Crude, cruder, crudest

[velmarin]

Regarding the Board, Ippo and Stockfish are equal.



typedef enum IPPOLIT
{
A1, B1, C1, D1, E1, F1, G1, H1,
A2, B2, C2, D2, E2, F2, G2, H2,
A3, B3, C3, D3, E3, F3, G3, H3,
A4, B4, C4, D4, E4, F4, G4, H4,
A5, B5, C5, D5, E5, F5, G5, H5,
A6, B6, C6, D6, E6, F6, G6, H6,
A7, B7, C7, D7, E7, F7, G7, H7,
A8, B8, C8, D8, E8, F8, G8, H8
} EnumSquares;

enum Square { STOCKFISH
SQ_A1, SQ_B1, SQ_C1, SQ_D1, SQ_E1, SQ_F1, SQ_G1, SQ_H1,
SQ_A2, SQ_B2, SQ_C2, SQ_D2, SQ_E2, SQ_F2, SQ_G2, SQ_H2,
SQ_A3, SQ_B3, SQ_C3, SQ_D3, SQ_E3, SQ_F3, SQ_G3, SQ_H3,
SQ_A4, SQ_B4, SQ_C4, SQ_D4, SQ_E4, SQ_F4, SQ_G4, SQ_H4,
SQ_A5, SQ_B5, SQ_C5, SQ_D5, SQ_E5, SQ_F5, SQ_G5, SQ_H5,
SQ_A6, SQ_B6, SQ_C6, SQ_D6, SQ_E6, SQ_F6, SQ_G6, SQ_H6,
SQ_A7, SQ_B7, SQ_C7, SQ_D7, SQ_E7, SQ_F7, SQ_G7, SQ_H7,
SQ_A8, SQ_B8, SQ_C8, SQ_D8, SQ_E8, SQ_F8, SQ_G8, SQ_H8,
SQ_NONE,

Ippolit regarding the values ​​are copied from a post by Don Dailey(in Talkchess) where Don copy these arrays in Komodo and ran some tests, I do not find, but very significant.

Originally Ippo used 8x8 arrays and 8x
8x8x combined, not 64 array.

[Lyudmil Tsvetkov]

Where is white and where is black? Where is a1 and where h8? If it is as I understand it, I must turn upside down both my thought and my vision.

Stockfish uses the same convention.

a1....h1
..
..
a8....h8

OR

h1....a1
..
..
h8...a8

// It doesn't matter. They are both laterally symmetrical.

I think you missed this and your values are all UPSIDE DOWN! This might be the reason why they were doing so badly.

Can you send an updated PSQT? Also, it would greatly help if it was a .txt. Use Notepad instead of Word.

It is much easier for you, Arjun.

You will not tell me that with completely reversed values Stockfish performed only 100 elo worse, will you?
I think I did it right the first time, as I read carefully the instructions for filling the table. I also think I understood correctly the Ippolit psqt. I would be surprised if it is otherwise. However, you never know, I already do not know where my left and my right are...

I would be afraid to look at any new results you might post if you retest with the reversed values Anyway, please share to unravel the enigma...

Ok, I did a quick test with the edited values.

The good news is that it is an improvement. The bad news is that its still pretty bad.

Code: Select all

TC = 2" + 0.02
340 - 565 - 295 [0.406] 1200
Elo difference: -66

Thanks, any longer TC games?

If it does not work better with longer TC, probably you should turn once more the board, this time from left to right. I think now that Miguel and Lucas are right, it does not matter what you do in an engine, you should just do something and then test it. Next time I will fill the table will random values, starting from the middle of the board, one column from bottom to top, and another from top to bottom. Then I will mix all values again. You never know, probably this time it will work, you do some tests and - bingo!

[arjuntemurnikar]

I didn't do longer TC as it seems to be hopeless. The other time I just did a couple of 5"+0.05 games to make sure the TC wasn't affecting the result. As I thought, it wasn't.

[Lyudmil Tsvetkov]

I didn't do longer TC as it seems to be hopeless. The other time I just did a couple of 5"+0.05 games to make sure the TC wasn't affecting the result. As I thought, it wasn't.

But still it showed 30 elo difference.

[arjuntemurnikar]

I didn't do longer TC as it seems to be hopeless. The other time I just did a couple of 5"+0.05 games to make sure the TC wasn't affecting the result. As I thought, it wasn't.

But still it showed 30 elo difference.

Yes, but the number of games were also much much less. 200 games is just not accurate enough. I stopped it because the number of losses still far outweighed the number of wins, and so it was not much of an improvement. If the test was anyway close, like if it was hovering around +-15 elo, I would have given it a shot. It seems hopeless to me. I much rather give my resources to the fishtest framework.