at least the white correspondance game player would have beenDon wrote:That is pretty bizarre - I cannot figure out why Komodo (who also likes Be1) would so happily give up a passed pawn. Komodo does give a relatively high score for the king side pawn majority but that should not trump a nice passed pawn.mclane wrote:ok. i do begin.Don wrote:I have a very hard time believing that. Any top engine should be play better correspondence chess than one hundreds of ELO weaker.hgm wrote:Someone who actually has experience with this in correspondence Chess (Uly on Rybkaforum) tells a completely different story, though. Accoring to him Houdini is virtually useless, because it initially produces moves that, although winning in blitz, are losing in correspondence Chess. By the time (many hours) it switches to moves that are any good, other engines have already settled on those moves for a long time. So he does no longer use Houdini, but HIARCS, which suggests moves no other engine suggests, he considers a valuable asset...
Maybe this is a style issue. Maybe, for whatever reason, Uly does not like the playing style of Houdini for correspondence chess.
I would like to see examples of what he is talking about, not just one but several - but even one would be better than nothing.
this is a real mail chess / correspondance game.
1.e4 c5 2.Nf3 e6 3.d4 cxd4 4.Nxd4 a6 5.Nc3 Qc7 6.Be2 Nf6 7.O-O Bc5 8.Kh1 b5 9.f4 Bb7 10.Bf3 O-O 11.e5 Bxf3 12.Nxf3 Ng4 13.Ne4 Be7 14.Nfg5 f5 15.exf6 Nxf6 16.Qd3 g6 17.Qh3 h5 18.Nxf6+ Bxf6 19.Qd3 Kg7 20.Ne4 Nc6 21.c3 Be7 22.Be3 d5 23.Nd2 Rf5 24.Nf3 Bf6 25.a4 bxa4 26.Rxa4 a5 27.h3 h4 28.Bf2 Rh8 29.Re1 Kf7 30.Qd2 Qb8 31.Qe2 Qd6 32.Nh2 Bg7 33.Ng4 Qd7 34.Qd1 Rfh5 35.Rae4 Re8 36.R4e3 Rf5 37.Qa4 Rh5 38.R3e2 Qc7 39.Qb5 Qd7 40.Qd3 Qd6 41.b3 Qd7 42.c4 Re7 43.Rd2 Qd6 44.Rdd1 Rc7 45.Qf3 Kg8 46.Qe4 Ne7 47.c5 Qd7 48.Qxe6+ Qxe6 49.Rxe6 Kf7 50.Rd6 d4 51.Ne5+ Bxe5 52.fxe5 Rxe5 53.Bxd4 Re6 54.Bf2 Nf5 55.Rd7+ Rxd7 56.Rxd7+ Ke8 57.Rd1 Re2 58.Re1 Nd4 59.Rxe2+ Nxe2 60.Be1 Nd4 61.Bxa5 Nxb3 62.Bb6 Nd4 63.Kg1 Kd7 64.Kf2 Ke6 65.Ke3 Ke5 66.Bc7+ Kd5 *
[D]4k3/8/6p1/B1P5/7p/1n5P/6P1/7K w - - 0 62
Houdini moves Be1.
this gives up the important pawn c5.
in opposite to houdini2.0, hiarcs14 changes to the right move bb6 that defends
the important pawn. hiarcs increases evaluation from 0.80 to 1 pawn and later 1.20.
humans know how important pawns are.
chess programs need knowledge. and if the program
has not implemented this knowledge in the evaluation function,
search depth will NOT find it although
we have fast computers and very fast chess programs (fast searchers).
It may be too much emphasis on bishop mobility but that doesn't really explain it either. This is a puzzler. I think Bb6 probably wins.
fracked with analyzing with houdini/komodo.
I have no clue why they prefer Be1. But i would guess hiarcs has static knowledge about pawns that overwrite the search stuff in this example.
hiarcs is a slow searcher. it is doing few NPS, fewer than other engines.
I guess it is because it is using more stuff in the evaluation function and prefers static knowledge instead of trusting the search will find it.
this helps a lot on SLOW machines, e.g. on the ipad.
i remember that in the old days of dedicated chess computers we had similar situations.
e.g. Fidelity Mach III knew how to mate with K and B+N vs. K.
Also Mephisto II ESB had this knowledge.
Important is also to know that your bishop is of wrong color to protect that passed pawn in the endgame. these situations happen often.
of course these "knowledge" comes rare. but if you play a correspondance chess game, that is beeing played in
months ! or more.
And you lose although you have the strongest chess programs,
because you TRUST them, and then such a stupid
GAP is the reason you lose your long long long played correspondance
game, its very disapointing.
of course humans who 100% trust chess programs instead of using their own brain, are naive.
correspondance chess IS about thinking yourself. the computers help in tactics.