BertTuyt wrote:I observed the same, that when playing only with a material based function and endgame databases, I could sometimes draw against competition.
However when losing it was most related to long-term disadvantages far beyond the search horizon (like breakthrough, wing locks, and outposts).
Material only always looses, at least against the top programs. I tried a few times to play against Kingsrow with material only and after each move I see the score from Kingsrow go up and the game is over in less than 20 moves.
On the other hand i also belief when one (for example) constantly could reach 30 ply, than the evaluation knowledge which always should guarantee a draw, could be minimal.
In other words, there must be diminishing returns in the evaluation when search depth explodes.
With current hardware it is not doable to reach 30 ply during the midgame, 22 ply is reachable and with pruning like LMR you will get a few plies more but this is what I call fake depth, something out of one's own pocket, it helps a little because you examine the PV somewhat deeper but you can easily miss a refutation along the line. To get at 30 ply we need computers at least a 100 times faster than we have right now.
I have the impression that many lines in draughts are forced lines, very small paths, the PV almost never changes no matter how deep you look. For instance this is what I see when I look at the position above.
Code: Select all
m m m m m
m m m m m
m m m - m
- m m - -
m - m M m
- - - M -
M M - M -
M M M M -
M M M M M
M M M M M
info depth 1 score 34 time 0.00 nodes 25 nps 2096269 21-26
info depth 2 score 20 time 0.00 nodes 116 nps 975456 21-27 31x22
info depth 3 score 122 time 0.00 nodes 270 nps 1202412 23-28 32x23 21-27
info depth 4 score 122 time 0.00 nodes 516 nps 1347282 23-28 32x23 21-27 31x22
info depth 5 score 122 time 0.00 nodes 1194 nps 2052797 23-28 32x23 21-27 31x22 17x30
info depth 6 score 111 time 0.00 nodes 2221 nps 2116279 23-28 32x23 21-27 31x22 17x30 45-40
info depth 7 score 139 time 0.00 nodes 5227 nps 3306764 23-28 32x23 21-27 31x22 17x30 36-31 10-14
info depth 8 score 127 time 0.00 nodes 9163 nps 3607158 23-28 32x23 21-27 31x22 17x30 44-40 30-35 39-33
info depth 9 score 127 time 0.00 nodes 20388 nps 4699516 23-28 32x23 21-27 31x22 17x30 44-40 30-35 39-33
info depth 10 score 127 time 0.01 nodes 37119 nps 5598807 23-28 32x23 21-27 31x22 17x30 44-40 30-35 39-33
info depth 11 score 135 time 0.01 nodes 73721 nps 6042077 23-28 32x23 21-27 31x22 17x30 44-40 30-35 39-33
info depth 12 score 137 time 0.02 nodes 147604 nps 6630008 23-28 32x23 21-27 31x22 17x30 44-40 10-14 37-31
info depth 13 score 138 time 0.04 nodes 293878 nps 6582256 23-28 32x23 21-27 31x22 17x30 45-40 10-14 40-35
info depth 14 score 132 time 0.09 nodes 661227 nps 7341795 23-28 32x23 21-27 31x22 17x30 44-40 10-14 37-32
info depth 15 score 136 time 0.15 nodes 1213086 nps 8256232 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 16 score 129 time 0.23 nodes 2445233 nps 10664527 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 17 score 135 time 0.31 nodes 5047768 nps 16350136 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 18 score 131 time 0.37 nodes 9816334 nps 26331942 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 19 score 136 time 0.60 nodes 27964746 nps 46517172 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 20 score 129 time 0.81 nodes 46266014 nps 57061703 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 21 score 130 time 1.10 nodes 70860855 nps 64145878 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 22 score 131 time 1.71 nodes 124061777 nps 72423342 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 23 score 127 time 3.02 nodes 239948226 nps 79335806 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 24 score 127 time 4.77 nodes 396015913 nps 82967064 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 25 score 127 time 7.95 nodes 680145502 nps 85537166 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 26 score 123 time 18.88 nodes 1673693929 nps 88660482 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 27 score 126 time 29.90 nodes 2660314561 nps 88961789 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 28 score 124 time 62.84 nodes 5596364835 nps 89062791 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 29 score 123 time 112.39 nodes 10057648407 nps 89487688 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 30 score 122 time 211.20 nodes 18962371747 nps 89784704 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 31 score 123 time 361.23 nodes 32575815922 nps 90181268 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 32 score 121 time 772.74 nodes 69914708720 nps 90476713 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 33 score 123 time 1346.77 nodes 121652659669 nps 90329011 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 34 score 121 time 2808.52 nodes 251428779302 nps 89523521 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
info depth 35 score 122 time 5634.75 nodes 497460485349 nps 88284428 23-28 32x23 21-27 31x22 17x30 37-32 10-14 41-37
But it has often stated here, not losing is not the problem, the challenge is winning....
Maybe the game is not winnable at all when the opponent doesn't make gros mistakes. When I see that I can get several draws against the top engines with a basic search and a simple evaluation function it makes me wonder how large the draw margin is.
Joost