MikeGL wrote: ↑Thu May 09, 2019 12:39 pm
Again, CF is showing incorrect lines, after 41.Rfe1 above, 41...Rb3 is not optimum but 41...Nb6! with a possible winning line.
This is Multi-PV with the best move and ends with a 0-1
If you follow the Multi-PV of AB engine you would agree, that below line is all top moves, and ends in a win for black.
[pgn]
[Event "?"]
[Site "?"]
[Date "2019.05.09"]
[Round "?"]
[White "SF10"]
[Black "SF10"]
[Result "*"]
[BlackElo "2400"]
[WhiteElo "2400"]
[TimeControl "5400+5"]
[SetUp "1"]
[FEN "8/2k5/3p4/p2p1p2/n2P1Pp1/P1rqP1Pp/Q3R2P/5RK1 w - - 0 1"]
[Termination "unterminated"]
[PlyCount "17"]
1. Rfe1 Nb6 2. Qb2 Rb3 3. Qc2+ Qxc2 4. Rxc2+ Nc4 5. a4 Ra3 6. Rcc1 Rxa4 7.
Ra1 Rb4 8. Rab1 Kb6 9. Kf2 {-+}
[/pgn]
Multi- or sinlge pv, could you please start doing some Backward with you lines of "top moves" before simply taking them for granted, just because at short hardware-time they are assorted top of the list of arbitrarily chosen number of primaries, dear Mike?
After 7(47) Ra1?:
8/2k5/3p4/p2p1p2/r1nP1Pp1/4P1Pp/7P/R3R1K1 b - - 0 1
Analysis by CF EXT 060519 x64 POP NUMA:
47...Rb4 48.Rab1 Nd2 49.Rbc1+ Kb6 50.Kf2 Rb2 51.Re2 a4 52.e4 Nxe4+ 53.Ke1 Rb3 54.Rec2 Kb5 55.Rc7 Kb4 56.Rf7 Nc3 57.Kd2 Nb5 58.Rxf5 Rb2+ 59.Kd3 a3 60.Rxd5 a2 61.Ra1 Rxh2 62.Rg5 Rg2 63.Rxg4 h2 64.Rh4 d5 65.Ke3 Rxg3+ 66.Kf2 Rg1 67.Rxa2 h1Q 68.Rxh1 Rxh1 69.Ra6 Nxd4 70.Kg2 Rh8
-+ (-4.07) Depth: 37/64 00:00:40 1051MN
Before that blunder:
8/2k5/3p4/p2p1p2/r1nP1Pp1/4P1Pp/7P/2R1R1K1 w - - 0 1
Analysis by CF EXT 060519 x64 POP NUMA:
47.Kf2 Rb4
-+ (-2.06 --) Depth: 48/84 00:01:29 2165MN
Still losing, but not at once as after your "top move".
After 6(46). Rcc1?:
8/2k5/3p4/p2p1p2/P1nP1Pp1/r3P1Pp/7P/2R1R1K1 b - - 0 1
Analysis by CF EXT 060519 x64 POP NUMA:
46...Rxa4 47.Kf2 Rb4 48.Rb1 Na3 49.Ra1 Rb3 50.Rec1+ Kd7 51.Kg1 Rxe3 52.Ra2 Nc4 53.Raa1 Rb3 54.Rcb1 Rd3 55.Rd1 Rc3 56.Rdb1 Kc6 57.Rc1 Rd3 58.Rxa5 Rxd4 59.Ra6+ Kc5 60.Ra5+ Kb6 61.Ra2 Rd3 62.Rac2 Kc5 63.Kf2 Rb3 64.Ke1 Kd4 65.Ke2 Re3+ 66.Kf2 Ra3 67.Ke1 Rd3 68.Ra1 Rb3 69.Rd1+ Ke4 70.Rdc1 Rb4 71.Kf2 Ne3 72.Rd2 d4 73.Re1 Rb3 74.Ra2 Rc3 75.Ra4 Kd5 76.Ra5+ Kc4 77.Ra2 Kc5 78.Ra5+ Kb4 79.Ra2 Nc4 80.Rb1+ Kc5
-+ (-2.62) Depth: 48/66 00:01:04 1671MN
Before this one blunder:
8/2k5/3p4/p2p1p2/P1nP1Pp1/r3P1Pp/2R4P/4R1K1 w - - 0 1
Analysis by CF EXT 060519 x64 POP NUMA:
46.Rb1 Rxa4 47.Rcc1 Kd7 48.Rb7+ Kc6 49.Rb8 Ra3 50.Re1 Nxe3 51.Reb1 Nc4 52.Rc8+ Kd7 53.Ra8 Nd2 54.Ra7+ Kc6 55.Ra6+ Kc7 56.Ra7+
= (0.00) Depth: 58/21 00:03:24 5015MN
Notice, if you're analysing in mv mode, you have to give even more hardware- time to get later in line arising moves into correct calculation and evaluation, the broader the plies are searched with primary or with non primary- variant parameters, is the less deeper it gets in same time.
Otherwise of course nobody would let any engine play in single variant mode, but always only with as much multiple primary lines as possible.
If you play out or analyse forward in mv-mode, you don't have to be surprised, if blunders arise with not deep enough search. That the side, that has to defend against losing, is blundering first and more often than the attacking one, is chess- immanent.
All that might be not as true as for MCTS as for A-B-search, as komodo MCTS shows, but analysing with SF, you should be aware of it and at least give the lines a Backward, before you think them to best for sure,
To keep longer lines in hash, especially the ones, you found in mv- mode only, single variant mode will be faster for that even more.
Peter.