Repetitions in the Search Tree

[bob]

I see - hmm. But, that assumes that me and my opponent played our best so far until before the search-root. Is that an okay assumption to make in practice?

One example would be that my opponent blundered and let me repeat the position once - but won't allow it a second time? I know that's quite unlikely with chess engines....

How could it not be best? You have to cooperate to repeat, unless you have no choice in the matter. If you have no choice, then it is a draw. If you do have a choice, you should exercise it the first time the repetition is seen.

[Luis Babboni]

I see - hmm. But, that assumes that me and my opponent played our best so far until before the search-root. Is that an okay assumption to make in practice?

One example would be that my opponent blundered and let me repeat the position once - but won't allow it a second time? I know that's quite unlikely with chess engines....

How could it not be best? You have to cooperate to repeat, unless you have no choice in the matter. If you have no choice, then it is a draw. If you do have a choice, you should exercise it the first time the repetition is seen.

In for example 5/0 with time advantage being the opponnent near to lost by time, the 2nd repetition could be used to make him lose a little more time.
I couldnt imagine another situation to use it for the moment.

[flok]

I'm puzzled: if I always use score 0 and not do any searching when there's a history-hit in my history std::unordered_set, then I get this (from the starting position):

Code: Select all

info depth 1 seldepth 1 score cp 40 time 1 nodes 27 pv e2e4
info depth 2 seldepth 2 score cp 0 time 1 nodes 115 pv e2e4 e7e5 f1b5
info depth 3 seldepth 3 score cp 34 time 2 nodes 242 pv e2e4 e7e5 d2d4
info depth 4 seldepth 4 score cp 5 time 10 nodes 940 pv a2a4 e7e5 e2e4 a7a6
info depth 5 seldepth 5 score cp 31 time 61 nodes 5336 pv e2e3 e7e5 d1g4 h7h5 a2a4 h5g4
info depth 6 seldepth 6 score cp 29 time 152 nodes 12522 pv d2d4 e7e6 d1d3 h7h5 d3h7 h8h7 h7h8 d6h2
info depth 7 seldepth 7 score cp 33 time 310 nodes 24620 pv d2d4 e7e6 e2e4 b8c6 b1c3 f8d6 d1f3 c6d4 f3f7
info depth 8 seldepth 8 score cp 29 time 607 nodes 47067 pv d2d4 e7e6 d1d3 b8c6 b1c3 f8d6 d4d5 c6b4 d5e6
info depth 9 seldepth 9 score cp 22 time 1469 nodes 108242 pv d2d4 e7e6 d1d3 d8f6 b1c3 f8d6 d4d5 g8e7 g1f3 e6d5 c3d5 e7d5 d3d5

That 4th ply is worrying me; a2-a4?!

[jwes]

I'm puzzled: if I always use score 0 and not do any searching when there's a history-hit in my history std::unordered_set, then I get this (from the starting position):

Code: Select all

info depth 1 seldepth 1 score cp 40 time 1 nodes 27 pv e2e4
info depth 2 seldepth 2 score cp 0 time 1 nodes 115 pv e2e4 e7e5 f1b5
info depth 3 seldepth 3 score cp 34 time 2 nodes 242 pv e2e4 e7e5 d2d4
info depth 4 seldepth 4 score cp 5 time 10 nodes 940 pv a2a4 e7e5 e2e4 a7a6
info depth 5 seldepth 5 score cp 31 time 61 nodes 5336 pv e2e3 e7e5 d1g4 h7h5 a2a4 h5g4
info depth 6 seldepth 6 score cp 29 time 152 nodes 12522 pv d2d4 e7e6 d1d3 h7h5 d3h7 h8h7 h7h8 d6h2
info depth 7 seldepth 7 score cp 33 time 310 nodes 24620 pv d2d4 e7e6 e2e4 b8c6 b1c3 f8d6 d1f3 c6d4 f3f7
info depth 8 seldepth 8 score cp 29 time 607 nodes 47067 pv d2d4 e7e6 d1d3 b8c6 b1c3 f8d6 d4d5 c6b4 d5e6
info depth 9 seldepth 9 score cp 22 time 1469 nodes 108242 pv d2d4 e7e6 d1d3 d8f6 b1c3 f8d6 d4d5 g8e7 g1f3 e6d5 c3d5 e7d5 d3d5

That 4th ply is worrying me; a2-a4?!

I would worry more about the 5th ply. It looks like your program thinks dropping its Queen is good. Might there be a problem with printing or storing the PV?

[flok]

I reverted the 1-history-hit thing and this problem happens also without it. So in some change a bug was introduced.
Bring on the bisect!
Thanks, I had not noticed it.

----

Ok it was a too restrictive LMR setting.

info depth 9 seldepth 9 score cp 29 time 5676 nodes 417544 pv d2d4 d7d5 c1f4 c8f5 e2e3 f5c2 d1c2 a7a6 f4c7 d8c7

I wonder about that a7a6 though. I looked at the moves in xboard (great help) but I have no idea if it is sane or not.

[flok]

Uch. In my previous post I talked about revision 2565.
Just tried rev 2483:

Code: Select all

info depth 7 seldepth 7 score cp 20 time 1347 nodes 165262 pv d2d4 d7d5 c1f4 b8c6 b1c3 [b]a7a6[/b] d1d4
info depth 8 seldepth 8 score cp 14 time 2985 nodes 346492 pv d2d4 d7d5 d1d3 d8d6 d3b5 c7c6 b5d3 [b]a7a6[/b] h1h2
info depth 9 seldepth 9 score cp 13 time 5479 nodes 616912 pv d2d4 d7d5 d1d3 d8d6 e2e4 b8c6 e4e5 d6b4 b1c3 c6d4 d3h7

Revision 2205 too:

Code: Select all

info depth 1 seldepth 1 score cp 40 time 3 nodes 21 pv e2e4
info depth 2 seldepth 2 score cp 0 time 5 nodes 158 pv e2e4 e7e5 f1b5
info depth 3 seldepth 3 score cp 34 time 10 nodes 472 pv e2e4 e7e5 d2d4
info depth 4 seldepth 4 score cp 30 time 50 nodes 3416 pv d2d4 e7e5 d4e5
info depth 5 seldepth 5 score cp 19 time 194 nodes 20543 pv e2e3 d7d5 c2c3 c8d7
info depth 6 seldepth 6 score cp 21 time 604 nodes 65752 pv e2e3 d7d5 f1d3
info depth 7 seldepth 7 score cp 9 time 1997 nodes 210546 pv e2e3 d7d5 b1c3
info depth 8 seldepth 8 score cp 14 time 4910 nodes 517304 pv e2e4 d7d5 e4e5 a7a6
info depth 9 seldepth 9 score cp 20 time 20359 nodes 2060685 pv d2d4 e7e6 g1f3 d8f6 e2e4 b8c6 b1c3 f8b4

sigh