I will check.  I do remember some discussion about it previously though, so you might be right (I may have it turned off).

Still, in all, it would be nice if there was a parameter in the "Analysis Quality Settings" that would allow direct control of the formula... that would alleviate any need for the dev team to worry about it... let the end-user set it, as he/she wish.
Parent - - By pawnslinger (****) [us] Date 2018-11-22 18:44
Here is what my "time_tuning.xml" file has:

============================
<?xml version="1.0"?>
<tune_options first_interval="10">
  <if delta="20" time_coeff="0.7"/>
  <if delta="10" time_coeff="0.8"/>
  <if delta="5" time_coeff="0.95"/>
  <if delta="-5" time_coeff="1"/>
  <if delta="-10" time_coeff="1"/>
  <if delta="-20" time_coeff="0.9"/>
  <if delta="-100" time_coeff="0.8"/>
</tune_options>
============================

I am not sure what all that means... but it doesn't look like I have it turned off.  The way I read it, if the eval is going more than 1 pawn out of bounds then use only 80% of the time??  In my case, I was -1.90 pawns outside of the bounds for the project and I had set my max time to 3600 seconds, so the time limit should have been reduced to 2880 seconds??  Am I reading that correctly?

The table has 8 entires (depending how you count them, I guess), I wonder if I could add an entry like this:
<if delta="-175" time_coeff="0.5"/>

That would probably solve my issue, if I could do that.  Giving a more severe reduction for the more out of bounds eval.  I may try it just to see what happens.  Things are a bit hectic around here now, with Thanksgiving and general holiday prep.  I am too old and frail to do anything, but it is driving me nuts with stress.
Parent - - By dickie (**) [gb] Date 2018-11-22 21:35
Yes, time tuning appears to be on, but you seem to have made some changes to the original settings. The detailed explanation is here http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=26250
Parent - - By pawnslinger (****) [us] Date 2018-11-22 23:39
I may have modified it, but I do not recall doing so.  Brain goes to mush the older I go.
Parent - By pawnslinger (****) [us] Date 2018-11-23 00:06
So if I read that right...

Pseudo Code
---------------
if (WhiteToMove) {
        delta = RootEval - CurrEval
}
else {
        delta = CurrEval - RootEval
}

Kind of hard to wrap my mind around it... but I think that would work.  Maybe not... but close enough so I can use it, I guess.

The explanation given by Dadi was written in 2012 and I am using AQ 2017, so perhaps that is why my settings appear to be different (as if I modded them).

I think it would defintely be simpler if there was (as I suggested above) a separate evaluation score bounds.  Easier to understand and put to use.
Parent - By pawnslinger (****) [us] Date 2018-11-23 01:57
If Dadi is correct, the negative delta values represent "better" than root moves, and the positive delta "worse" than root moves.  Hmmmm... not intuitive.

Am I reading that right?

If so, then why wouldn't I extend the time limits for "better" moves?  Seems like I might want to do that.

If I could set evaluation score bounds, I might set something like +5.00 to -2.00, if White was on-move, and the reverse if Black was on-move.  In other words, I would naturally like to spend more time on promising continuations (for both sizes).  But even now, I don't think I can vary the score bounds based on who is on-move.... be nice if I could though.
Parent - By pawnslinger (****) [us] Date 2018-11-23 02:45
Okay, based on what I read from Dadi, and what I feel I would like to try... here is my proposed setup for this xml file.

<?xml version="1.0"?>
<tune_options first_interval="10">
  <if delta="150" time_coeff="0.2"/>
  <if delta="100" time_coeff="0.6"/>
  <if delta="50" time_coeff="0.7"/>
  <if delta="20" time_coeff="0.8"/>
  <if delta="5" time_coeff="1"/>
  <if delta="-5" time_coeff="1"/>
  <if delta="-20" time_coeff="1.2"/>
  <if delta="-50" time_coeff="1.3"/>
  <if delta="-100" time_coeff="1.4"/>
  <if delta="-150" time_coeff="1.8"/>
</tune_options>

As you can see, I have added some lines and changed the coeff values to lengthen the time limit for "better" moves, shortening more aggressively for "worse" moves.  Please know that I have not tested this yet... heck I might even crash my system.  So this is the first cut.  This is not even "alpha" code... it is just off the top of my head code.  Use at your own risk and Backup backup backup... if you try my example, you are responsible for what happens.
Parent - By pawnslinger (****) [us] Date 2018-11-23 02:53
Okay, I am running the new code now.  It didn't crash and so far it seems to be running smoothly.

But I wonder, what happens if the root node has a value of 0.00??  According to what Dadi wrote, the delta values are percentages... Kind of hard to calculate a percentage of 0.00... right??

Oh well, maybe I read the explanation wrong.
Up Topic Rybka Support & Discussion / Aquarium / SUGGESTION FOR AQ 2019

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill