Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / Seeding when using Komodo 13 MCTS for IDeA
- - By RSeanOConnell (**) Date 2020-03-27 01:21
Does it make sense to seed a position using Komodo 10 or Stockfish 10 for K13 MCTS to analyze in IDeA?
Parent - - By pawnslinger (****) Date 2020-03-27 06:10
Yes.  I often do this sort of thing.  I don't use Komodo, but I often use different engines to seed the tree.  As long as you get a good spread of alternatives to be evaluated by your main engine.  The worst thing is to get a surprise move from your opponent, one that IDeA didn't bother considering.  It used to happen to me often, when I was rated around 2000, not so often anymore (now over 2400).
Parent - - By RSeanOConnell (**) Date 2020-03-27 16:11
Even though I am using MCTS with Komodo 13 vs Alpha-Beta or whatever it is.  Won't Stockfish 10 give Komodo 13 the type of "non human" moves that MCTS may not consider?  Or am I over thinking this? Awesome to hear you are over 2400.  When I get over 2000 I will be super happy.
Parent - - By pawnslinger (****) Date 2020-03-27 17:03
I started out at 1800 and it seemed like players on 2000 were impossible to beat... then 2300 was the next hurdle... they just seemed too strong for me.  Now I am struggling with players at the 2500 level and they seem like stone... really hard to even get a draw with them.

When it comes to seeding, almost any engine will work, because what you want is a full range of moves for your primary engine to "look at".  Often times IDeA will not consider good moves that your opponent might make, thus giving the tree a "blind spot".  You want to eliminate all such blind spots.  When you play a strong player and they make a move that doesn't show up in your tree... or has not been given adequate evaluation... well, that's your clue, that you may be in trouble.  So always try to be on the guard for surprise moves your opposition might be considering (their pet line).

I cannot tell you how many times that a stronger player has surprised me.  In the past, when this happened, I said to myself, ah... looks like he made a mistake!  Well, that was my mistake.  If you are playing a stronger player, never assume he has made a blunder in such a situation -- you will be lucky to pull out a draw in such situations.
Parent - By RSeanOConnell (**) Date 2020-03-27 17:41
Thank you!
Parent - - By ChiefPushesWood (**) Date 2020-03-28 12:31
I think I would NOT recommend Komodo MCTS for IDEA for CC. Komodo MCTS is at it's strongest when it is running multi pv (I believe recommended at least 4 variations).

I don't know that I'm clear on what your usage of Aquarium is. Are you preparing for OTB play or Correspondence play. This is an important distinction.

If OTB play, one of our friends here on this forum uses Komodo MCTS extensively, from what I understand, to prepare ideologically for his OTB opponents. I'd defer to him.

If you're preparing for Correspondence play, well Komodo MCTS is just worse than regular Komodo and certainly worse than SF 11. This is by Larry and Marks (K authors) own admissions. 

Parent - - By RSeanOConnell (**) Date 2020-04-01 01:01
Strictly OTB.  I take key positions from my games and plug them into IDeA.  Adding my what if's as it hums along.  I then take those positions and put them into my personal tactics database.  I play through the positions over and over.

Also, using IDeA for opening prep.
Parent - By Ghengis-Kann (***) Date 2020-04-22 13:21
I'm a little late to the game on this thread but pretty sure CPW was referring to me so I will add what I have found.
This refers to preparation for OTB play.

The MCTS engines (both Komodo MCTS and Leela), are useful in positions where long term strategic ideas are important.
They will often find variations in the sidelines that are completely playable against a human who is on the clock and tend to be solid enough that you won't easily go wrong.

Those engines are inferior to alpha/beta search engines in highly tactical positions, and inferior in general for computer versus computer play because they miss some tactical trick that is 30 moves deep.

I have used  Komodo MCTS or Leela to prepare for closed positions like the Colle or Nimzo-Indian, but lately have been moving my repertoire towards more risky and dynamic systems like the Morra, Urosov, and Jaenish gambits because I believe that piece activity is more important in human versus human play than computer versus computer.

With this in mind I have been making use of "Crazy K", which is my name for Komodo 13 with the dynamism value cranked up to 200.
It is a lot more optimistic about things like sacrificing a pawn for piece activity and is helping me find some fighting variations to try over the board.
Up Topic Rybka Support & Discussion / Aquarium / Seeding when using Komodo 13 MCTS for IDeA

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill