2. I use 30 s and depth 1 ,wait for next depth, FA check 4 , prolongation 100 %. Whats wrong with it ?
3. Why Idea is Missing tactics ?
4. If to find Tactics Interaction is Necessaary then why use IDeA and not IA or DPA as IDeA is wastage of cpu time?
> 2. I use 30 s and depth 1 ,wait for next depth, FA check 4 Whats wrong with it ?
wait for next depth has no meaning if you have depth_1 (next depth is 2 and its always reached within 30 secs)
>, prolongation 100 %.
do you mean width 100%? if yes its too much
>3. Why Idea is Missing tactics ?
because chess is an infinite space and it will always miss them
>If to find Tactics Interaction is Necessaary then why use IDeA and not IA or DPA as IDeA is wastage of cpu time?
1) Not all tactics, but some deep tactics that are beyond a certain horizon
2) because IDEA will do 90% of what you do with IA, but 24hours a day (which is not possible for a human)
3) Because that 10% is the Chess Skill of the player. Intuition. Search where the engine says is bad here...
The engine is IDEA's compass. You use the compass to explore. If the compass misses things (by far) then you are on your own.
>wait for next depth has no meaning if you have depth_1 (next depth is 2 and its always reached within 30 secs)
You used this setting to solve a postion related to zugzwang. Hence I used that.
>do you mean width 100%? if yes its too much
I think Its advised to use 100% in advanced tab in IDeA when using multiple single core engines. I use Rybka 4 1 cpu x 6 on amd 1090T X6. I did not understand about your "too much"
>The engine is IDEA's compass. You use the compass to explore. If the compass misses things (by far) then you are on your own.
That I know. But still I think There may be good auto mated solutions. I have problems with 10 %. Is other weak engines good in Idea specially Zappa Mexico II ?
This means it is not good to use with IDeA. You can start it with Zappa but the resulting tree is probably useless.
The top free engines are all OK with IDeA, except for Stockfish; it may support 'searchmove' but the big evaluation jumps make it less suitable for IDeA.
In my opinion a 100% width is perfectly valid, especially with two factors combined : 1) wide trees due to many alternatives close in eval, and 2) low search depths or a few seconds per position. In such a scenario you would be looking to limit the tree depth in order to find the right tree eval window before increasing the tree depth.
My current preferred practise is this :
a) start with limited tree depth and search depth while keeping the tree as wide as possible. For this purpose I use zero seconds AND a fixed number of plies (0 seconds AND 13p for example)
b) wait until no more tasks are generated given a certain eval window for the tree (experiment with settings to find your sweet spot on your machine given the time span you are away from your terminal)
c) enlarge the eval window slightly in favour of the defender (lower limit becomes a little lower/more negative) to automatically look for 'a few' black holes. With a small change IDeA will normally run out of additional tasks in no time. If at some point the queue is filled with many tasks you know you have hit a big black hole early on in the tree.
d) next stage is to increase the tree depth but I would also advice to tighten the eval window again back to what is was or go even tighter while at the same time increase the search depth. If the primary line is slowly increasing in eval then adjust your window accordingly.
e) wait until no more tasks are generated and repeat the cycle.
f) At some point it will take too much time to wait for the next tree depth to complete. Try to anticipate that tree depth (by experience with stable settings and the same engine) and then switch to a different approach in narrowing the tree by decreasing the tree width setting ( < 100%) and increasing the search depth at the same time.
This approach could be an interesting approach to use if you have an average of more than 4 days per move choice / position. More days is better but if you only have 1 or 2 days then I guess you won't be at the terminal often enough to complete a few cycles.
My main advice as I have posted many times: most of times it is not worth to use very deep settings, say d=20, if you want to get different results than IA results. IDeA algorithm seems to get much more profitable results from múltiple iterations rather than going very deep into a position.
But at the end, it all depends on your available time to be in front your computer and interact with it.
Also, you state "I think the key thing to understand is that a d=15 search conducted 7 ply off the root position is almost always more accurate than a d=22 search conducted at the root position."
How would one conduct a search off the root position? Is that a separate IDeA run after the first one or a 2nd node that is run simulataneously with the main node?
Also, how can one have two nodes running at the same time, since IDeA would not know which node to use or why to use it?
>Moz, what is a "mid-ply" search?
It's subjective of course but I always think of depth=15-17 as happy medium. Deep enough to be fairly reliable without taking too long to reach with a single core engine.
>How would one conduct a search off the root position?
IDeA does this automatically.
Thank you for the explanation. 15-ply seems to be a happy medium between those calling for 1-play and those calling for 20-ply initially.
Can you articulate your reasons? Do you think that by listing your specific settings you will be elevating the rest of the world too much? :)
> Can you articulate your reasons? Do you think that by listing your specific settings you will be elevating the rest of the world too much? :)
Because I don't want an opponent to use them against me. Is it really that hard to understand? It's no different than telling someone what engine you use and exactly how you use it. If I know that information about someone, I'll exploit it. If I know the IDeA settings someone uses I'd exploit that as well. Like poker, the less info your opponents have about you, the better off you are.
Seriously, it's not hard to infer the settings I use from reading this thread. I use mid-ply searches, the width varies a lot depending on the position and I rarely, if ever, adjust the advanced settings. I do a lot of interaction and make heavy use of color marking to shape my tree as I go.
I don't think there's any IDeA secret sauce. I think a multifaceted approach is always best, one that combines interactive IA and DPA with lots of IDeA interaction.
> What you are saying would be true if people you play know you by the name you use to post here.
I'm playing in the WBCCC tournament here at the Rybka forum. There's money on the line.
> A truly light setting is probably d=11.
yes, I agree.
> This 'pyramid' approach of starting wide and ending narrow in my opinion is a good way to sidestep the pitfalls in infinite analysis with today's engines, which seem to collectively sacrifice the concept of finding the optimal move by their aggressively tuned pruning and search defaults
Yes, although in some endgames where I cannot help the engine with my chess knowledge my pyramid gets suddenly inverted again!
> You are contradicting your own 'philosophy' with your '7 ply off the root' statement
I did no such thing. If 100% width and incessant micromanaging of Idea works for you - great! Have at it.
I only know that the settings I use work great for me and they're drastically different than the settings you use. I'm also pretty sure that your approach would be a disaster for Omprakash since he specifically mentioned that he's looking for an approach that doesn't require a lot of user interaction. Since 100% width essentially says that you're interested in ALL moves irrespective of the evaluation, it's a terrible approach for someone who isn't willing to invest a lot of time interacting with the analysis. I suspect that the "holes" that he's mentioned in previous posts concerning Idea are probably the direct result of too much width w/ not enough interaction. I posted here not to get into a debate about Idea settings but in order to help him get more out of Idea. We'll have to agree to disagree on the "philosophy".
> Your intentions to help may be good but imho you got it all wrong.
Yeah, you're right. I've got it all wrong. I don't know what I'm doing. My corr results are terrible. Why do I even bother playing when I keep embarrassing myself with faulty Idea analysis...
Unlike you, I've played games here. People can look at those games and decide for themselves if my results are effective.
Personally I would listen to Moz and create your own settings based on your needs and hardware.
I do a lot of interaction, have 14 cores available and my best settings are D=18 or T=300 secs, tree width 60%, max moves 10 and gradually increase if needed.
I never use IDeA as a single analysis method, IA sometimes produces different and better moves. Opening theory (books or games annotated by titled players) are most trustworthy than any engine output in the opening stage.
When I get to the middlegame I increase the depth to 20, in the endgame I use 22 or 24.
Any engine output without interaction is subject to significant improvement, correspondence chess without interaction is simple computer chess. If your opponent knows what he is doing you may improve your results by interacting with IDeA or IA. In the end, the best player usually wins in about 40% of the games.
I'm tied up with games right now. Glad to see you in this forum.
We choose to walk on paths of some gradients (moves with certain scores) and ignore others. Most times that is the correct way to do it. Some times there are hidden paths that the compass (gradient) does not show.
A good explorer is needed to sense those hidden paths. The more complex the position (manifold) the more chances to find such a move
If one wishes auto solutions, then he simply does not now the sport at all. IDEA is just a robot to do most of the exploring "common usage of the compass" would do. The hidden paths are left for the explorer.
There are no magical settings. What you gain by better quality search (through engine depth, since engine has far the better smarter search) you loose it by creating a big tree to work with later (interactively)
I prefer normal depths (fast times) and 5-6 alternatives per move. Then go through the paths and try to find new paths by using my skills. Better skills, better results...
I started analyzing a complex position defined as a position where my IDeA settings did not find the move I had chosen, as a matter of fact my choice was not even in the top 5 moves in the tree. Your method finds it after the first stage, I'm running the other stages and will report back but this is very interesting. Only 3-4 hours as opposed to my old method interacting with IDeA and IA (and my chess intuition) where I had spent more than 2 days!
1.Nf3 d5 2.c4 d4 3.b4 c5 4.e3 dxe3 5.fxe3 cxb4 6.d4 g6 7.Be2 Bg7 8.O-O Nh6 9.a3 bxa3 10.Rxa3 O-O 11.e4 Bg4 12.Be3 f5 13.Nc3 fxe4 14.Nxe4 Bf5 15.Nf2 Bc8 16.d5 Nf5 17.c5 e6 18.d6 Bd7 19.Bc4 Nc6 20.Bg5 Qc8 21.Ne4 h6 22.g4 hxg5 23.gxf5 Rxf5 24.Nfxg5 Rxg5+ 25.Nxg5 Ne5 26.Ne4 Qd8 27.Qb3 Nxc4 28.Qxc4 Qh4 29.Kg2 Qh5 30.Rg3 b5 31.Qd3 Bc6 32.Kg1 Rc8 33.Ng5 Bd7 34.Rf7 Rxc5 35.Rxg7+ Kxg7 36.Qd4+ Kg8 37.Qxc5 Qd1+ 38.Kg2 Qe2+ 39.Qf2 Bc6+ 40.Nf3 Qd3 41.Qc5 Qd5 42.Qxd5 exd5 43.Nd4 Bd7 44.Re3 b4 45.Re7 Ba4 46.Kg3 Kf8 47.Ne6+ Kg8 48.Rg7+ Kh8 49.Rxa7 Bc6 50.Nd4 Ba4 51.Rxa4 Kg7 52.d7 Kf6 53.d8=Q+
Position after move 21 of black
IDEA will probably not find g4 in a logical amount of time, and even if it does , not for the correct reasons.
Second position is in the same line, some moves later , position after 25th move of black
by the way i should find time polishing a script i wrote some time ago, that does a "full tree" for the positions you want.
IDEA will never find the correct move without interaction (i think its pretty easy to find the correct move or to suspect the good move if you are a good OTB player.)
My point is that the pyramid approach is a safe approach (although time consuming), but i prefer having anand beside me to help me interact
>Of course the final decision takes intensive interaction, most likely with IA since Anand will not return my calls :)
> I'm very impressed so far.
Me too! I've been playing around with Sciolto's settings and found them to be extremely helpful in really complex positions, especially those that are dynamic or unbalanced in some way. For "dry" positions, I have a lot of trouble with repeated moves when I use them. But for opening analysis or early middle game analysis where the tree often explodes and Idea suffers from tunnel-vision, I think the settings are very useful. Good stuff!
Just like with any other settings I've seen or tried you have to be careful, the least preferred move at the base of the pyramid might be the best move overall. In my view these settings need to be checked with other methods before committing to a move.
Still a very interesting and enjoyable way to build trees.
"My current preferred practise is this :
a) start with limited tree depth and search depth while keeping the tree as wide as possible. For this purpose I use zero seconds AND a fixed number of plies (0 seconds AND 13p for example)..."
is a very useful guide to getting some value out of IDeA.
It would be very valuable if Sciolto or some expert could explain, step by step, how to implement his methodology.
To give an example of why this very advanced post needs explanation for non-expert IDeA users, setting 0 seconds would seem to result in no analysis or completely worthless analysis, so why would he do that?
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill