Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / Automatic expansion of tree in IDeA
- - By cma6 (****) Date 2020-03-10 01:52
Is it possible or even a good strategy to turn off automatic expansion of the tree in IDeA; or should that be done selectively, depending on the position, assuming one can turn it off for particular projects.
Parent - - By pawnslinger (****) Date 2020-03-10 06:37
Many people have discussed this... the option is the first option on the IDeA option panel.  A simple checkbox.

When to turn it on or off is a strict judgement call.  Personally I use it as little as possible, but I do use it.  It is very dangerous to use as it gives many false positive results.  So beware of any sudden change in the project evaluation... it is very likely to be due to a bad choice performed by the automatic prolongation algorithm.

They don't call it "expansion", it is called "prolongation".  You must disable "Automatic prolongation", it is the first option on the IDeA properties panel.
Parent - - By ChiefPushesWood (**) Date 2020-03-10 14:34
Slinger, I apologize for being dense here...

You said:

>Personally I use it as little as possible, but I do use it.


Are you saying that you use Auto Prolongation as little as possible or manual prolongation as little as possible? And, in turn, are you saying the auto prolongation gives many false positive results?

The reason I ask is this would be a clear deviation from your prior methodologies and a recent (and interesting to me) development for you.

Chief
Parent - - By pawnslinger (****) Date 2020-03-10 16:48
Chief:

First, my methodologies change almost twice a day.  Consistency is NOT part of my method.  Judgement and intuition and experience are.

That said, I don't know what "manual prolongation" is exactly.  The program has an option called "automatic prolongation" and I switch it on/off frequently.  So I DO use automatic prolongation, but I am extremely leery of it.  I do not trust it.  As you know , I like to scrub the line of play that I believe is critical.  I do this by using IA and/or the Sandbox to run back and forth thru that critical line.  You know this.  You know the basic form I use to analyze a position.  Doing this prolongates the line of play.  It also adds many alternative lines of play that may need to be investigated.

This method can be use with or without automatic prolongation enabled.  Currently (i.e. always changing, remember) I do not use automatic prolongation (at all) on my large trees.  It just creates a mess.  With way too many blind alleys. 

When I transition out of opening analysis, I create individual trees for each game, and use a very different approach.  Greater depth of analysis, greater use of Custom Tasks (I almost constantly use Custom Tasks to supplement Aquarium's alternative generation, which is pathetic).  And less use of IA or Sandbox (still used but not nearly as much)...  As I cut back on IA and Sandbox, I increase the use of automatic prolongation ... with much anxiety and worry.... it is not a set and forget option.  Not if you hope to play at the highest levels anyway.  And that has always been my goal... to compete at the very highest level that I have the ability to reach.  At present, I think I have reached that level (as I have in the past), so I am looking for ways to improve and possibly create a breakout (as I have done before).  I absolutely hate when my play is not improving and I am currently struggling with just such a period... no matter what I have done, over the last year or so, my play has plateaued, so I make changes in the hope of doing better.
Parent - - By cma6 (****) Date 2020-03-10 17:06
PS:
So you for the most part have "automatic prolongation" set to Off?
  Among the pre-packaged Custom Tasks, do you have any favorites?
Parent - By Ghengis-Kann (***) Date 2020-03-10 17:20
You could try a few from the tree strategy library I posted up a few months ago.

3X5 with 5 ply depth and 3 alternatives at each ply is one I use a lot.

Manually moving around an IA engine and using it to seed IDEA is probably the most efficient way to generate a high quality tree if you are willing to put in the time.
Parent - - By pawnslinger (****) Date 2020-03-10 17:41
I have tried several of the new strategy methods for Custom Task.  But honestly I do not use any of them.  I use the original Custom Task, where I use, normally, number of plies set to 1.  The other parameters can vary quite a bit.  Depending on the situation, I can even set the depth to be extremely shallow, to get a quick spread of alternatives added, in which case I go back afterwards and manually use more Custom Tasks to prolongate each "interesting" alternative... in this way I try to use Custom Task in 2 different ways, first to get a lot of different alternatives into an interesting position and 2nd to add depth of analysis to those (perhaps many) that I think are interesting.
Parent - By cma6 (****) Date 2020-03-10 17:46
Thanks to PS, Ghengis, and CPW for the suggestive "ideas". I have tried a few of Ghengis' Custom Tasks with interesting results.
Parent - - By rov (*) Date 2020-03-11 16:34

> I do this by using IA and/or the Sandbox to run back and forth thru that critical line.


I have heared about this method, but I didn't see any example how it is work. Do You have any position that in normal engine's evaluation  can't find the right move, and after run back and forth it can do it ?
Parent - By pawnslinger (****) Date 2020-03-11 18:27
You make me smile.

You can pick just about any chess position, except those that lead to a forced mate.

The critical variations of most positions need to be analyzed.  I use Aquarium to save my analysis and use it as a starting point for my next session after my opponent moves.  So scrubbing the critical variations helps me, even when my current move choice is obvious.
Parent - - By pawnslinger (****) Date 2020-03-11 20:06

> the right move


In many positions such a thing does not seem to exist.

Rather, only the move you prefer... or are forced to make.
Parent - - By rov (*) Date 2020-03-11 23:01

> only the move you prefer...


Is it possible that my move can be ok if it is worse 0,5 than engine move on root position ?
Parent - - By pawnslinger (****) Date 2020-03-11 23:49
It is not only possible, but in the Kings Indian some moves such as that, or even worse, can actually be winning!!  I have seen moves as bad as -0.75 actually be the winning move!!!  My mind is mush at the moment, I can't remember the name of the variation... but it leads to very dynamic positions (where evals change often and by large amounts).

By -0.75, I mean the engine thinks that it is that amount relative to the root node, i.e. it thinks it is a very bad move... but it is actually an excellent move.

A couple of years ago, in this forum, a position was presented from a ICCF GM game, where the winning move was thought to be so bad that Stockfish didn't even find it in the top 15 or 20 alternatives.  The person who presented the move here was not the author of the move, but a person who saw the game and wanted to show it here in the hopes of determining how the move had been found.  The move was made by Dadi (I am not sure I should reveal his last name here).  He is a prominent Aquarium user.  I do not know how Dadi found the move, nor what tools he used, but the move was brilliant and I do not think anyone here was able to figure out how he worked it out.  As you can imagine, Dadi is a very creative player and I think his judgement is superior (to most any player, man or machine).
Parent - - By ChiefPushesWood (**) Date 2020-03-12 13:38 Edited 2020-03-12 13:41

>My mind is mush at the moment, I can't remember the name of the variation...


I could be wrong but I think you're referring to the Mar del Plata.

>The move was made by Dadi (I am not sure I should reveal his last name here).  He is a prominent Aquarium user.  I do not know how Dadi found the move, nor what tools he used, but the move was brilliant and I do not think anyone here was able to figure out how he worked it out.


Dadi's full name is all over the forum and his articles.. I don't think you have any worries :).

I remember an article that Dadi posted and referenced a pretty incredible move that wasn't found by IA or IDEA by themselves.. only with the players' (Shahar Tzafrir (Highendman)) substantial interactive analysis was he able to find the move 27. d5 in this position “2rnb1k1/4r1p1/1p1qp2p/p3N3/1nPP2p1/1Q5P/1B1RBP2/3R2K1 w – – 27”. You can read Highendman's own analysis in this forum link.

There are PLENTY of opportunities to improve on 'computer moves' if one is willing to put in the effort and work. Interactive analysis is ALWAYS, and I mean ALWAYS, better than either IA alone or IDEA alone.

Chief
Parent - By pawnslinger (****) Date 2020-03-12 15:53
Thank you, Chief.,  Yes, Mar del Plata!!  I don't know why I could not remember, but at 74 lots of things seem to fade.

Well, then... without further ado, GM Dadi Jónsson is the player in question.
Up Topic Rybka Support & Discussion / Aquarium / Automatic expansion of tree in IDeA

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill