The progress from #174 to #254

[hwj-111]

I extract the data from http://www.yss-aya.com/cgos/19x19/bayes.html
based on the results of all "LZ_???_????_p400" robots.
The progress rate in Bayes ELO is about 4/model (and looks like slowing down recently)
ELO +330 in 14 months!

[diadorak]

https://cloudygo.com/leela-zero-eval/eval-graphs

[cryptsport]

if extrapolate these values, then after how many games will 300 more elo be added?
(given the slowdown)

[Marcin1960]

@cryptpark

Probably couple years or more? Be patient ;)

[cryptsport]

@Marcin1960 with my pace of studying igo, I have enough KataGo for 200 years :)
it's a sports fan interest

[hwj-111]

I added another data set, which use all "lz???_3ghwj" (my computer with 2xGTX1070 + 1xGTX1060, roughly p2400 for 40b models). It is about +350 ELO higher than p400

[l1t1]

nearly 90 matches, no weight promotes

[yssaya]

This is Selfplay games winrate from LZ_200 to LZ_254.
These are from raw training data.
Black winrate is declining. But it looks flat from LZ_235 to LZ_254.
Current black winrate is around 0.425.

And this is ELFv2_p800 winrate against LZ_xxx_yyyy_p400.
http://www.yss-aya.com/cgos/19x19/cross/LZ_05db_ELFv2_p800.html
It looks like progress is small from LZ_235 to LZ_254.