The 8 opening positions I choice will follow. They are about equal positions with chances for both sides. I would be open to suggests form Uly and the crowd if they would like to give one.
That's why I think for this challenge, subsequent analysis methods is my best option, these take about the same time as the previous one, so I get meaningful results ASAP, here's the outline:
4 Layer Analysis (all methods based exclusively on Infinite Analysis and interaction):
Layer 1 - Carousel. This analysis method focuses on rotating different engines on the position to check, and cross check their ideas, until eventually one stands up to get it better, which becomes the Main engine. At the start of the game I'll also focus in finding the Lancer engine, one which complements the Main engine the best (lowest redundancy).
Layer 2 - Pyramid. This analysis method is radically different depending on what engine was found best on the carousel, as it usually focuses on the exclude moves feature, but there are engines that don't support it so I'd need to rely on MultiPV (and also if the main engine turned out to support Persistent Hash). The method focuses on getting as wide as possible near the root, the tree ends being shaped like a pyramid (umop apısdn).
Layer 3 - Ping Pong. The Main engine and the Lancer engine crosscheck the game, the main engine plays my side, the Lancer plays yours, trying to find holes in the pyramid. If enough holes are found and the main engine is refuted, the Lancer becomes main engine, or I may go back to the carousel, since the main engine turned out to be a bad choice after all.
Layer 4 - Simulacron. Usable only after I have a pyramid built and the Lancer defeated, this focuses on finding tactical positions and going as deep as possible on the mainlines, I can keep using this method endlessly as refuted mainlines go back to Ping Pong (if I refuted your lines), or the Pyramid (if I refuted mine) so I don't need a fifth layer.
These methods don't allow me to dig for human moves or squeeze centipawns (my generic analysis), I am also not allowed to abandon the methods or introduce new ones (so, no Kitchen Sink, etc.) I think this is fair since you won't be using Infinite Analysis at all, which sounds impressive.
Each layer takes about 15 minutes to do (except for L1 that tends to explode, but L4 isn't important until later on), so I'd very much like the idea to restrict myself to 1 hour of analysis. I don't know the speed of your hardware, but we'd probably like to level things in round numbers, such as you analyzing for 2 hours, or 3 hours, depending on hardware disparity, and we have to have a move ready at the time out and make it (going by the honor system of time taken). No time forfeit (we can start these analysis hours at our discretion). I can tell what I did each move, but considering I wouldn't be giving details, it's probably not very useful.
So we're almost ready to rumble! Thanks for the challenge, and good luck.
Can i ask in Layer 1 how to you allow for the fact that some engines are inaccurate with the position score.
Engine One likes move A with score 0.10
Engine Two likes move B with score 0.60
That doesn't mean move B is better, the highness of 0.60 doesn't mean I give preference to the move, I'll just cross check:
Engine One dislikes move B with score 0.00
Engine Two likes move A with score 0.64
There, move A is better (both engines agree), engine One is better (it found it by itself).
I am new to this computer chess and have a couple of question.
DPA, i know the Fritz GUI has an option for this but you appear to be doing it manually.
Aquarium only has IA or idea, so if you are doing it manually are you not just doing IA on different positions.
How to you store your result, inserting the analysis into the game notation would become messy.
I bet this one brings back memories
I'll give 3 ideas here.
1) Use one of these I posted and have limited book use, say 3 to 5 moves
2) Use one of these with and disable book use.
3) If you wish we can use more longer lines, but the reason why I did it this way is there can be a lot more possibles instead of setting the course of the game down to 15 or 20 moves. The choice is yours, I do have a few others and I will like to get a chance to play the oppsite color of another game.
As far as time control goes. I can probably make a move or 2 a day or so. Just let me and I'll let you know what things look like form the day to day bases.
By the way my specs are daul 2.13 Ghz 4 GB of Ram. I think you can figure out a decent to control here.
I like this approach, as you posted positions you'd be happy playing and I picked one I'll be happy playing
Just give me the PGN that leads to this position and we're ready to go. [Edit] (I think I got it, probably 1. d4 f5 2. c4 Nf6 3. g3 e6 4. Bg2 d5 5. Nf3)
>2) Use one of these with and disable book use.
I like this one the best, the idea of position presets is that we get rid of the book factor, which in this case is as easy as not checking any book nor database, and it's also the simplest. Of course I wouln't mind if Nelson gave statistics about this position.
>I do have a few others and I will like to get a chance to play the oppsite color of another game.
Sure, I actually would like to play the black side of this position that you posted as a second game:
>By the way my specs are daul 2.13 Ghz 4 GB of Ram
Well, then I think giving you 3 hours of DPA per move sounds about right? Otherwise we'd need to run several tests to equalize our times to the last Hz, but doing so didn't have much of an effect in the Alan match.
I don't know how often I can play moves, but if you're willing to wait as much as necessary, we can start the match at April 12.
> I was really only worried about playing the ruy position.
Oh, I was very tempted to pick the Ruy and play it as black, since I main it and have a mountain of analysis ready for it. Then I remembered I couldn't use any of it , so, yeah, better something new than rehash something I have already analyzed in depth.
And since it's your turn to move, I think you can do it at any time (as if you chose any of the resulting positions as the preset, and I'd move at game start).
> I like this approach, as you posted positions you'd be happy playing and I picked one I'll be happy playing
> Just give me the PGN that leads to this position and we're ready to go. [Edit] (I think I got it, probably 1. d4 f5 2. c4 Nf6 3. g3 e6 4. Bg2 d5 5. Nf3)
Saying it like that it will be my 5th move here. In 2 days.
Black to play, 44.1% with 31.9% draws. ECO A90, which somewhat favors white, is somewhat less-than-avearge in being drawish.
Black to play, 48.9% with 42.7% draws. ECO D45, slightly favors black on a relative basis, is one of the more drawish openings.
This may not be your cup of tea.
Oh, and, I don't think time has to be taken consecutively, if you have 3 analysis sessions of 1 hour, or want to distribute your time differently, there's no problem. The time added is what matters.
I'm looking at my move now by the way. In phase 1
(vulneraries are my banked time, that I never get to use...)
I also hope that without banking time, we keep the playing level constant thorough the game, I wouldn't like to see a player thinking a position is easy and moving too fast, only to find he missed something big, specially if that player is me.
Phase 1-Search for root moves(25 min)
Phase 2-Make lines for root moves(2hr 25min)
I actually trusted what come out and I didn't really have time to polish it up. But in the future I will as the root moves go down.
O-O found best in 6 minutes (though, the alternative wasn't bad.)
1. d4 f5 2. c4 Nf6 3. g3 e6 4. Bg2 d5 5. Nf3 Bd6 6. O-O
Last night I got my root moves-20m
Today making variations-2h 45m
Main move found in 1 minute 36 seconds.
Qb3 found in 18 minutes.
Main engine falls from the carousel at minute 32 (wrong engine ordering bug ).
Qb3 found best at minute 37, when the engine that found it becomes new main engine.
Lancer engine found at minute 48 (lines remain not refuted at time out, one engine not checked at all)
1. d4 f5 2. c4 Nf6 3. g3 e6 4. Bg2 d5 5. Nf3 Bd6 6. O-O c6 7. Qb3
> Full hour on the carousel again (wow, these positions do allow for creativity )
I agree that is one reason why I choice it. I think Nc3 is probably the main move here, but I don't see anything wrong with Qb3.
And, my analysis methods don't let me force and examine moves like Nc3 until it becomes best, just like Jimmy isn't allowed to use Infinite Analysis. We're restricted (that's why it's "Battle of the methods").
But I've been thrilled to do the best I can do in one hour, and it was decided before the game started, on a normal game I guess I'd add MegaScout before the Carrousel (that didn't like Nc3), since the main engine is changing, but I had to think about that possibility before the game started.
Anyway, MegaScout is an obsolete analysis method of mine that I haven't used in ages, and the reason is, that if your other analysis methods are going to miss a move that MegaScout would find, there's something wrong with them at the core, so I've switched to methods that do that without wasting so much time at the root.
By now MegaScout, Carrousel, Follow The Leader and King Of The Hill starting analysis methods are already obsolete. I have to thank TheHug for that, regardless of game result he definitively defeated Layer Analysis.
It would be great if you could explain Follow the leader and King of the hill methods, in that way I (/someone else) could increase my learning curve. I'm saying this beacause you said that they are obsolete, so no harm for you in doing so. Steel, if you can't, that's totally ok. Thanks for explaining Megasout (and the reason why is obsolete!).
The Carrousel, King Of The Hill, and Follow the Leader are analysis methods that vary on the same idea: To get to know what engines are the best on the position, and use them to lead the game. So, say, if Rybka 4.1 is really bad on the positions, you don't use it, and save time because that time is used with an engine that DOES know what is going on.
This is done by cross checking with several different engines. At the start of the game, you throw your all your arsenal, which quickly refutes the worst engines, and you end with the cream of the crop.
Cross checking is rather simple, you check what lines the engines propose, and where they deviate from each other. The critical deviations are those that happen at the root, that is, the engines suggest what move you should play on the game, and they disagree with one another.
Then, you play the moves they didn't see into each other, what'll you see is that one of the engines will eventually go "oh, you're right, that move was beyond my horizon, I like this line better than the one I suggested".
How to handle these situations varies with the analysis method, in the Carrousel, if an engine gets a huge fail low in some variation, I search for an engine that doesn't, and replaces this engine. In King of the Hill, this is done depending on the variations refuted from an engine, and how many variations this engine refutes from other engines, so getting a huge fail low isn't that important if this engine caused huge fail lows in other two engines. In Follow The Leader, what is taken into account is the line lengths before refutation, no matter how small of a refutation. Say, if an engine takes 40 moves to realize its line is worse than the one from another engine, you're wasting a lot of time, so you ax the engine and search for one that realizes what's the best line sooner.
These methods also vary in the way refuted lines are handled (keep refuting lines until the engine agrees with the other one, keep refuting lines until they reach a threshold, or assume all of them would be refuted and jump to the root), and how ties on positions that don't transpose are handled (whether ties at the root are treated differently than ties deeper on the line, and if engines that think all moves are equal in scores are used for their move choices or dropped).
At the end of the process, I end with 6 engines, which seems to be the sweet spot, though in this game against TheHug, very often I just ended with 5, because the Carrousel was taking too much time to finish and there's no point to find a 6th engine I'm not going to use (since the carrousel restarts on the next move).
What I did with those 6 engines now is different depending on the method. The Carrousel searches for 2 main engines with low redundancy because another analysis method will depend on them. King of the Hill only needs to find one top engine, which is the main one that is leading the game. The other engines try to refute it and become the new king of the hill, but the order of the other 5 engines isn't important unlike in the Carrousel or Follow The Leader. In Follow The Leader I separate the engines in 2 groups of 3 engines, in one, there's the Main engine, and two engines that have the best understanding of the position, while in the other, I have 3 engines with lowest redundancy with the other 3.
While in the Carrousel 2 lowest redundancy engines lead the game, in Follow the Leader the two best engines for the job lead the game, regardless of the redundancy of them. This allows FtL to find a main line the fastest, and then, if the first 3 engines agree on this line, then I search for alternative lines with the other 3. That is, 3 different plans from main line. If one engine from the second group suggests the main line again, I may drop the engine and search for one more original, unless it found it faster than the first 3 engines, in such case it may replace one of them. King of the Hill is usually more straight forward and easy on this, allowing for going deeper as there aren't so many main line changes as in the other two, but the main line takes longer to find (that means there are many instances in where the process continues without a main line) and engine changes are less frequent so it depends heavily on the results of previous moves.
While these methods were good in telling me what engines to use, they weren't good in in telling me what moves to play, that's the responsibility of other analysis methods. Remember the most important thing is that you understand why the plans of the engines you're not using failed, and why the variations of the 6 engines that remained worked. Afterwards you can discover or extend the most interesting variations on your own.
The reason they're obsolete is that they were superseded by the Hammertime analysis method, that not only is good at telling what engines to use, but it's good to hold its own against other methods for move picking, so when I'm done with it I have an excellent engine order and move choice, though finding the best one is harder, at least I have the best engines chosen to do it.
Just a quick question: how do you select engines for your team? Trail and error (experience in pratice)? Through CCRL ponder hits and eval diff? Do you establish a minimum elo?
For new positions I just go with the last set that won me a game, and it hasn't turned very well , I'm still experimenting on what engines I should use to start a game.
>this is one of the best posts in the forum!!
Heh, well, I just depicted 3 analysis methods that do not work, but maybe it'll give someone ideas of their own.
Indeed. I think I've found some improvements ;).
The process I used was basically the same.
About 15 minutes for root moves and the rest to make my main lines.
Main move found in Ponder Hit.
Since lines remained not refuted in last move I jump directly to Ping Pong.
Lancer beats main engine and becomes new main engine in minute 24. Main move discarded, a second move becomes best.
Back to carousel for a long time as an engine gets hard to refute, and then, about 30 minutes later, the Hero (the engine that found 6.O-O first and fell from the carousel last move) refutes the Lancer.
Bg5 (the third one) becomes best move at minute 55. The Hero becomes main engine at minute 58, and the old main engine from this move disagrees with the move, but, time out!
Now, that was intense, if the time out was 54 minutes my move would have been different. Oh, and I still don't have a mainline...
1. d4 f5 2. c4 Nf6 3. g3 e6 4. Bg2 d5 5. Nf3 Bd6 6. O-O c6 7. Qb3 O-O 8. Bg5
A second point why Qb3 is bad is that white usually advances his pawns on the queenside which is now impossible or at least delayed.
Anyway, I didn't check b3 at all, but at least, Jimmy had problems picking his move as well.
> OTB, I don't see why white would go after black's dark bishop, since it's not hitting anything and white's dark bishop it's at least pinning a knight.
Because it's his good bishop.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill