New networks 20 blocks

[And]

new network of 20 blocks b7e1fc49, stronger than lz # 173.
http://zero.sjeng.org
https://github.com/leela-zero/leela-zero/issues/2504
it would be interesting to compare with elf v2, time parity

[And]

new network of 20 blocks 5282a90e http://zero.sjeng.org/networks/5282a90e ... 853bf62.gz

5282a90e VS b7e1fc49 219 : 188 (53.81%) 407 games
http://zero.sjeng.org

221:179 over b7e1fc49
48:54 against elfv2
https://github.com/leela-zero/leela-zero/issues/2504

[And]

new network of 20 blocks 6b5a964e https://zero.sjeng.org/networks/6b5a964 ... f7fad3c.gz

6b5a964e VS b7e1fc49 237 : 163 (59.25%) 400 games
https://zero.sjeng.org

226:174 over b7e1fc49
126:274 vs elfv2
https://github.com/leela-zero/leela-zero/issues/2504

[And]

can anyone suggest why such a result - 126:274 is against the elf?

[wineandgolover]

Could it be a “new move” that it thinks is good, and can’t refute in self-play, but that Elf2 can debunk?

I agree the reduction in win rate is dramatic!

[And]

new network of 20 blocks a8f4d271 https://zero.sjeng.org/networks/a8f4d27 ... 3b7fd07.gz
a8f4d271 VS 6b5a964e 200 : 184 (52.08%) 384 games
https://zero.sjeng.org

221:179 over 6b5a964e (55.25%)
https://github.com/leela-zero/leela-zero/issues/2504

[And]

new network 1fd6ac21 https://zero.sjeng.org/networks/1fd6ac2 ... 5b197d4.gz
1fd6ac21 VS 6b5a964e 240 : 160 (60.00%) 400 games
https://zero.sjeng.org

185 : 237 (43.84%) against elfv2

file size:
1fd6ac21 - 168 mb, gz - 47 mb
elfv2 - 323 mb, gz - 117 mb

[And]

new network of 20 blocks ae853694 https://zero.sjeng.org/networks/ae85369 ... fd5eab4.gz
ae853694 VS 1fd6ac21 163 : 160 (50.46%) 323 games
https://zero.sjeng.org

231:169 (57,75%)
https://github.com/leela-zero/leela-zero/issues/2504

[And]

new network of 20 blocks eee8895c https://zero.sjeng.org/networks/eee8895 ... 8fb7b64.gz
eee8895c VS 1fd6ac21 142 : 152 (48.30%) 294 games
https://zero.sjeng.org

233:167 (58,25%)
https://github.com/leela-zero/leela-zero/issues/2504

[And]

new network of 20 blocks 1e5061af http://zero.sjeng.org/networks/1e5061af ... fb3adb7.gz

249:151 over 1fd6ac21 (62,25%)
https://github.com/leela-zero/leela-zero/issues/2504

[And]

new network of 20 blocks d4ae73bb https://zero.sjeng.org/networks/d4ae73b ... 2957150.gz
d4ae73bb VS 1fd6ac21 229 : 187 (55.05%) 416 games
https://zero.sjeng.org

58:41 (58,58%) against elfv2 https://zero.sjeng.org/match-games/5e10 ... 23ebc2e8ec

[TelegraphGo]

In Game 41 of the match against ELF v2, the result is given as B+38.5... but black is totally dead in the bottom right? I think whatever scorer you're using made a mistake.

[And]

new network of 20 blocks 5c29e7d5 https://zero.sjeng.org/networks/5c29e7d ... f7cf380.gz
5c29e7d5 VS 1fd6ac21 258 : 142 (64.50%) 400 games

145 : 155 (48.33%) 300 games against ELFv2
https://zero.sjeng.org

[Amigo]

new networks expected?