Ratinglist- and regression-testruns of Stockfish 220422 finished.
https://www.sp-cc.de/stockfish-regression.htm
https://www.sp-cc.de
Important: From now (Stockfish 15), the regression-testrun is not 20000 games with bullet-timecontrol (30sec+300ms) anymore, but 2000 games with VLTC of 10min+3sec (average game duration 30 minutes, around 3x more time than in the SPCC-ratinglist, using my UHO 2022 openings (otherwise 95%+ draws...)). I believe, this makes sense, because of 2 facts:
1) In February, we saw a huge regression of Stockfish-dev with short time-controls, but in Fishtest, VLTC-tests had shown a progress. And most chess-players and computerchess-fans use Stockfish with longer timecontrols, so a regression-test with long timecontrol makes sense to me.
2) In Fishtest, regression-test with short time-control (and sometimes using UHO-openings) are done, too, so there is no need, to do more of these tests here IMO.
Of course, 2000 games is not so much and the errorbar is quite huge, but 1000 gamepairs will give a reasonable result. And each regression-testrun will get a gamepair-rescoring by my Gamepairs Rescorer Tool, besides the classical Elo-performance-list.
(Perhaps you have to clear your browsercache or reload the website)
SPCC: Testruns of Stockfish 220422 finished
Moderators: hgm, Dann Corbit, Harvey Williamson
-
pohl4711
- Posts: 2388
- Joined: Sat Sep 03, 2011 7:25 am
- Location: Berlin, Germany
- Full name: Stefan Pohl