Mate Score

[tomitank]

Hi!

Which one is better? (Why?)
I'm using Fail-Soft, Aspiration Windows (50), PVS, NMP, Hash Table, LMR, Razoring, Futility Pruning, LMP.

Vice 1.0:

Code: Select all

#define MAXDEPTH 64
#define INFINITE 30000
#define ISMATE (INFINITE - MAXDEPTH)

Sungours 1.4:

Code: Select all

#define MAX_PLY         64
#define INF             32767
#define MATE            32000
#define MAX_EVAL        2999

Tamas

[hgm]

The first seems wrong: mate distance can easily be larger than MAXPLY (assuming that the latter is the ultimate limit t the search depth). Engines with TT can see mates beyond the horizon through grafting.

[tomitank]

For the first one:

King value: 50000

Save tt:

Code: Select all

if (score > ISMATE) {
	score += boardPly;
&#125; else if &#40;score < -ISMATE&#41; &#123;
	score -= boardPly;
&#125;

Score from tt:

Code: Select all

				
if (value > ISMATE) {
	value -= boardPly;
&#125; else if &#40;value < -ISMATE&#41; &#123;
	value += boardPly;
&#125;

Video:
https://youtu.be/clLaVOnvUvA?t=2m17s

But Vice use fail hard.

With fail-soft, is can it still wrong?

Tamas

[Nordlandia]

How does excessive king value influence overall gameplay?

For example King value: 5,000,000,000,000

[tomitank]

I do not know.

Maybe, if there is no king for some mistake?

This is all misty for me.

[hgm]

But Vice use fail hard.

With fail-soft, is can it still wrong?

Yes, it is still wrong. The necessary adjustment for mate scores would not take place for mate scores between -ISMATE and ISMATE.

[hgm]

How does excessive king value influence overall gameplay?

For example King value: 5,000,000,000,000

King value should not affect game play. Fairy-Max uses king value = -1, and in search recognizes victims with negative value as royal, terminating the branch with +INF score if the value is -1.

[Nordlandia]

How does excessive king value influence overall gameplay?

For example King value: 5,000,000,000,000

King value should not affect game play. Fairy-Max uses king value = -1, and in search recognizes victims with negative value as royal, terminating the branch with +INF score if the value is -1.

Some earlier computer chess programs aligned the king an arbitrary large value (1,000,000,000 points) As a result the King remained on the first rank throughout the entire game

That number indicate king safetey is paramount, so the King shouldn't be centralized and rule out the possibility to get mated.

In a nutshell - in a regular pawn ending, the king messed around in the corner instead of heading to the center of the board.

Those days are over.

[hgm]

The value of a piece should not determine where it prefers to go. In my first chess program I assumed it would be sufficient to give the king a very high value to keep it out of check. Instead it just started to reply to checks with counter-checks, in an attempt to trade kings. You really have to program that there is no opportunity for retaliation after you lost your king.

[JVMerlino]

How does excessive king value influence overall gameplay?

For example King value: 5,000,000,000,000

King value should not affect game play. Fairy-Max uses king value = -1, and in search recognizes victims with negative value as royal, terminating the branch with +INF score if the value is -1.

If you allow the King to capture a piece in SEE, then isn't a very high King value necessary?