Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / Which engine for IDeA?
1 2 Previous Next  
- - By Checks Date 2017-05-14 20:37
Dear all,

is there a consensus which of the contemporary engines is most suited for IDeA?

At the moment, I use ASMFish/Stockfish which is very fast, but I'm not sure whether it would be better to use Komodo or Houdini for IDeA and maybe ASMFish/Stockfish as an addition in IA, sending lines to IDeA. Maybe there is a logical solution to this question, considering how the different engines work?

What are your preferences?

Kind regards

Checks
Parent - By pawnslinger (***) [us] Date 2017-05-15 07:33
A tough question... in general I believe the engine with the best rating should be used in IDeA.  A very small edge in rating can be magnified to a large difference in the performance of IDeA.  However, with that said, slower, more "positional", engines do have their place -- probably in IA or Sandbox.  I tend to go by the TCEC listings.

Setup of parameters in IDeA can also have a big impact... possibly greater than rating.  But that has yet to be shown clearly.  The horizon effect that all engines suffer from is the biggest obstacle to successful use of IDeA... and that is somewhat mitigated by a stronger, more highly rated, engine.
Parent - By Carl Bicknell (*****) [gb] Date 2017-05-15 18:23
Go for the top rated engine for IDeA and something different for sending lines to it (Houdini / Komodo / whatever)
Parent - - By Checks Date 2017-05-15 20:38
Thanks for your answers.

The three top engines are pretty close in rating at the moment. I figure you mean ASMFish/Stockfish by the top rated engine?

Just an idea, correct me if I am wrong:
ASMFish usually has greater depth than Komodo given the same computing time. If I use Komodo as my IDeA Engine and ASMFish as my IA engine, sending lines to IDeA, ASMFish will send more new positions to IDeA than if I would use them the other way round, given the same time. Is that true? Does it matter?

Concerning the IDeA parameters, do you have any hints for me?
Parent - - By pawnslinger (***) [us] Date 2017-05-16 16:38
No hints about the parameters.  They are a continual work-in-progress.

As to the feeding of positions into IDeA... recently I have come to the conclusion that more is not always better.  The big deal of IDeA is that it tends to spend A LOT of time going down "blind alleys".  I now try to avoid that at all costs.  In other words less but higher quality is ALWAYS better.  Analyzing the better moves sooner in the process is ALWAYS better.  I have been playing around with Sandbox settings lately of 3-5 minutes min time and 30+ min depth.  Usually this generates very few candidate moves, but depending on the minimums you use, you can generate better quality moves.

In my opinion, the weakest part of IDeA is alternative generation.  Left to its own, IDeA sometimes has a very hard time even finding the key move.  So I spend a lot of time on alternative generation, to try and make sure that the key move is at least being considered.
Parent - - By mist (**) [gb] Date 2017-05-28 09:23
I know you are a BIG user of IDeA, but I would like to ask you if you have been using the new Tree Generation Strategies. and if so do you find that of any use?
Parent - By Radost (***) [de] Date 2017-05-28 17:33
I use TGS. I want to know e. g. what is the evaluation of a variation position after 10 moves ... :-)

Best, Frank
Parent - By pawnslinger (***) [us] Date 2017-05-28 17:54
Hi Mist:

As a rule, I do not use them.  I have tried them, but find seeding IDeA thru the Sandbox is better for me... just my opinion, I could be wrong.  I have disable the "root" and other automaticly run TGS.  They are interesting, but seem to do a lot of unnecessary work... IMHO.
Parent - - By Checks Date 2017-06-04 21:20
I have another question concerning IA vs. IDEA:

When using IA with the same engine as in IDEA, isn't it better to send the lines directly to the tree instead of creating new idea tasks?

Especially when using Multi-Variation Mode, I think this is much faster and avoids double processing work.

Is that correct?
Parent - - By pawnslinger (***) [us] Date 2017-06-05 04:44
Yes, but at least in the case of Stockfish, there are bogus evals generated that way also.  For some reason Stockfish sends +/-300 as a bogus eval.  It later corrects it, but Aquarium does not properly record it, until a minimax overwrites the bad eval... which can sometimes take a LONG TIME.

So I stopped doing it and requested a bug fix quite awhile ago -- which I think was ignored.
Parent - - By Ghengis-Kann (***) [us] Date 2017-06-07 16:54
I have experimented extensively with both automatic IDEA expansion and tree generation strategies.
A few of my custom strategies have been posted on this forum in case you would like to try them or just understand better how that system works.

My current assessment is that both automatic expansion and tree strategies generate trees that are bloated with garbage.
The main problem with any form of chess analysis is that the final evaluation of every position is back propagated from a shallowly analyzed leaf node.
The quality of your analysis tree is entirely dependent on finding and fixing bad evaluations out on the leaves.

My current method is to seed the tree with database games, and then use my best computer to run infinite analysis and send all of the positions to the IDEA queue where they are processed by the other computers.

Infinite analysis is better than automatic expansion at identifying the best candidate moves, and sending all of the positions to IDEA provides a reality check on the evaluation of each position. It also creates a permanent structure that I can look at and build upon.

If you don't have multiple computers you can do the same thing with one engine per processor on a single computer.

I have also seen this bogus +/- 300 evaluation but did not realize it is coming from Stockfish.
I bought the Houdini Pro version of AQ2017, but the license is only good for 4 computers.
There are 4 strong computers on my home network, but I have an old Dell laptop that I stuck on there with Stockfish due to an insane greed for processing power.
I will take that offline and see if the issue goes away.
Parent - - By pawnslinger (***) [us] Date 2017-06-07 23:53
I agree with your analysis.  Another interesting, and easy, technique to help control bad leaf nodes is to use the "Min Depth" parameter.  Getting the right value for this parameter is an art.   When I want lots of possible moves generated, I usually set Min Depth to 5-10, this gives lots of depth and eliminates a lot of the noisy moves.  When I want only the best alternatives, I set Min Depth high... I have experimented with various depths, and the higher the better seems to be a pretty true statement.  However depth of analysis tends to suffer in direct proportiuon.  So you can have quality or depth, but not really both, without spending the necessary time to plow thru the variations.
Parent - - By Ghengis-Kann (***) [us] Date 2017-06-08 02:43
The min depth is a useful parameter, but there doesn't seem to be any way to stop IDEA itself from generating bad leaf nodes as "alternatives".
It will even do this if you send positions from the database with the "Generate Alternatives" box unchecked.

Have you found any way to prevent this?
Parent - - By pawnslinger (***) [us] Date 2017-06-08 06:35
No.  I believe that the method used to generate alternatives is basically flawed.  Disabling automatic prolongation will stop it, and then it is all manual... not very pleasant either.  I wish that Aquarium used the normal engine multi-variation mode to generate alternatives at depth (user specified, both depth and number of alternatives)... this would be a clearly superior method.  I am not sure why Aquarium (IDeA) does not use this method.  It is difficult to tell precisely how IDeA generates alternatives, but it is not "at depth"... alternatives are chosen at a lower depth and only evaluated later at full depth.  This leaves many good moves to be looked into only much later after MUCH time has been spent on lesser moves.

This is why I spend a lot of time and effort generating alternatives in the Sandbox using a large min depth and multi-variation mode.  Usually something like min depth around 30-ish and 5-12 variations.  This generates a low number of quality candidate moves.  Always the best way to begin any leaf node - fill it with the best candidate moves ASAP.  Then if IDeA adds its own alternatives later, they will get less attention.
Parent - - By Ghengis-Kann (***) [us] Date 2017-06-08 13:50
That is a good strategy for trying to find the truth in the position.

As an OTB player I need to be prepared for sub-optimal responses, so my trees are wider and shallower than yours.
I use anywhere from 1 to 4 variations and a min depth of 1.
That sends everything the engine even briefly considers to the IDEA queue, which provides me with guidance on how to respond to the "human moves".

Disabling automatic prolongation does not completely stop alternative generation.
When I send database games that have no annotation to IDEA it still generates alternatives on some of the moves in a seemingly random fashion.

I also have no idea what you mean by "Then if IDeA adds its own alternatives later, they will get less attention", since all of the leaf nodes are weighted equally.
If they are outliers then that is what will propagate back to your root nodes.
Parent - - By pawnslinger (***) [us] Date 2017-06-08 17:31
=====
"I also have no idea what you mean by "Then if IDeA adds its own alternatives later, they will get less attention", since all of the leaf nodes are weighted equally.
If they are outliers then that is what will propagate back to your root nodes."
=====

For example, take my Nimzo-Indian tree, it is quite large, so that if I turn on "automatic prolongation" in the early stages of the game, I usually get more than 50,000 alternatives generated in a single pass.  So I have to restrict the generation process, or disable it entirely.  At the extreme, I allow the prolongation, but limit the process to +/- 0.17 (this figure must be adjusted or tuned to get a reasonable number of alternatives generated).  With that tight a limit on prolongation, not many "outliers" are considered... generally only very pertinent moves enter into the process.

However, in a situation where the tree is not as full, the point restriction should be less restrictive, more normal would be +/- 2.00.  In this case, it is imperative to have the tree well seeded before prolongation, because the prolongation will naturally concentrate on the moves and nodes that already exist,  Yes, as new alternatives are added automatically by IDeA, they will be less pertinent (perhaps even very poor), but the tree will already be seeded with many more good moves, so the actual prolongation will be dominated by those sheer numbers.  In other words, when IDeA goes looking for nodes to prolongate, it will have more pre-seeded nodes to work with.  It is absolutely still possible for a poor move to move to the top of the node list and dominate the minimax, but this generally won't las long, if the treee is seeded correctly.  What I am trying to say is... the seeding process is not a one and done!  I have to continually comb thru the tree and make sure key nodes are properly seeded - it takes a lot of time and effort.  Not actually sure it is worth it, but over time your trees begin to mature, and your play improves markedly.
Parent - - By Ghengis-Kann (***) [us] Date 2017-06-08 18:06
Like any good gardener you must tend to your trees by hand...

The huge number of alternatives generated is why I stopped using automatic prolongation.
Instead I control a single mobile "root node" by hand that uses infinite analysis to find the alternatives and sends all of the positions to IDEA.
This creates a permanent record of my work and improves my own understanding of the positions.

Of course we will do different things because my interest in Aquarium is to use it as a teacher, while you are using it for competition.
Parent - - By pawnslinger (***) [us] Date 2017-06-08 22:25
Speaking of competition... I am hoping to complete my CCM this year.  CCM is ICCF's title for Correspondence Chess Master.  A title that is nearly the lowest one they offer, and it has not been around very long.  I earned my Master of Correspondence Chess from USCF around 10 years ago (don't recall the exact date of the award).  I got that after a 4th place finish in the Golden Knights.  Computers were not allowed by the USCF, so it was a big change in my approach when I started with the ICCF.

I have tried OTB, but I just don't have the nerves for it.  The last time I tried OTB, I cracked a tooth and had to have a root canal... needless to say I could not even complete the event.
Parent - - By mist (**) [gb] Date 2017-06-09 08:34
As someone who managed to get/earn the CC Expert and CC.Master titles this year can I wish you well in your quest ...but although both titles are new, it is still not that easy to earn two norms in events in this day and age when everyone and his cat are using engines.

>The last time I tried OTB, I cracked a tooth and had to have a root canal...


Now that is what you call stress!!  The last time I tried OTB play at a club (many a year ago) the thing I noticed was the noise..which I found off putting.
Parent - By pawnslinger (***) [us] Date 2017-06-09 15:52
Thank you, mist.  I think I will be able to garner 3 norms for CCM, possibly 1 CCE and 2 CCM almost at the same time.  I am currently playing in a section of a Master Promotion event, which I may earn a CCE, but chances are good for a CCM.  Then I am playing in a final of the Veteran's World Cup (I am 71, soon to be 72), and I am very sure I will earn a CCM, perhaps even an IM norm in that event.  Lastly I just started a section of the semi-final of the WCCC, and the ratings of my opponents are so high, that the CCM norm is low enough I think everyone in the event will surpass it.

And lest I forget, congratulations on your new titles!  I agree with you, with more than 400 ICCF games under my belt, I can testify that no title norm is easily earned.  I recently told my local ICCF Secretary that I hoped to earn ANY ICCF title before I passed.
Parent - - By pawnslinger (***) [us] Date 2017-06-09 16:13
I see that you are located somewhat in the north... are you in Scotland?  During my youth, I spent a couple of years near Felixstowe, on the Suffolk coast.  I was in the US Air Force at RAF Bentwaters.  Around the time I was there they had that UFO scare.  Never did get to the bottom of that one.  But I had a lot of fun playing Chess for the Woodbridge Chess Club.  I got to see a lot of pubs that way <grin>.
Parent - - By SchackMatt (**) [se] Date 2017-06-10 15:32
Did you ever had Haggis? Then you comprehend their taste of whiskey!
Parent - By pawnslinger (***) [us] Date 2017-06-10 20:10
No, never Haggis.  But Kidney Pie... I was amazed that I actually liked it a lot!!  While I was in England, I got to visit the southern coast twice, for Chess tournaments (I played both CC and OTB in my youth).  Those are very fond memories.  My worst and best memory is getting the tar beat out of me by some English GM in an event called the Barnstaple Open (I am not sure I spelled the name of that town correctly).  Anyhow, it got my first game published as a result.  I forget his name, but I got crushed, the game was published because of him (not me).  In the notes published by the event organizer, they said the favorite had an easy game <grin>... and he did... I played a very weak defense to e4.
Parent - - By mist (**) [gb] Date 2017-06-10 15:53 Edited 2017-06-10 15:55

>are you in Scotland?


No...England.

> (I am 71, soon to be 72),


Already there!

>more than 400 ICCF games under my belt


300 more than yourself.

>Then you comprehend their taste of whiskey!


Horrible stuff...need to drink loads of it to make the taste go away!!
Parent - By pawnslinger (***) [us] Date 2017-06-10 20:50
I found the game... it was in the Chessbase database.  Played in 1971.  Chessbase shows his first name as George, but I always thought it was "Geof" or something like that.  Anyhow, GM Botterill had an easy time of it that night.  <Grin>

[Event "Barnstaple op"]
[Site "Barnstaple"]
[Date "1971.10.01"]
[Round "1"]
[White "Botterill, George Steven"]
[Black "Meiners, Edwin"]
[Result "1-0"]
[ECO "B09"]
[WhiteElo "2408"]
[BlackElo "2860"]
[PlyCount "59"]
[EventDate "1971.??.??"]
[BlackTeam "USA"]

1. e4 d6 2. d4 Nf6 3. Nc3 g6 4. f4 Bg7 5. Nf3 O-O 6. Be2 c5 7. dxc5 dxc5 8.
Qxd8 Rxd8 9. e5 Nd5 10. Nxd5 Rxd5 11. Be3 b6 12. Ng5 Rd7 13. Bf3 Bb7 14. e6
fxe6 15. Bxb7 Rxb7 16. O-O-O Nc6 17. Nxe6 Bf6 18. Bf2 Kf7 19. Rhe1 h6 20. g4 g5
21. f5 Ke8 22. Bg3 Rc8 23. c3 Nd8 24. Be5 Nxe6 25. Bxf6 exf6 26. Rxe6+ Re7 27.
Rxf6 Rd8 28. Rxh6 Rxd1+ 29. Kxd1 Kf8 30. Rf6+ 1-0

Parent - By cma6 (****) Date 2017-06-13 02:26
To see how strong engines have made it very difficult to achieve norms, consider WC Final 29 with 13 GMs of 17 participants, some of the GMs very highly rated.
It is quite possible that Dronov will be the only 1 of the 13 GMs to earn a GM norm and that just barely.
https://www.iccf.com/event?id=52852
Parent - - By SchackMatt (**) [se] Date 2017-06-10 15:37
Why not stick to infinite analysis? What does IDeA contribute? I am currently very concerned - do I waste o lot of time or how does IDeA contribute? Three questionmarks - it reflects my own current status!
Parent - - By Ghengis-Kann (***) [us] Date 2017-06-10 17:54
Even if you exclusively use infinite analysis there is still a very important use for IDEA.
It is what I have been describing in this thread.

Sending all of the positions to IDEA creates a permanent record of the analysis that you can review and build on.
Parent - - By pawnslinger (***) [us] Date 2017-06-10 20:01
Ditto!!  You are exactly correct.  I think that the "historical" record of my games/analysis is very valuable.  It builds over time and as it does, my game improves.
.
Parent - - By Checks Date 2017-06-10 21:03 Edited 2017-06-10 21:13
I agree 100% that Aquarium delivers great value in terms of saving the analysis in the min-maxed tree structure.

What I still don't understand (besides the "Stockfish 300" bug):

Why not write the results of IA directly in the tree?
What is the benefit of sending them to the IDEA queue which consumes a lot of processing time.
Parent - - By SchackMatt (**) [se] Date 2017-06-10 22:49
This is also one of my concerns, as long as I don't use different engines in IDeA and Infinitie analysis.

I do appreciate the historical record in Aquarium.
I am using Infinite Analysis more and more and it seems I eventually will end up using it exclusively and only using IDeA when analysing interesting games and other external analysis e.g from books. Since a week ago I am actually using the same engine in IDeA and Infinite Analysis. When I want a second opinion I start Infinite Analysis again with another engine.
Parent - By pawnslinger (***) [us] Date 2017-06-10 23:28
The main benefit is the depth of analysis.  For example, let's say that I send all positions from Stockfish to IDeA, after 3 minutes of Stockfish analysis (or whatever), then some of those moves will have a lot of analysis, and some will have very little.  Stockfish front-loads analysis on the early moves, the later moves (let's say the last 16 ply) receive very little.  Whereas IDeA has a level amount of analysis throughout.

To prevent this you can use the "Min Depth" parameter and set something reasonable, such as 16 ply in my example above, this would ensure that all moves forwarded to IDeA's would have at least that much analysis, BUT still it would be uneven... just not as bad as sending ALL moves no matter how deep the analysis.
Parent - - By cma6 (****) Date 2017-06-13 02:15
" I spend a lot of time and effort generating alternatives in the Sandbox using a large min depth and multi-variation mode.  Usually something like min depth around 30-ish and 5-12 variations."
  So when you send i.a. evals to IDeA, you typically are running i.a. in multi-variation mode?
Parent - - By pawnslinger (***) [us] Date 2017-06-13 06:48
Yes, I use multi-variation mode in the Sandbox.  But I send the moves to IDeA, not the evals.  I tried sending evals at one time and it caused too many problems.
Parent - - By cma6 (****) Date 2017-06-13 11:56
Your are sending moves (not evals?) from i.a. of a position in a game linked to a particular IDeA project, whose IDeA project properties/Common/ dialog has a check box for "Insert Infinite Analysis from linked games". 
I always assumed that meant the moves/positions together with the associated evals are sent. I don't see how you separate out sending the moves but not the associated evals.
  Also, mechanically how does one send these moves/evals? While i.a. is running in Sandbox or Database pane, I right-click in the analysis pane on one of the lines shown and then click on "Send line to IDeA: game X". But that sends only one line at a time; so I assume that if you are running 2 variations, it takes two right clicks for you to send the lines to IDeA?
Parent - - By pawnslinger (***) [us] Date 2017-06-13 18:05
A good question.  In older versions of Aquarium (I am running 2017) there used to be project options that controlled whether just moves or moves+evals were sent during linked analysis.  I cannot find that option in 2017 anymore.  Since it only sends the moves, that is what I have done for a long time, I did not notice that the other options were removed or moved someplace I cannot find.  I am completely confused - but not really concerned, as it performs in the manner I wish, sending only moves... the evals are done by IDeA.

The mechanism is automatic, I just link the project to the Sandbox and bingo, the moves are automatically transfer as the Min Depth and Min Time criteria are met.

Of course, these criteria apply to IA as well... i.e. the moves are automatically transferred once the criteria are met.
Parent - - By Checks Date 2017-06-13 19:59
It's in the (advanced) engine options of each engine.
Parent - By pawnslinger (***) [us] Date 2017-06-13 23:31
Thank you.  I guess I forgot where it was -- for some reason I thought it was associated with the project options.  Since I don't work with this particular option very often, I just plain forgot where it was!!  As I get older my memory seems to begin to suffer.

Thank you for the help.
Parent - - By cma6 (****) Date 2017-06-14 00:48
Each engine has 3 choices:
  a) Send positions from analysis line into IDeA queue.
   b)   Add moves & evaluations to IDeA tree.
   c) Add analysis to IDeA tree and generate alternative tasks.

For years, I have unthinkingly been using c), probably because of something that Dadi said or wrote.
It seems to me that b) should be best, since one is doing much deeper ply analysis in i.a. than in IDeA and one wants these good evals sent to the IDeA tree.

Which one of these 3 choices do you guys use and why?

Thanks, CMA
Parent - - By pawnslinger (***) [us] Date 2017-06-14 14:55 Edited 2017-06-15 02:15
I always use (a) Send positions from analysis line into IDeA queue.  I have tried (b) with not much good luck.  I think I tried (c) but found that it generated too much computation for my taste.  I could be wrong, it has been a long time since I tried it.
Parent - By cma6 (****) Date 2017-06-14 15:30
Pawnslinger:
I  am a heavy user of IDeA, although not so heavy as you. I will try a) & b) and see if there is any improvement in the quality of the IDeA queue.
Parent - - By cma6 (****) Date 2017-06-14 00:52
Yes, I know that sending the whatever--a), b), or c)--from i.a. to IDeA is automatic once minimum depth/time criteria have been met. However, I have found that right-clicking the analysis pane while doing i.a. and "Sending to IDeA" seems to send more information to IDeA than the automatic method alone.
Parent - By pawnslinger (***) [us] Date 2017-06-14 14:51
Maybe, but I have never done it.  So I don't know from my own experience.
Parent - - By cma6 (****) Date 2017-06-13 02:11
When I open the "Analysis Quality Settings" dialog, if I set 300 seconds and 40 plies, I am unable to check either "And" or "Or".
Am I right in assuming that IDeA defaults to "AND" in that case?
  I assume that that is the "Min Depth" parameter to which you are referring?
Parent - - By pawnslinger (***) [us] Date 2017-06-13 06:55
It should tell you in the Stage Status pane whether AND or OR is selected.  I would assume nothing.

"Min Depth" is found on the Project Properties Dialog Box.  Not the Analysis Quality Settings.  Min Depth does not affect the quality settings at all, it controls how much of the analysis is transferred from the Sandbox to the IDeA Project queue for later evaluation (where the quality settings would be applied).  It also controls the same thing for IA proper in the IDeA pane.
Parent - - By cma6 (****) Date 2017-06-13 12:09 Edited 2017-06-14 00:42
In IDeA's Stage Status pane, I do not see any reference to AND or OR. Since I am unable to click either AND or OR in "Analysis Settings"--are you able to do that?--what Analysis Settings does IDeA default to?
I figured out that in this case AND will be the default.

"Min Depth does not affect the quality settings at all, it controls how much of the analysis is transferred from the Sandbox to the IDeA Project queue for later evaluation (where the quality settings would be applied).  It also controls the same thing for IA proper in the IDeA pane."
   
If "Analysis Quality Settings" are at 300 seconds/40 plies/Max time 300 seconds/"Wait for next depth", you are saying that those settings are confined to i.a. in Sandbox or DB pane? In that case, how does one best harmonize "Analysis Quality Settings" with IDeA's "Min. Time" and "Min. depth"?
Parent - By pawnslinger (***) [us] Date 2017-06-13 18:11 Edited 2017-06-14 19:57
I am sorry, it is the "Project Status" pane.  The very first line in the Project Status pane gives the AND / OR settings, as follows:
Parent - By pawnslinger (***) [us] Date 2017-06-13 18:19
Quote:
======
If "Analysis Quality Settings" are at 300 seconds/40 plies/Max time 300 seconds/"Wait for next depth", you are saying that those settings are confined to i.a. in Sandbox or DB pane? In that case, how does one best harmonize "Analysis Quality Settings" with IDeA's "Min. Time" and "Min. depth"?
======
I am saying that these parameters apply only to the IDeA evaluation function.  IA and Sandbox analysis are by definition (almost) infinite, that is they continue until you stop them manually.  So they do not use the "Analysis Quality Settings", they use YOU to decide when enough is enough.

The Min Depth and Min Time parameters control how much of the IA or Sandbox analysis is queued for IDeA to process.  Only those moves that meet the Min Depth and Min Time parameters are allowed into the IDeA queue.  And of course, from there they are evaluated using the "Analysis Quality Settings".
Parent - - By cma6 (****) Date 2017-06-13 02:05
"The quality of your analysis tree is entirely dependent on finding and fixing bad evaluations out on the leaves."

Do you turn off automatic tree generation?

   It is common to find bad evals with a relatively small # of nodes (if that is the right term), e.g., if the position after 21 Bh4 is .12 to .14 for W, with evals in at range like:
         21...Nf6   .13    112 nodes
         21...Bd7  .13       96 nodes
         21...Be6  .14       81  nodes
         21...Bf5  0.01        3 nodes

A brief i.a. check of 21...Bf5 will show that the true eval to depth 39 is actually 0.35 
     So are you saying that all the evals up to 21 Bh4 will be messed up because of the faulty eval of 21...Bf5, and if so, how does one fix in IDeA the bad eval for 21...Bf5?
Parent - - By pawnslinger (***) [us] Date 2017-06-13 07:00
I do sometimes turn off automatic prolongation, but not usually.  It is quite common to find nodes such as you describe, however they can only be fixed by prolongation of leaf nodes.  So you must navigate down stream from these nodes to the leaves that are under it and run IA or Sandbox there to get the correct evaluation.  This must be done for all leaf nodes that can affect the minimax of the node under study.  It could be anywhere from 1 leaf node to many, depending on the situation.

A leaf nodes being defined as a node with no candidate moves yet.  It is the point at which the minimax can be changed most quickly.
Up Topic Rybka Support & Discussion / Aquarium / Which engine for IDeA?
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill