Sep-17-19
 | | AylerKupp: <<spingo> I like McShane's original protection of his b-pawn from the black queen's threats. 11.Ra2, then when the queen goes to a6 he plays 20.Ra1. He then plays 22.Bc1 and 23.Rb1.> After 11...Qb6
click for larger viewI frankly didn't see much difference between 11.Ra2 and the latter 20.Ra1 ... 22.Bc1 ... 23.Rb1 as played in the game and the immediate 11.Bc1. In both cases the White rook winds up on a1 and White's DSB winds up on c1. So, for fun, I used Stockfish 10 with MPV=5 and threads=4 to see how it might approach the defense of the Pb2 after 10...Qb6. To my surprise, at relatively low ply for Stockfish (d=35) neither 11.Ra2, 11.Rb1, or 11.Bc1 were among Stockfish's top 5 moves. Stockfish preferred to offer the Pb2 and it's top choice was 11.g4 although it only rated it as equal, [0.00] after 11...dxe4 12.dxe4 Qxb2 .... Its other top 4 moves were 11.Qf4 (evaluating it also at [0.00] after 11...Qxb2 ...), 11.b4 (evaluating it at [-0.05] after 11...h5 12.Qf4 Nbd7 ... with Black never capturing ...Qxb2), 11.Na4 directly protecting the Pb2 (evaluating it at [-0.13] after 11.Qb5 12.b3 ...), and 11.b3 immediately (evaluating it at [-0.22] after 11...h5 12.e5 Nfd7 ...). So Stockfish evaluated the positing as effectively equal after 10...Qb6. So, again for fun, I had Komodo 12.2 MCTS evaluate the position after 10...Qb6. Komodo using MCTS instead of its standard alpha-beta pruning (ABP) is claimed to play more "human like" just like AlphaZero is claimed to play. Since when using MCTS there's supposedly no time penalty at higher MPVs, I set MPV=20 to see at what point it would select either 11.Ra2, 11.Rb1, or 11.Bc1 as its approach to defend the Pb2 or whether it would offer it just like Stockfish did. I need not have bothered. At d=30 it selected 11.Ra2 as its top move (evaluating it at [+0.07]), 11.Rb1 as it's 2nd top move (evaluating it at [-0.01]), and 11.Na4 as it's 3rd top move (evaluating it at [-0.06]). It only considered 11.Bc1 as its 18th top move (evaluating it at [-0.46]) and considered its first offer of the Pb2, 11.Qf4, as only it's top 4th move (evaluating it at [-0.46]). Since multi-core chess engines are notoriously non-deterministic (right, <RandomVisitor>?), I let Stockfish 10 analyze the position after 10...Qb6 again except for setting MPV=20 and only let it run to d=32 because I needed to get unto other things. This time it selected 11.Qf4 as its top move (evaluating it at [0.00]), 11.g4 as its 2nd top move (evaluating it at [-0.01]), and 11.Rb1 as its 3rd top move (evaluating it at [-0.11]). For contrast in the previous analysis at d=32 using MPV=5 Stockfish selected 11.g4 as its top move (evaluating it at [0.00]), 11.Qf4 as its 2nd top move (evaluating it at [0.00]), and 11.Na4 as its 3rd top move (evaluating it at [0.00]). So it seems that Stockfish prefers to offer the Pb2 as its top 2 moves and protecting it (in different ways) as it's top 3rd move while Komodo MCTS prefers to protect the Pb2 as its top 2 moves and only offer it as it's top 3rd move. So indeed Komodo MCTS player more "human-like" (or at least more accurately "McShane-like") than Stockfish, although this was a Blitz game and McShane might have decided to offer the Pb2 if he had more time to think about it. Finally, to see if there was a difference between Komodo 12 MCTS and standard Komodo 12 (using ABP), I let Komodo ABP analyze the same position with MPV=5 (since there is a time penalty with additional MPVs in chess engines using ABP. I let it run until d=38 and this version considered 11.0-0-0, 11.Qg3, 11.Na4, 11.Qf4, and 11.Be2 all to be White's top move, evaluating each of them at [0.00], the order that they were listed being due only to the order that the moves were found on its search tree. So Komodo 12 ADP seems to consider giving up the Pb2 to be about as strong as defending it. So, if we want to find out the best 11th move for White, these engines have no clue. |
Kibitzing
sac: 49.Kg2
