null-move + verification

[bob]

Tried two tests overnight.

test 1, normal null-move + verification search. This is again too close to call without more games, but seems to be about 2 Elo worse.

test 2. unrestricted null move (as discussed in the other thread) + verification search. This is _also_ about 2 Elo worse than the normal crafty.

It seems that any of these changes has a very small negative effect, whether it be verification, unrestricted nulls or both. And all three of those cases seem to be about the same, Elo-wise...

Verification is "off my to-do list" now.

BTW, for those that care, if you do the verification, then KP vs K is a trivial win, and KR vs K is a faster win. But for those that care about overall effect, it is worse.

[jwes]

Tried two tests overnight.

test 1, normal null-move + verification search. This is again too close to call without more games, but seems to be about 2 Elo worse.

test 2. unrestricted null move (as discussed in the other thread) + verification search. This is _also_ about 2 Elo worse than the normal crafty.

It seems that any of these changes has a very small negative effect, whether it be verification, unrestricted nulls or both. And all three of those cases seem to be about the same, Elo-wise...

Verification is "off my to-do list" now.

BTW, for those that care, if you do the verification, then KP vs K is a trivial win, and KR vs K is a faster win. But for those that care about overall effect, it is worse.

Have you tried verification only in positions where each side has pawns but no or only one piece? These are the positions where zugzwang is most likely to occur, and it makes sense to me to only try verification there.
Is there an available version of crafty with null move verification so I could try it myself?

[michiguel]

Tried two tests overnight.

test 1, normal null-move + verification search. This is again too close to call without more games, but seems to be about 2 Elo worse.

test 2. unrestricted null move (as discussed in the other thread) + verification search. This is _also_ about 2 Elo worse than the normal crafty.

It seems that any of these changes has a very small negative effect, whether it be verification, unrestricted nulls or both. And all three of those cases seem to be about the same, Elo-wise...

Verification is "off my to-do list" now.

BTW, for those that care, if you do the verification, then KP vs K is a trivial win, and KR vs K is a faster win. But for those that care about overall effect, it is worse.

You just proved that you should include verification in your program! 2 elo points does not justify the problems that may arise when the engine is used for analysis.

Miguel

[bob]

Tried two tests overnight.

test 1, normal null-move + verification search. This is again too close to call without more games, but seems to be about 2 Elo worse.

test 2. unrestricted null move (as discussed in the other thread) + verification search. This is _also_ about 2 Elo worse than the normal crafty.

It seems that any of these changes has a very small negative effect, whether it be verification, unrestricted nulls or both. And all three of those cases seem to be about the same, Elo-wise...

Verification is "off my to-do list" now.

BTW, for those that care, if you do the verification, then KP vs K is a trivial win, and KR vs K is a faster win. But for those that care about overall effect, it is worse.

Have you tried verification only in positions where each side has pawns but no or only one piece? These are the positions where zugzwang is most likely to occur, and it makes sense to me to only try verification there.
Is there an available version of crafty with null move verification so I could try it myself?

I have not made it public, but could certainly send you a copy. I can try the verification only when pieces(side) == 0 to see if it makes a difference.

[bob]

Tried two tests overnight.

test 1, normal null-move + verification search. This is again too close to call without more games, but seems to be about 2 Elo worse.

test 2. unrestricted null move (as discussed in the other thread) + verification search. This is _also_ about 2 Elo worse than the normal crafty.

It seems that any of these changes has a very small negative effect, whether it be verification, unrestricted nulls or both. And all three of those cases seem to be about the same, Elo-wise...

Verification is "off my to-do list" now.

BTW, for those that care, if you do the verification, then KP vs K is a trivial win, and KR vs K is a faster win. But for those that care about overall effect, it is worse.

You just proved that you should include verification in your program! 2 elo points does not justify the problems that may arise when the engine is used for analysis.

Miguel

2 here, 2 there, before you know it, you are a patzer. But in any case, you are focusing on a position. But what about all the positions where null-everywhere is better?