Razoring

[lucasart]

I did a few experiments with Razoring, and would like to share some results.

my first attempt at razoring was:

Code: Select all

	// Razoring
	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= alpha && --depth <= 0&#41;
				return score;
		&#125;
	&#125;

So, when all the usual pre-conditions are met, I reduced if qsearch failed low. With this method, I've found that it's best to reduce rather than prune. I lost the actual result of my self-play test, but I remember that they were clear as day. It's probably because this approach is flawed, and puts too much trust in the qsearch.

So I then did like everyone else (Hyatt Razoring) and compared qsearch to alpha - RazorMargin(depth) instead. However, I still wanted to test reducing against pruning in this situation. And I found that reducing is actually *better* than pruning. Result in self-testing: 321-249-297 (stopped by sequential Wald test algorithm).

Hyat Razoring & Reduce

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta
				&& --depth <= 0&#41;	// reduce
				return score;		// prune at horizon
		&#125;
	&#125;

Hyat Razoring & Prune

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta&#41;
				return score;
		&#125;
	&#125;

I wonder if other engine developers have tested it, and whether or not they came to the same conclusion. In theory, reducing is safer than pruning, so even if the elo gain is zero this modification could be added on the basis that it should improve (slightly) the results in tactical test suites.

[bob]

I did a few experiments with Razoring, and would like to share some results.

my first attempt at razoring was:

Code: Select all

	// Razoring
	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= alpha && --depth <= 0&#41;
				return score;
		&#125;
	&#125;

So, when all the usual pre-conditions are met, I reduced if qsearch failed low. With this method, I've found that it's best to reduce rather than prune. I lost the actual result of my self-play test, but I remember that they were clear as day. It's probably because this approach is flawed, and puts too much trust in the qsearch.

So I then did like everyone else (Hyatt Razoring) and compared qsearch to alpha - RazorMargin(depth) instead. However, I still wanted to test reducing against pruning in this situation. And I found that reducing is actually *better* than pruning. Result in self-testing: 321-249-297 (stopped by sequential Wald test algorithm).

Hyat Razoring & Reduce

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta
				&& --depth <= 0&#41;	// reduce
				return score;		// prune at horizon
		&#125;
	&#125;

Hyat Razoring & Prune

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta&#41;
				return score;
		&#125;
	&#125;

I wonder if other engine developers have tested it, and whether or not they came to the same conclusion. In theory, reducing is safer than pruning, so even if the elo gain is zero this modification could be added on the basis that it should improve (slightly) the results in tactical test suites.

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

[lucasart]

I did a few experiments with Razoring, and would like to share some results.

my first attempt at razoring was:

Code: Select all

	// Razoring
	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= alpha && --depth <= 0&#41;
				return score;
		&#125;
	&#125;

So, when all the usual pre-conditions are met, I reduced if qsearch failed low. With this method, I've found that it's best to reduce rather than prune. I lost the actual result of my self-play test, but I remember that they were clear as day. It's probably because this approach is flawed, and puts too much trust in the qsearch.

So I then did like everyone else (Hyatt Razoring) and compared qsearch to alpha - RazorMargin(depth) instead. However, I still wanted to test reducing against pruning in this situation. And I found that reducing is actually *better* than pruning. Result in self-testing: 321-249-297 (stopped by sequential Wald test algorithm).

Hyat Razoring & Reduce

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta
				&& --depth <= 0&#41;	// reduce
				return score;		// prune at horizon
		&#125;
	&#125;

Hyat Razoring & Prune

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta&#41;
				return score;
		&#125;
	&#125;

I wonder if other engine developers have tested it, and whether or not they came to the same conclusion. In theory, reducing is safer than pruning, so even if the elo gain is zero this modification could be added on the basis that it should improve (slightly) the results in tactical test suites.

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

i found it in the chess programming wiki, under the name Hyatt Razoring.

[lucasart]

I did a few experiments with Razoring, and would like to share some results.

my first attempt at razoring was:

Code: Select all

	// Razoring
	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= alpha && --depth <= 0&#41;
				return score;
		&#125;
	&#125;

So, when all the usual pre-conditions are met, I reduced if qsearch failed low. With this method, I've found that it's best to reduce rather than prune. I lost the actual result of my self-play test, but I remember that they were clear as day. It's probably because this approach is flawed, and puts too much trust in the qsearch.

So I then did like everyone else (Hyatt Razoring) and compared qsearch to alpha - RazorMargin(depth) instead. However, I still wanted to test reducing against pruning in this situation. And I found that reducing is actually *better* than pruning. Result in self-testing: 321-249-297 (stopped by sequential Wald test algorithm).

Hyat Razoring & Reduce

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta
				&& --depth <= 0&#41;	// reduce
				return score;		// prune at horizon
		&#125;
	&#125;

Hyat Razoring & Prune

Code: Select all

	if &#40;UseRazoring && depth <= RazorDepth
		&& !is_pv && !is_mate_score&#40;beta&#41; && !in_check&#41;
	&#123;
		if &#40;current_eval + RazorMargin&#40;depth&#41; <= alpha&#41; &#123;
			const int razor_beta = beta - RazorMargin&#40;depth&#41;;
			const int score = qsearch&#40;B, alpha, beta, 0, ply+1, is_pv, si+1&#41;;
			if &#40;score <= razor_beta&#41;
				return score;
		&#125;
	&#125;

I wonder if other engine developers have tested it, and whether or not they came to the same conclusion. In theory, reducing is safer than pruning, so even if the elo gain is zero this modification could be added on the basis that it should improve (slightly) the results in tactical test suites.

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

i found it in the chess programming wiki, under the name Hyatt Razoring.

http://www.talkchess.com/forum/viewtopic.php?t=24286

[Gerd Isenberg]

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

I coined Hyatt's Razoring in cpw, because you proposed it as mentioned by Lucas. Should I credit somebody else?

[lucasart]

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

I coined Hyatt's Razoring in cpw, because you proposed it as mentioned by Lucas. Should I credit somebody else?

the thread I mentionned suggests that it was Tord's idea in Glaurung in the first place.

[Sven]

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

I coined Hyatt's Razoring in cpw, because you proposed it as mentioned by Lucas. Should I credit somebody else?

the thread I mentionned suggests that it was Tord's idea in Glaurung in the first place.

I don't think so. It seems Bob had temporarily tried to insert very old Crafty razoring code into current Crafty (back in 2008) after he heard that Tord used razoring, and did some cluster testing to check whether that old code might improve Crafty 23.x. But the code quoted by Bob in that thread must be very old, it is not contained in any Crafty source version since 15.17 (from ~1998) available on the "cis.uab.edu" site. So there might be some misunderstanding around here. Nevertheless, the name "Hyatt razoring" does not appear to be completely wrong, only the century is missing ...

Sven

[lucasart]

I was more hoping for a discussion on the *actual* topic... If anyone has done some actual testing and has something useful to say on this problem, feel free...

So let's just call it razoring i/o Hyatt razoring, and avoid these discussions about who invented the wheel and who first called it a wheel...

[mcostalba]

the thread I mentionned suggests that it was Tord's idea in Glaurung in the first place.

Yes it is a Tord's idea.

Perhaps now Bob says that in the 70' he tried already this in an experimental Cray computer in assembly, but it doesn't matter.....it is a Tord's idea.

[bob]

Where does that "hyatt code" come from? It does not appear to look anything like the code in recent Crafty versions... I don't do "razoring" at all. I prune or reduce...

I coined Hyatt's Razoring in cpw, because you proposed it as mentioned by Lucas. Should I credit somebody else?

No, that was what I did a good while back. I thought he was looking at Crafty source, which doesn't do razoring any longer, resorting on actual forward pruning instead...