Searching using slow eval with tactical verification

[thomasahle]

Then you can cut it off at any point where all the remaining smaller terms can't add up to be larger than any of the bigger terms.

But I can't do that because I have no idea how my eval() works. It's just a bunch of matrices.

Can't you still evaluate the terms in order of largest weights? I wonder if the weights in your matrices are mostly uniform, or some are much larger than others?

[matthewlai]

Can't you still evaluate the terms in order of largest weights? I wonder if the weights in your matrices are mostly uniform, or some are much larger than others?

It would be possible if the network has no hidden layers. With 2 hidden layers it's impossible to do something like this.

Also, when evaluating full matrices I can use the CPU's SIMD instructions, which are many times faster than scalar instructions. If I had to evaluate more than 10% or so of the weights, it would be more efficient to just evaluate the whole thing.

Also, matrix-vector multiplications are usually bound by memory bandwidth (as opposed to matrix-matrix multiplications, which are usually bound by CPU), and since weights are loaded in cache line units, evaluating some of the weights may require loading most of the matrices anyways.

[Sergei S. Markoff]

SmarThink eval function is very slow so I've implemented an original lazy eval approach.

Code: Select all

int LazyEval(void)
{
	int material = thread_board.Material[thread_board.SideToMove] - thread_board.Material[ChangeSide(thread_board.SideToMove)];
	int min_lazy, max_lazy;

	if (CurSide == thread_board.SideToMove)
	{
		min_lazy = root.min_material_lazy;
		max_lazy = root.max_material_lazy;
	}
	else
	{
		min_lazy = -root.max_material_lazy;
		max_lazy = -root.min_material_lazy;
	}

	if &#40;material < min_lazy&#41;
	&#123;
		if (&#40;thread_board.PieceCount&#91;PAWN&#93;&#91;ChangeSide&#40;thread_board.SideToMove&#41;&#93; > 1 || thread_board.NumOfPieces&#91;ChangeSide&#40;thread_board.SideToMove&#41;&#93; > 9&#41; && &#40;thread_board.NumOfPieces&#91;ChangeSide&#40;thread_board.SideToMove&#41;&#93; > 5&#41;)
		&#123;
			if &#40;thread_board.SideToMove&#41;
			&#123;
				if &#40;black_pawns_bb & FULLROW&#40;1&#41;) material += 100;
				else if &#40;black_pawns_bb & FULLROW&#40;2&#41;) material += 50;
			&#125;
			else
			&#123;
				if &#40;white_pawns_bb & FULLROW&#40;6&#41;) material += 100;
				else if &#40;white_pawns_bb & FULLROW&#40;5&#41;) material += 50;
			&#125;

			if &#40;material < min_lazy&#41; return material;
		&#125;
	&#125;
	else if &#40;material > max_lazy&#41;
	&#123;
		if (&#40;thread_board.PieceCount&#91;PAWN&#93;&#91;thread_board.SideToMove&#93; > 1 || thread_board.NumOfPieces&#91;thread_board.SideToMove&#93; > 9&#41; && &#40;thread_board.NumOfPieces&#91;thread_board.SideToMove&#93; > 5&#41;)
		&#123;
			if &#40;thread_board.SideToMove&#41;
			&#123;
				if &#40;white_pawns_bb & FULLROW&#40;6&#41;) material -= 100;
				else if &#40;white_pawns_bb & FULLROW&#40;5&#41;) material -= 50;
			&#125;
			else
			&#123;
				if &#40;black_pawns_bb & FULLROW&#40;1&#41;) material -= 100;
				else if &#40;black_pawns_bb & FULLROW&#40;2&#41;) material -= 50;
			&#125;

			if &#40;material > max_lazy&#41; return material;
		&#125;
	&#125;

	return INFINITY;
&#125;

And at any next IID iteration I'm going to update this "lazy window":

Code: Select all

void RecalcLazyBounds(int root_depth)
{
	root.root_material = thread_board.Material[CurSide] - thread_board.Material[ChangeSide(CurSide)];
	root.min_material_lazy = MIN(root_move_eval[0], root.root_material) - 200 - root_depth * 5 + MIN(150, MAX(-150, root_move_eval[0] / 5));
	root.max_material_lazy = MAX(root_move_eval[0], root.root_material) + 200 + root_depth * 5 + MIN(150, MAX(-150, root_move_eval[0] / 5));
}

So if the root eval is 0 at any iteration I will have window [-200;200] for depth 0, [-205;205] at depth 1, [-250;250] at root depth 10 and so on. All evaluation outside this window will be replaced with only material eval. Surprisingly it works for me. If you're interested you can try it)
I will be glad for any feedback)