Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / Search depth in IdEA vs. in infinite analysis
- - By cma6 (****) Date 2020-07-21 16:28
What is the proper way to understand true search depth in IDeA vs. in infinite analysis. For a given amount of computer resources devoted to chess analysis, let us say that in i.a. one searches to 54 plies for a typical middlegame position. Assume also that for this setup and position, in IDeA one can use the following analysis quality settings: 400 seconds AND 45 plies with maximum time of 600 seconds plus "Wait for the next depth". Automatic Tree Expansion is unchecked. 
In understanding true search depth, is is correct to say, "54 plies in i.a. beats 45 plies in IDeA; therefore, don't bother with IDea."?
Parent - - By pawnslinger (****) Date 2020-07-21 23:56
I think you still can use IDeA... because it searches for the best move with a constant quality at each step.  Whereas IA engines use lesser quality at each step.

I don't know how to explain myself better, I apologize.  I hope you can see what I mean.
Parent - - By cma6 (****) Date 2020-07-22 12:20
I do see what you mean: much more thorough search in IDeA.

However, I was getting at another question about IDeA. The analysis settings--400 seconds AND 45 plies with maximum time of 600 seconds plus "Wait for the next depth"--don't they in effect amount to a deeper search, on average, than 45 plies?
Parent - - By pawnslinger (****) Date 2020-07-22 12:32
It all depends on how deep your tree goes... for example, I often use something like 30 ply (it varies a lot depending on the situation)... so just an example.  At 30 ply for each step in IDeA, if my tree extends the critical variation for 30 nodes (or 15 moves), then my overall depth of search is 30+30 or 60 ply.

Of course, the quality of the search, at each step, either IA or IDeA, must be put to the question mark.  I find IDeA can easily be thrown off my a single evaluation that is incorrect.  So I am always careful to examine the critical variation to insure quality.  Of course, mistakes are still made... even combining IA with IDeA, which I often do.
Parent - - By cma6 (****) Date 2020-07-22 13:02 Edited 2020-07-23 15:08
"if my tree extends the critical variation for 30 nodes (or 15 moves), then my overall depth of search is 30+30 or 60 ply".
That means that you have automatic tree expansion enabled. The plus side to that would be that you could have overall depth of search for some lines quite a bit longer than 30 ply.

But what about the downside of enabling automatic tree expansion: very lengthy IDeA sessions where much of the analysis is taken  up with automatically expanded trees, e.g., nonsense lines? For years I had automatic tree expansion enabled by default but have disabled it in the last few months.
Parent - - By pawnslinger (****) Date 2020-07-23 04:09
No.  I do not have to have automatic expansion enabled.  I generate the tree just from IA in the Sandbox.
Parent - - By cma6 (****) Date 2020-07-23 15:12
Pawnslinger:
That is great, the best of both worlds. How do you generate the tree tasks from i.a. in Sandbox?
I create the IDeA tasks by i.a. (up to 60 ply depth) in DB pane and then run IDeA from the IDeA window. Is there any difference in our procedures, other than the seemingly insignificant one that I create the tasks in DB window and you do it in Sandbox pane?
Parent - - By mattchess (**) Date 2020-07-23 15:45 Edited 2020-07-23 17:14 Upvotes 1
I use brute force and do analysis in sandbox, copy lines to the notation window, then paste the PGN into the IDEA project, add all tasks (positions), and run the IDEA analysis when I want to add tasks generated from IA.  I've never managed to figure out anything more sophisticated although I know there are more efficient ways.  But I find just pasting the PGN and then adding all positions works very well and I feel I have visibility into what exactly I am doing :P  Also remember the tree function in Sandbox that lets you add the best line from IDEA from a position into the notation.  I use this to then go to end of that line and run IA and also to run IA from intermediate positions.  So in effect I am manually expanding lines of interest to me using IA and can do that using multiple engines.
Parent - - By cma6 (****) Date 2020-07-25 03:39 Edited 2020-07-26 02:00
Hi Matt:
As for your procedure:"copy lines to the notation window...and add all tasks (positions), and run the IDEA analysis when I want to add tasks generated from IA"; is not this done automatically (adding tasks) in IDeA Project Properties/Common if you have already checked "Insert infinite analysis from linked games", say with a setting like "Min time 60/ min depth 34, tree config IDeA?
  Or are you doing something in addition to the automatically added tasks?
   This part is new to me: "Also remember the tree function in Sandbox that lets you add the best line from IDEA from a position into the notation.  I use this to then go to end of that line and run IA and also to run IA from intermediate positions.  So in effect I am manually expanding lines of interest to me."
   It seems that one can do everything in Database window, while runining i.a.,  that you and PS are doing in Sandbox?!
While right-clicking the top line in the analysis pane, while running i.a., I see various choices like Copy and Insert. Wouldn't Insert be more efficient than Copy/Paste (into notation window)/and Join?
Also, I noticed what you must be referring to in the Database (or Sandbox) Window: Tree/copy best line from tree. That is an interesting concept, especially if one has a well populated tree and the best line is significantly better than the second line. I assume that "best line" means the one on top with best eval for the side on the move?
Parent - - By mattchess (**) Date 2020-07-25 05:18 Edited 2020-07-25 05:20 Upvotes 1
I am sure that linking and adding IA automatically is more efficient - I just haven't taken advantage of that and am embarrassed to admit I have not really explored it.  My approach is probably clunky but it works for me and I feel like I have control and visibility for what I am doing and choosing to insert into IDEA.  Basically I do IA with one engine and paste lines into notation, explore those lines, possibly do IA from a later position, etc.  Rinse, wash and repeat with alternative engines and keep populating the notation window.  This also lets me easily go back to earlier positions and reanalyze after having analyzed from a later position ("backsliding analysis") to explore alternative lines.  Then I paste all of that over into my idea project and add all the positions to be analyzed in IDEA.

The "Best Line from Tree 1" and "Tree 2" depend on how you configured the trees for those functions (little wrench in the corner of the ribbon).  In my case my Tree 1 is IDEA and you could choose one of your IA trees for Tree 2, or CAP, etc.  I basically just use Tree 1.  From a given position inserting the best line from Tree 1 will insert the best (or multiple nearly equal) lines from my IDEA tree so I can go straight to a position of interest or the end of those lines and extend the analysis using IA myself.  You can increase or decrease the number of lines that get inserted by the best line function by adjusting the delta % up or down.  The best line as you said is determined by the best eval for each side to move.

EDIT:  I should add that the trees you can select from depend on what you have set up in your CTG options.
Parent - By mattchess (**) Date 2020-07-25 21:52
OK looking at it I think the big difference with the automation if sending tasks to IDEA is that when the engine changes its eval or drops a candidate move using my method those lines may be lost whereas with the automation they get retained for later analysis in IDEA.  SO I am actually now combining the methods.  Still manually pasting lines at depth to my sandbox so I can explore them and backslide - but now also automatically sending tasks to linked project so that I don't lose lines along the way :P
Parent - - By cma6 (****) Date 2020-07-26 02:07
"You can increase or decrease the number of lines that get inserted by the best line function by adjusting the delta % up or down."
 

Matt: where does one adjust delta%?
Parent - - By mattchess (**) Date 2020-07-26 02:30 Edited 2020-07-26 04:23 Upvotes 1
The little wrench under the insert best line from tree functions opens settings that allow you to indicate what trees to use and there is a field at the bottom of that window to adjust the %

EDIT:  Here is a screenshot
Parent - - By cma6 (****) Date 2020-07-26 18:05
Matt, that's very useful. So if you use 5%, what does that mean when you select "Best Line from Tree 1"?
Parent - By mattchess (**) Date 2020-07-26 22:00
I believe it pastes lines where the end nodes are within 5% on the eval.  I would be a bit careful and start low and then creep up because it could blow up on you if there are too many lines.
Parent - - By pawnslinger (****) Date 2020-07-23 17:55 Upvotes 1
I do not understand the question.

I link the Sandbox to my IDeA tree and the tasks are fed thru from Sandbox to IDeA.  It is a pretty basic Aquarium operation.  There are plenty of tutorials explaining it.  I even saw a YouTube video about it once.  If you have some specific question... maybe I could help with troubleshooting it.

As things progress, you just navigate in Sandbox thru the critical lines.  Pretty straight forward stuff.
Parent - - By cma6 (****) Date 2020-07-25 01:40
Thanks, PS, I think that you have answered my question. It should not matter if one feeds the task generated by i.a. from Sandbox or from DB pane.
Parent - By pawnslinger (****) Date 2020-07-25 05:27 Upvotes 1
I am sure you are correct.  However, I do not use the DB portion of Aquarium... so I cannot do as you do.  I tried it a long time ago, but found it very buggy.  So I use Chessbase for my DB, and copy games between it and IDeA using PGN... then link to the Sandbox to do my move generation.  Probably about the same as linking to the DB function in Aquarium, but that is just an assumption on my part.
Up Topic Rybka Support & Discussion / Aquarium / Search depth in IdEA vs. in infinite analysis

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill