Fringe- Special Engine and Book Tests

[Fringe407]

Test book: UHO 2024
Time:1+1
Threads:1
Hash:64

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 killfish190324                 30      35     140   54.3%   62.9% 
   2 Marauders 3.0_avx2             15      32     140   52.1%   68.6% 
   3 Brainlearn27-x86-64-avx2        5      38     140   50.7%   57.1% 
   4 Big-SF 210324-avx2              0      39     140   50.0%   54.3% 
   5 stockfish_24032015             -2      34     140   49.6%   65.0% 
   6 INCOGNITO 7-avx2              -10      34     140   48.6%   64.3% 
   7 corchess_x64_avx2             -15      36     140   47.9%   61.4% 
   8 Raid depth--                  -22      37     140   46.8%   57.9%  

PGN:https://pixeldrain.com/u/JrqXmsuU

[Fringe407]

Test suit :UHO2024
Hash:64
Thread:1CPU
Time:1+1

Score of corchess220324 vs SF-PB-200324: 36 - 29 - 85 [0.523]
... corchess_x64_avx2 playing White: 35 - 0 - 40 [0.733] 75
... corchess_x64_avx2 playing Black: 1 - 29 - 45 [0.313] 75
... White vs Black: 64 - 1 - 85 [0.710] 150
Elo difference: 16.2 +/- 36.7, LOS: 80.7 %, DrawRatio: 56.7 %
SPRT: llr 0 (0.0%), lbound -inf, ubound inf
150 of 150 games finished.

pgn: https://pixeldrain.com/u/nCzcMEhp

[Fringe407]

Test book: UHO 2024
Time:1+1
Threads:1
Hash:64

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Big-SF 220324-avx2             22      37     140   53.2%   57.9% 
   2 Light 26.3-avx2                15      38     140   52.1%   57.1% 
   3 HypnoS-190324_x64_AVX2         15      38     140   52.1%   57.1% 
   4 killfish190324                  7      35     140   51.1%   62.1% 
   5 TacticaL Gear 3.1_avx2          0      38     140   50.0%   55.7% 
   6 Marauders 3.0_avx2            -15      37     140   47.9%   58.6% 
   7 Deep Blue                     -15      40     140   47.9%   52.9% 
   8 stockfish16.1                 -30      42     140   45.7%   47.1%   

pgn: https://pixeldrain.com/u/NNfmA9ge

[Fringe407]

Test book: UHO 2024
Time:1+1
Threads:1
Hash:64

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 corchess220324                 10      47     100   51.5%   53.0% 
   2 INCOGNITO 8                     7      45     100   51.0%   56.0% 
   3 HypnoS 240324                 -17      48     100   47.5%   51.0%    

pgn: https://pixeldrain.com/u/iDiPTXWm

[Solista]

Test book: UHO 2024
Time:1+1
Threads:1
Hash:64

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Big-SF 220324-avx2             22      37     140   53.2%   57.9% 
   2 Light 26.3-avx2                15      38     140   52.1%   57.1% 
   3 HypnoS-190324_x64_AVX2         15      38     140   52.1%   57.1% 
   4 killfish190324                  7      35     140   51.1%   62.1% 
   5 TacticaL Gear 3.1_avx2          0      38     140   50.0%   55.7% 
   6 Marauders 3.0_avx2            -15      37     140   47.9%   58.6% 
   7 Deep Blue                     -15      40     140   47.9%   52.9% 
   8 stockfish16.1                 -30      42     140   45.7%   47.1%   

pgn: https://pixeldrain.com/u/NNfmA9ge

I'm glad you're having fun testing! I looked at the source code of Hypnos 190324. The engine is not as new as the date. Most of the code is about 2 months old. It was mixed, old code with new code. That doesn't have to be bad, but old parameters were mixed with new parameters. From what I've heard, this version of Hypnos is not MZ himself. MZ provided only a basis and others are now experimenting with it.

[Fringe407]

Test book: UHO 2024
Time:1+1
Threads:1
Hash:64

Code: Select all

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Big-SF 220324-avx2             22      37     140   53.2%   57.9% 
   2 Light 26.3-avx2                15      38     140   52.1%   57.1% 
   3 HypnoS-190324_x64_AVX2         15      38     140   52.1%   57.1% 
   4 killfish190324                  7      35     140   51.1%   62.1% 
   5 TacticaL Gear 3.1_avx2          0      38     140   50.0%   55.7% 
   6 Marauders 3.0_avx2            -15      37     140   47.9%   58.6% 
   7 Deep Blue                     -15      40     140   47.9%   52.9% 
   8 stockfish16.1                 -30      42     140   45.7%   47.1%   

pgn: https://pixeldrain.com/u/NNfmA9ge

I'm glad you're having fun testing! I looked at the source code of Hypnos 190324. The engine is not as new as the date. Most of the code is about 2 months old. It was mixed, old code with new code. That doesn't have to be bad, but old parameters were mixed with new parameters. From what I've heard, this version of Hypnos is not MZ himself. MZ provided only a basis and others are now experimenting with it.

My dear friend Eduard,

I honestly didn't know that. I'm going to use the Hypnos240324 version in my new test. Where I got the engines is actually from the source file. Link: https://github.com/Zerbinati/HypnoS/releases/tag/240324

Besides, your engine series are very valuable to me. I beat some cowards with your engines in lichess.
On the other hand, his tactical solving percentage surprises me. Following you and testing your engines is a fun activity for me mate.

I use various special books for my lichess bot. I came across your book on a torrent site a long time ago and downloaded it. I tested it and it was very powerful. If I'm not mistaken, it was solista attack 8.1. Then the file was deleted. Until now, if that book is yours, it has never been defeated, my dear friend.

If I talk about Light 26.3, it beat your other engine, which was the 1st engine in the test, by a small margin in another test. I sent you a private e-mail with the file and the result, I hope you saw it.

Best regards,
Fringe

[Solista]

Like I said, it's a mix of old and new in the Hypnos 190324 version. Just two quick comparisons, the hypnos code here is 2 months older than the SF dev code.

SF dev:

for (size_t i = 1; i < reductions.size(); ++i)
reductions = int((19.80 + std::log(size_t(options["Threads"])) / 2) * std::log(i));

Hypnos dev

for (int i = 1; i < MAX_MOVES; ++i)
Reductions = int((20.37 + std::log(Threads.size()) / 2) * std::log(i));

---

SF dev.

Depth Search::Worker::reduction(bool i, Depth d, int mn, int delta) {
int reductionScale = reductions[d] * reductions[mn];
return (reductionScale + 1091 - delta * 759 / rootDelta) / 1024 + (!i && reductionScale > 950);

Hypnos dev.

Depth reduction(bool i, Depth d, int mn, int delta, int rootDelta) {
int reductionScale = Reductions[d] * Reductions[mn];
return (reductionScale + 1346 - int(delta) * 896 / int(rootDelta)) / 1024
+ (!i && reductionScale > 880);

----------------------------------------------------------------------------------------------------------------------------------

My Solista Attack book is commercial, I sell it, it's not free. I have two versions, BIN and CTG. Which one do you use? The current version is 8.4.

[Solista]

HypnoS 240324
Allign to: "VVLTC search tune"

That is not correct.

Here only one example, there are many more:

SF dev. "VVLTC search tune":

Depth Search::Worker::reduction(bool i, Depth d, int mn, int delta) {
int reductionScale = reductions[d] * reductions[mn];
return (reductionScale + 1107 - delta * 725 / rootDelta) / 1024 + (!i && reductionScale > 956);


HypnoS 240324:

// Reductions lookup table initialized at startup
int Reductions[MAX_MOVES]; // [depth or moveNumber]

Depth reduction(bool i, Depth d, int mn, int delta, int rootDelta) {
int reductionScale = Reductions[d] * Reductions[mn];
return (reductionScale + 1346 - int(delta) * 896 / int(rootDelta)) / 1024
+ (!i && reductionScale > 880);


SF dev. VVLTC search tune

for (size_t i = 1; i < reductions.size(); ++i)
reductions = int((19.80 + std::log(size_t(options["Threads"])) / 2) * std::log(i));

HypnoS 240324 VVLTC search tune

for (int i = 1; i < MAX_MOVES; ++i)
Reductions = int((20.37 + std::log(Threads.size()) / 2) * std::log(i));


HypnoS 240324 is (fake) OLD SF CODE "VVLTC search tune"


-------------------------------------------------------------------------

[Fringe407]

HypnoS 240324
Allign to: "VVLTC search tune"

That is not correct.

Here only one example, there are many more:

SF dev. "VVLTC search tune":

Depth Search::Worker::reduction(bool i, Depth d, int mn, int delta) {
int reductionScale = reductions[d] * reductions[mn];
return (reductionScale + 1107 - delta * 725 / rootDelta) / 1024 + (!i && reductionScale > 956);


HypnoS 240324:

// Reductions lookup table initialized at startup
int Reductions[MAX_MOVES]; // [depth or moveNumber]

Depth reduction(bool i, Depth d, int mn, int delta, int rootDelta) {
int reductionScale = Reductions[d] * Reductions[mn];
return (reductionScale + 1346 - int(delta) * 896 / int(rootDelta)) / 1024
+ (!i && reductionScale > 880);


SF dev. VVLTC search tune

for (size_t i = 1; i < reductions.size(); ++i)
reductions = int((19.80 + std::log(size_t(options["Threads"])) / 2) * std::log(i));

HypnoS 240324 VVLTC search tune

for (int i = 1; i < MAX_MOVES; ++i)
Reductions = int((20.37 + std::log(Threads.size()) / 2) * std::log(i));


HypnoS 240324 is (fake) OLD SF CODE "VVLTC search tune"


-------------------------------------------------------------------------

My dear friend Eduard,

It seems that these codes are really about 2 months old. We have a small request to Marco Zebinati, we want a new version of your own.

Also, I didn't know your book was for sale and as I said, I found it on a torrent site. It says solista attack 8.1.bin. I can delete the book if you don't want me to use it. I trusted the book when I saw that it belonged to you. It's an old version, but it's still very good.

Best regards,
Fringe

[Solista]

Keep the book! I would give it to you for free too.

---------------------------------------------------------------

This line is funny, which was changed by the developers to VVLTC search tune:

SF dev "VVLTC search tune" changed:

Depth Search::Worker::reduction(bool i, Depth d, int mn, int delta) {
int reductionScale = reductions[d] * reductions[mn];
- return (reductionScale + 1091 - delta * 759 / rootDelta) / 1024 + (!i && reductionScale > 950);
+ return (reductionScale + 1107 - delta * 725 / rootDelta) / 1024 + (!i && reductionScale > 956);


And what does it say in HypnoS 240324 VVLTC search tune? So:

Depth reduction(bool i, Depth d, int mn, int delta, int rootDelta) {
int reductionScale = Reductions[d] * Reductions[mn];
return (reductionScale + 1346 - int(delta) * 896 / int(rootDelta)) / 1024
+ (!i && reductionScale > 880);

Absolute wrong (very old SF code, no VVLTC search tune).