mbabigian wrote:I'm not using SF for analysis. I use Aquarium's IDeA system and Komodo 11.x
IDeA works in increments. First it creates tasks to generate alternative moves starting from some root position, say the Winawer positions. It then sends those positions to Komodo for analysis. After all those alternatives have been scored. It generates prolongation tasks, to extend lines that look promising. Komodo then analyzes those. It then goes back and generates alternatives again along the lines it found, and after they are analyzed, generates prolongation tasks. Rinse and repeat. Using this process, it broadens and lengths lines, moving Komodo further along the line or into side variations. The examination of alternatives helps catch variations the engine might have pruned if left on its own. It is completely different from letting an engine sit there on infinite for days - which generally accomplishes nothing after 24hrs. It has completed more than 30 complete cycles (alternatives/prolong) since I put it on the Winawer.
In this variation alone since I diverted it from my other interests, it has analyzed more than 46,000 positions at 2+ minutes per move on quad processor machines (currently 9 of them). About 57 Quad CPU days of analysis. It stores all data in a tree with scores for the best lines found.
So if you have interest regarding particular lines and their scores, you need only give them. They may already be in the data examined. If not, I can add the lines to verify their scores versus those already found.
The system has it's flaws (many of them - LOL), but it hasn't found any way for white to push an out sized opening advantage.
Unfortunately, I'll need to put the array of machines back to more pressing matters soon, but I'm happy to look up variations in the tree even after it is no longer actively analyzing this line.
Mike
thanks for the explanation, Mike.
now I already know how it works.
the problem is, how you choose what alternatives to analyse?
for example, in my suggested g3 line(after black's Ne7), how do you proceed to find best moves for both sides deep down the tree?
for, my SF, immediately after g3 is played, wants to proceed with Nf3 as next white move, which totally makes no sense(the white knight should go to e2).
and you should be able to input correcting moves for some 30 plies or so at least, because that is when SF score starts changing from negative to 0.0 and then climbing to over 50cps white advantage.
for 30 plies, SF insists black is better, and then starts suddenly seeing white advantage.
Komodo should not make any bigger difference, both engines fail to understand the intricacies of such closed positions.
so that, unless you input white moves on each and every ply, I don't see how you would get a correct evaluation of the position. (please, correct me, if I am wrong)
again, the main position, after black's Ne7, strognly favours white evaluation-wise:
- pair of bishops, 50cps
- e5 white pawn, with e6 and f7 black pawns, with f7 being backward shelter pawn, and closed center, another 50cps
- a3 isolated pawn, -20cps
- c3 doubled pawn, -20cps
adding the total(other factors are more or less equal), you get some 50-60cps white advantage, just on or slightly above the winning margin in the mg(I would say slightly above)
that is my maths, and I don't go any further.
I have checked this line in thousands of games and analytical sessions, to know white wins by force, maybe not all 100% of available lines, but at least the very vast majority.
anyway, looking forward to additional feedback from you.(and, if you give me each black response, and I input white moves, you will certainly get to over 50cps in some 30 plies or so)