Hotspots

[Lyudmil Tsvetkov]

It looks like it will never pass.
I do not like such patches at all. At one point it was at 2.72 LLR, just 10 more wins and would be through, but now back again at 1.42.


From the info we also have from Joerg's patch, I know what needs to be corrected, so that the patch performs better - increase somewhat queen mg and eg value, or maybe just not decrease it from original values.

If I had another attempt to push, I would go with values like:

PawnValueMg = 194, PawnValueEg = 256,
KnightValueMg = 817, KnightValueEg = 846,
BishopValueMg = 840, BishopValueEg = 865,
RookValueMg = 1278, RookValueEg = 1294,
QueenValueMg = 2518 QueenValueEg = 2555

or, even a bit more cps for the queen:

PawnValueMg = 194, PawnValueEg = 256,
KnightValueMg = 817, KnightValueEg = 846,
BishopValueMg = 840, BishopValueEg = 865,
RookValueMg = 1278, RookValueEg = 1294,
QueenValueMg = 2520 QueenValueEg = 2559

Btw., now, already too late, well beyond 6 futile attempts, I also learned what was wrong with the rook psqt: lack of synchrony between mg and eg values.

As older values were almost flat for the mg and fully flat fro the eg, this worked quite normal, although values were more or less crude; when we add rank bonus, but do this just for the mg, already inconsistency problems arise with interpolation. So, the natural thing to be done for the tables to perform better, is to just use the same tables for mg and eg with rank-based approach.

Unfortunately, it is too late now.

[Lyudmil Tsvetkov]

Actually, my preferred try is this one, with the bigger queen value:

PawnValueMg = 194, PawnValueEg = 256,
KnightValueMg = 818, KnightValueEg = 846,
BishopValueMg = 840, BishopValueEg = 865,
RookValueMg = 1278, RookValueEg = 1294,
QueenValueMg = 2520 QueenValueEg = 2560

One way or another, the original SF values look very ugly, for example the bonus for the bishop increases from mg to eg less than the bonus for the knight, very ugly indeed, but obviously one can not go very far apart from original values without disturbing drastically the overall balance.

[lantonov]

These are the rook psqt values from the SPSA:

Code: Select all

{ // Rook
   S(-23, 3), S(-20, 3), S(-13, 3), S(-7, 3), S(-7, 3), S(-13, 3), S(-20, 3), S(-23, 3), 
   S(-18, 3), S(-13, 3), S(  3, 3), S( 2, 3), S( 2, 3), S(  3, 3), S(-13, 3), S(-18, 3), 
   S(-22, 3), S( -4, 3), S( -6, 3), S(-6, 3), S(-6, 3), S( -6, 3), S( -4, 3), S(-22, 3), 
   S(-18, 3), S(-13, 3), S( -5, 3), S( 6, 3), S( 6, 3), S( -5, 3), S(-13, 3), S(-18, 3), 
   S(-26, 3), S(-14, 3), S(  7, 3), S( 3, 3), S( 3, 3), S(  7, 3), S(-14, 3), S(-26, 3), 
   S(-21, 3), S(-16, 3), S( -2, 3), S( 3, 3), S( 3, 3), S( -2, 3), S(-16, 3), S(-21, 3), 
   S(-17, 3), S( -3, 3), S(  3, 3), S(12, 3), S(12, 3), S(  3, 3), S( -3, 3), S(-17, 3), 
   S(-17, 3), S(-19, 3), S(-19, 3), S(-5, 3), S(-5, 3), S(-19, 3), S(-19, 3), S(-17, 3)
  },

Do they make sense?
I suppose not much. These, however, will be improved by-and-by. Mindbreaker (on Fishcooking) proposes to make SPSA rank-by-rank or file-by-file. I think file-by-file is more reasonable. Also, not sure if psqt should be symmetric. I think not: a-file is not the same as h-file.

[Lyudmil Tsvetkov]

These are the rook psqt values from the SPSA:

Code: Select all

{ // Rook
   S(-23, 3), S(-20, 3), S(-13, 3), S(-7, 3), S(-7, 3), S(-13, 3), S(-20, 3), S(-23, 3), 
   S(-18, 3), S(-13, 3), S(  3, 3), S( 2, 3), S( 2, 3), S(  3, 3), S(-13, 3), S(-18, 3), 
   S(-22, 3), S( -4, 3), S( -6, 3), S(-6, 3), S(-6, 3), S( -6, 3), S( -4, 3), S(-22, 3), 
   S(-18, 3), S(-13, 3), S( -5, 3), S( 6, 3), S( 6, 3), S( -5, 3), S(-13, 3), S(-18, 3), 
   S(-26, 3), S(-14, 3), S(  7, 3), S( 3, 3), S( 3, 3), S(  7, 3), S(-14, 3), S(-26, 3), 
   S(-21, 3), S(-16, 3), S( -2, 3), S( 3, 3), S( 3, 3), S( -2, 3), S(-16, 3), S(-21, 3), 
   S(-17, 3), S( -3, 3), S(  3, 3), S(12, 3), S(12, 3), S(  3, 3), S( -3, 3), S(-17, 3), 
   S(-17, 3), S(-19, 3), S(-19, 3), S(-5, 3), S(-5, 3), S(-19, 3), S(-19, 3), S(-17, 3)
  },

Do they make sense?
I suppose not much. These, however, will be improved by-and-by. Mindbreaker (on Fishcooking) proposes to make SPSA rank-by-rank or file-by-file. I think file-by-file is more reasonable. Also, not sure if psqt should be symmetric. I think not: a-file is not the same as h-file.

Well, we will shortly see.

Considering however that squares on the 7th rank like a7 and b7 get penalised, 3rd rank gets bigger bonus than 4th and 5th ranks, c and f files for some squares get bigger value than d and e files...I really do not know

[lantonov]

Well, we will shortly see.

Considering however that squares on the 7th rank like a7 and b7 get penalised, 3rd rank gets bigger bonus than 4th and 5th ranks, c and f files for some squares get bigger value than d and e files...I really do not know

Bench was too high (9084418) compared to master (7086031) which does not bode well, but hope dies last ...

[lantonov]

It looks like it will never pass.
I do not like such patches at all. At one point it was at 2.72 LLR, just 10 more wins and would be through, but now back again at 1.42.

Piece values have chance to pass STC. More probably it will fail yellow which is the equivalent of Levski losing 0:1 against Barcelona in Spain.

[Lyudmil Tsvetkov]

It looks like it will never pass.
I do not like such patches at all. At one point it was at 2.72 LLR, just 10 more wins and would be through, but now back again at 1.42.

Piece values have chance to pass STC. More probably it will fail yellow which is the equivalent of Levski losing 0:1 against Barcelona in Spain.

Do not talk BS.

Levski never loses.

I do not see the point of a patch succeeding after more than 80 000 games; the book has already been spanned, and everything else is just external unwanted influence on the result.

If I had the chance, I would stop any patch after it spans the whole book and sprt is still not finished. Authors are there to find improvements to their patches.

However, if anyone ever pushes my last suggested values, with slightly increased queen value, I bet it will pass very fast.

Concerning rook psqt, the main problem is again not so much the inaccuracies of the values, but the discontinuity between flat eg and non-flat mg values, which seemingly strongly distorts interpolation. Pity I thought about it too late. Obviously, in the eg the rook could be bonused in very much the same way as in the mg, based on chess knowledge.

[lantonov]

Concerning rook psqt, the main problem is again not so much the inaccuracies of the values, but the discontinuity between flat eg and non-flat mg values, which seemingly strongly distorts interpolation. Pity I thought about it too late. Obviously, in the eg the rook could be bonused in very much the same way as in the mg, based on chess knowledge.

The flat eg values poked my eyes from the beginning but human reasoning stopped me from changing those. I thought: what does it matter which square is the rook in the endgame? It seems that it matters for the computer.
However, the problem with many values in SPSA is the noise, and maybe these raw first data are very noisy -- 32 values are on the upper end. In the beginning, I was going to do SPSA with all 128 values but people persuaded me against this. I wonder what should be next. The logical thing to do is to SPSA only the eg values against master.

[lantonov]

I do not see the point of a patch succeeding after more than 80 000 games; the book has already been spanned, and everything else is just external unwanted influence on the result.

If I had the chance, I would stop any patch after it spans the whole book and sprt is still not finished. Authors are there to find improvements to their patches.

However, if anyone ever pushes my last suggested values, with slightly increased queen value, I bet it will pass very fast.

If this patch passes, maybe it would not be too brazen to push a patch with changed queen values.

[karger]

You seem to make a lot of posts and then you are the only one answering them over and over again.