New unbalanced openings are used in the SF-Framework

[pohl4711]

From now, the Stockfish Framework (https://tests.stockfishchess.org/tests) uses my new Unbalanced Human Openings XXL openings-set for testing all new patches! This will lower the draw-rate there from around 90% to around 50% (!)

https://github.com/glinscott/fishtest/pull/996
(Switches to a new book by Stefan Pohl: UHO_XXL_+0.90_+1.19.epd)

"With the new unbalanced book we would favor a bit patches that perform better in those 'sharp' situations. We can't predict (and hard to test) what the result of such a different 'evolutionary pressure' will be. However, I think it would be nice to try out the change. I'm sure we'll continue to see progress, it will be just a different direction. Anyway. It could be a good idea to introduce an unbalanced book and then to redo the test after a few months. It if turns out that the noob book is still significantly better (in NormalizedElo), then we can switch back, since then the result can no longer be attributed to selection bias."

So, it is possible, that the SF-Framework will return to the classical, balanced noob_3mvs set in some months, but anyway, using an unbalanced openings-set for the first time in the SF-Framework is a very interesting experiment and a new direction in the development (and of course an honor for me, that my UHO_XXL set was chosen for this!): An unbalanced openings-set "would favor a bit patches that perform better in those 'sharp' situations" - IMHO, that is definitly not a bad thing, isnt it?!


Learn more about the UHO openings on my website:
https://www.sp-cc.de/uho_xxl_project.htm
https://www.sp-cc.de/anti-draw-openings.htm

[MikeB]

From now, the Stockfish Framework (https://tests.stockfishchess.org/tests) uses my new Unbalanced Human Openings XXL openings-set for testing all new patches! This will lower the draw-rate there from around 90% to around 50% (!)

https://github.com/glinscott/fishtest/pull/996
(Switches to a new book by Stefan Pohl: UHO_XXL_+0.90_+1.19.epd)

"With the new unbalanced book we would favor a bit patches that perform better in those 'sharp' situations. We can't predict (and hard to test) what the result of such a different 'evolutionary pressure' will be. However, I think it would be nice to try out the change. I'm sure we'll continue to see progress, it will be just a different direction. Anyway. It could be a good idea to introduce an unbalanced book and then to redo the test after a few months. It if turns out that the noob book is still significantly better (in NormalizedElo), then we can switch back, since then the result can no longer be attributed to selection bias."

So, it is possible, that the SF-Framework will return to the classical, balanced noob_3mvs set in some months, but anyway, using an unbalanced openings-set for the first time in the SF-Framework is a very interesting experiment and a new direction in the development (and of course an honor for me, that my UHO_XXL set was chosen for this!): An unbalanced openings-set "would favor a bit patches that perform better in those 'sharp' situations" - IMHO, that is definitly not a bad thing, isnt it?!


Learn more about the UHO openings on my website:
https://www.sp-cc.de/uho_xxl_project.htm
https://www.sp-cc.de/anti-draw-openings.htm

+1 and you should be honored, nice work!

[Terje]

Very cool, I look forward to seeing how it plays out!