go4thewin wrote:
Ah I see, thanks for running so many games. Just two last questions. 1) For example, I played this game that the optimized net policy beat zen 6 7 dan as white with no komi. But since winrate is not measured when using just 1 playout on my device, I hit ponder alot to see who was winning. Does this give katago more visits, or does it still just use the policy output?
2) It looks like a 60b net is being trained. If hypothetically someone were to make an optimized version of that net using the exact same methods you used, what would the size and speed be? Is it oversimplication to say that since 60*256 is 1.5 times 40*256, the size will be about 48*1.5 mb, so it will be about the size and speed of the non optimized 20b net?
Thanks!
1) I'm sorry, I don't know, ask @lightvector about this.
2) Other parameters being equal, the network size will be roughly proportional to the number of blocks while speed will be inversely proportional. So your size estimate seems reasonable, the speed depends on the speedup that the optimization provides which depends on device type:
Best speedup I heard of was factor 6, then an optimized 60b network will still be twice as fast as the unoptimized 20b.
Worst speedup was factor 2, then an optimized 60b network will be 33% slower than the unoptimized 20b.