New engine releases 2020

[Terje]

Weiss 0.10 released with +250 self-play elo

https://github.com/TerjeKir/weiss/releases/tag/v0.10

[Graham Banks]

Excellent progress!

Would have been way too strong in my Division 6 tourney.

[mvanthoor]

That'd put Weiss somewhere around 2900 ELO if I remember correctly. That's a huge jump. Congratulations.

[Gabor Szots]

Weiss 0.10 released with +250 self-play elo

Incredible! I hope I'll be able to confirm that.

[xr_a_y]

That'd put Weiss somewhere around 2900 ELO if I remember correctly. That's a huge jump. Congratulations.

Looks like new Weiss is around 2900 indeed

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 nemorino                      167      36     284   72.4%   29.2%
   2 igel-last                      71      33     283   60.1%   33.2%
   3 minic_dev                      47      35     284   56.7%   27.5%
   4 Winter                        -32      35     284   45.4%   27.5%
   5 rodentIV                      -90      34     285   37.4%   31.2%
   6 weiss                        -163      37     284   28.2%   26.1%

[Terje]

Thanks all

I hope so too @Gabor

Still too soon to hang with the big boys haha @xr_a_y

Tuning eval params by itself probably gained +-150 elo

[mvanthoor]

Still too soon to hang with the big boys haha @xr_a_y

At some point, you'll get there, because the big boys are experiencing diminishing returns. Their progress is going to be slower and slower.

The one obstacle to reaching that level is time and resources. When you get to that level, every single code change has to be tested by 5000 games or so.

Tuning eval params by itself probably gained +-150 elo

I'll have to look into that tuning stuff when I get there. I can imagine that settings for parameters that worked well when the engine was 2100 ELO, don't work well after it gained 30 more evaluation parameters, 5 new search enhancements, and is now over 2800 ELO.

[Terje]

Still too soon to hang with the big boys haha @xr_a_y

At some point, you'll get there, because the big boys are experiencing diminishing returns. Their progress is going to be slower and slower.

The one obstacle to reaching that level is time and resources. When you get to that level, every single code change has to be tested by 5000 games or so.

Tuning eval params by itself probably gained +-150 elo

I'll have to look into that tuning stuff when I get there. I can imagine that settings for parameters that worked well when the engine was 2100 ELO, don't work well after it gained 30 more evaluation parameters, 5 new search enhancements, and is now over 2800 ELO.

Thanks to Andrew Grant and noobpwnftw, I have access to the necessary resources.

[mvanthoor]

Thanks to Andrew Grant and noobpwnftw, I have access to the necessary resources.

Maybe, some day, we'll both be on that platform. It'll take a while though

First, I'll have to upgrade my current computer for local testing at some point. The 6700K that was released almost 5 years ago is still fine for everything I do, except for testing chess engines after a code change. I'm still amazed how people got anything written that actually worked in the 70's and 80's...

[voffka]

Thanks to Andrew Grant and noobpwnftw, I have access to the necessary resources.

With an access to OpenBench resources I think you will reach 3100+ quite easily, as the biggest challenge for "simple mortals" is the access to computer resources, it is indeed especially hard after 3050+ to get some elo quickly as every single change requires so much testing.