LCZero: Progress and Scaling. Relation to CCRL Elo

[Werewolf]

Saw this today.


https://docs.google.com/spreadsheets/d/ ... edit#gid=0

so elo 2552 on april 14th

now april 26 it is elo 2803. so a 251 elo gain in 12 days and 21 points a day gain. (including regression). To catch stockfish at 3563, another 760 points needed which should take about 5 weeks unless it slows down. so beginning of June. just maybe.

3rd june 'only' 3121 . so a 318 gain in 5 weeks so 9 points a day gain. To catch stockfish at 3563, another 442 points needed which should take about 7 weeks unless it slows down. so beginning of august. just maybe.

Everyone else here sees progress, I just see stalling. There's no way it'll catch SF anytime soon IMO.

[duncan]

so elo 2552 on april 14th

now april 26 it is elo 2803. so a 251 elo gain in 12 days and 21 points a day gain. (including regression). To catch stockfish at 3563, another 760 points needed which should take about 5 weeks unless it slows down. so beginning of June. just maybe.

3rd june 'only' 3121 . so a 318 gain in 5 weeks so 9 points a day gain. To catch stockfish at 3563, another 442 points needed which should take about 7 weeks unless it slows down. so beginning of august. just maybe.

Everyone else here sees progress, I just see stalling. There's no way it'll catch SF anytime soon IMO.

9 points a day is stalling. when has an alpha-beta program ever done that for even a week?

[Werewolf]

3rd june 'only' 3121 . so a 318 gain in 5 weeks so 9 points a day gain. To catch stockfish at 3563, another 442 points needed which should take about 7 weeks unless it slows down. so beginning of august. just maybe.

Everyone else here sees progress, I just see stalling. There's no way it'll catch SF anytime soon IMO.

9 points a day is stalling. when has an alpha-beta program ever done that for even a week?

9 elo a day is incredible, granted. But I suppose the issue is whether it continues or if there is a trend towards no improvement.

[yanquis1972]

why is that an issue if the trend projects to a 3600 elo engine within, say, a year?

[Milton]

For the purposes of not starting another LCzero thread, I am posing my question here.

Can someone explain to me why LCzero refuses to use its time resource? I am using the GPU version of Leela. I have a Nvidea 1080 video card. I am using the Fritz 16 GUI. No matter how much time I allow LCzero in the GUI, Leela moves instantly, essentially playing bullet chess. Is there a setting I can change in the engine parameters which will cause it think longer? Thanks.