ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

There was a pawn eval issue, you might want to re-import both versions.

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

There was a pawn eval issue, you might want to re-import both versions.

Just finished the second test, +29 -16 =5. Hopefully the issue wasn't something catastrophic…

I tried to share the Wolverine 3 DCI project, but for some reason Scratch won't let me. You can get it here instead, if you want early access: https://limewire.com/d/cOaKJ#9wR7n6N7vq

TheRandomFrenchGuy wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

There was a pawn eval issue, you might want to re-import both versions.

Just finished the second test, +29 -16 =5. Hopefully the issue wasn't something catastrophic…

I tried to share the Wolverine 3 DCI project, but for some reason Scratch won't let me. You can get it here instead, if you want early access: https://limewire.com/d/cOaKJ#9wR7n6N7vq

The json takes 7.5MB uncompressed. The limit is 5MB.

waabooboo wrote:

TheRandomFrenchGuy wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

There was a pawn eval issue, you might want to re-import both versions.

Just finished the second test, +29 -16 =5. Hopefully the issue wasn't something catastrophic…

I tried to share the Wolverine 3 DCI project, but for some reason Scratch won't let me. You can get it here instead, if you want early access: https://limewire.com/d/cOaKJ#9wR7n6N7vq

The json takes 7.5MB uncompressed. The limit is 5MB.

Pretty sure it's only a small fraction of that. But whatever, you can download it from the link I sent.

ArnoHu wrote:

waabooboo wrote:

TheRandomFrenchGuy wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

The tournament's final result:

#   engine             games   score      %     W    D    L    elo    ±
1   gok_nnue             180   170.5  94.7%   163   15    2   2500   60
2   black_crow_dev       180   151.5  84.2%   139   25   16   2304   44
3   gok_classic          180   111.5  61.9%    99   25   56   2039   34
4   Wolverine2_NNUE      180   105.5  58.6%    95   21   64   2006   33
5   delta_5_dev          180    97.0  53.9%    88   18   74   1961   33
6   white_dove_dev       180    91.5  50.8%    78   27   75   1932   33
7   Wolverine_2          180    79.5  44.2%    70   19   91   1868   33
8   delta_4              180    41.0  22.8%    40    2  138   1646   38
9   shallow_blue_dev     180    37.0  20.6%    30   14  136   1619   39
10  shallow_blue         180    15.0   8.3%    13    4  163   1442   49

Wait for Wolverine 3, it is coming out soon and should be the strongest HCE engine when it does.

I am trying to get it working with DCI now, having some issues with that. Wolverine has dominated GoK in the small number of manual test games I've done (both 1 sec and 5 sec per move) – but I need the DCI integration for a larger sample size.

Once I do that, I'll run some automated tests and report back. Just a few DCI corruption issues…

Cool, competition is always welcome, without it we wouldn't be at the current level. Please use the latest GoK versions, there have been several improvements in recent two days.

That's my hope, that the new Wolverine will spur a new wave of innovations

I completely rebuilt the project, that's why it's Wolverine 3. Texel tuned eval is the same, search is completely redesigned and much more aggressive (even more so for NNUE, which I haven't attempted to port yet).

https://scratch.mit.edu/projects/828094886/editor/

Is this DCI-compatible out of the box? If I just save your board sprite and put it in a DCI project?

I know there used to be a separate DCI GoK project, but I'm not sure where that went.

Yes, the Board sprite is DCI-compatible, that's why the DCI project wasn't needed any more.

Ok, I got it. I swapped the new GoK in for the old one and reset the counter. Wolverine vs old GoK was 7.5-1.5 when I stopped it, Wolverine vs new GoK is 2-1 as of now. I will do 100 1-sec games, that should give a reasonable sample size.

I'll release the DCI sprite later today, for the main Wolverine project it'll be a little longer as I want to make some changes to the UI.

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

There was a pawn eval issue, you might want to re-import both versions.

Just finished the second test, +29 -16 =5. Hopefully the issue wasn't something catastrophic…

I tried to share the Wolverine 3 DCI project, but for some reason Scratch won't let me. You can get it here instead, if you want early access: https://limewire.com/d/cOaKJ#9wR7n6N7vq

The json takes 7.5MB uncompressed. The limit is 5MB.

Pretty sure it's only a small fraction of that. But whatever, you can download it from the link I sent.

Congrats, looks very strong indeed!

And confirmed, uncompressed 7.5MB, over the limit, you might want to empty some lists that were filled at runtime.

I only had time for a quick testrun against the current GoK Classic stable (pawn eval bugfix), 1sec. GoK won by the smallest of margins: 10.5 - 9.5. I can send you the games.

2sec games ended 2.5 - 0.5 for Classic. Much too few games to draw conclusions. Also, my local system likely differs, performance-wise.

BTW, the old DCI project might write wrong results sometimes. I think it will get updated soon. I patched my version locally.

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

TheRandomFrenchGuy wrote:

waabooboo wrote:

ArnoHu wrote:

waabooboo wrote:

ArnoHu wrote:

ArnoHu wrote:

100 games makes sense. E.g., in the recent tournament Wolverine NNUE won five games in a row against Classic, but still lost the 20 games series. Esp. 1sec games are very volatile. But yeah, I also get impatient waiting for 100 5sec games.

You may alternatively want to try the Classic Dev version at https://scratch.mit.edu/projects/1033757275 , which has more aggressive LMR. That became possible due to recent improvements.

Ok, 50 game result was +25 -17 =8 for Wolverine. I will make a few tweaks and then try another 50 against dev version.

There was a pawn eval issue, you might want to re-import both versions.

Just finished the second test, +29 -16 =5. Hopefully the issue wasn't something catastrophic…

I tried to share the Wolverine 3 DCI project, but for some reason Scratch won't let me. You can get it here instead, if you want early access: https://limewire.com/d/cOaKJ#9wR7n6N7vq

The json takes 7.5MB uncompressed. The limit is 5MB.

Pretty sure it's only a small fraction of that. But whatever, you can download it from the link I sent.

Congrats, looks very strong indeed!

And confirmed, uncompressed 7.5MB, over the limit, you might want to empty some lists that were filled at runtime.

I only had time for a quick testrun against the current GoK Classic stable (pawn eval bugfix), 1sec. GoK won by the smallest of margins: 10.5 - 9.5. I can send you the games.

2sec games ended 2.5 - 0.5 for Classic. Much too few games to draw conclusions. Also, my local system likely differs, performance-wise.

BTW, the old DCI project might write wrong results sometimes. I think it will get updated soon. I patched my version locally.

Weird, I cleared the hash tables before saving and it was just 170 kb on my computer (and 140 as I shared it). Not sure what would cause it to balloon so much, but I'll take a look.

There's still a few issues, most costly is that Wolverine can accidentally allow a repetition of moves in DCI mode. Few things to fix, then I'll try to get the NNUE in there and see if I can keep it under 5 MB…