LC0 finally?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

Werewolf
Posts: 1795
Joined: Thu Sep 18, 2008 10:24 pm

Re: LC0 finally?

Post by Werewolf »

Hasn’t the current network saturated? Why haven’t we moved to a larger net?
yanquis1972
Posts: 1766
Joined: Wed Jun 03, 2009 12:14 am

Re: LC0 finally?

Post by yanquis1972 »

am wondering the same thing, if it's still a test-run they may want to be absolutely sure of the limit.
Werewolf
Posts: 1795
Joined: Thu Sep 18, 2008 10:24 pm

Re: LC0 finally?

Post by Werewolf »

yanquis1972 wrote: Fri Jun 22, 2018 8:39 pm am wondering the same thing, if it's still a test-run they may want to be absolutely sure of the limit.
I did actually wonder if the reason was another restart was due, but thought this time it would be different.

Ho hum. Maybe it’s called LC0 because it’s Elo keeps going back to zero :D
User avatar
Rebel
Posts: 6991
Joined: Thu Aug 18, 2011 12:04 pm

Re: LC0 finally?

Post by Rebel »

Werewolf wrote: Fri Jun 22, 2018 10:58 pm
yanquis1972 wrote: Fri Jun 22, 2018 8:39 pm am wondering the same thing, if it's still a test-run they may want to be absolutely sure of the limit.
I did actually wonder if the reason was another restart was due, but thought this time it would be different.

Ho hum. Maybe it’s called LC0 because it’s Elo keeps going back to zero :D
Elswhere
Id396 is a new net without input from broken games:

Net has been replaced with a new bootstrap net with 50 move rule normalization and learnable batch normalization beta bias. It will likely be weaker to start but will hopefully grow stronger quickly. Thanks to scs-ben!

scs-ben - 06/02/2018
192x15 network is being bootstrapped. Batch size is 512. Using games 7 million to current but excluding 10.5-14 million due to likely bad data. Batch normalization, 50 move rule plane normalization, test/train data separation are all included.
90% of coding is debugging, the other 10% is writing bugs.
Werewolf
Posts: 1795
Joined: Thu Sep 18, 2008 10:24 pm

Re: LC0 finally?

Post by Werewolf »

It’s good they keep trying things, but this net doesn’t look that good, unless the numbers are skewed. Perhaps another restart?
User avatar
Rebel
Posts: 6991
Joined: Thu Aug 18, 2011 12:04 pm

Re: LC0 finally?

Post by Rebel »

Well, there was a bug, hence the restart.

As said -
It will likely be weaker to start but will hopefully grow stronger quickly. Thanks to scs-ben!
This seems to be exactly the case.

Patience, patience and hoping for no new bugs
90% of coding is debugging, the other 10% is writing bugs.
Uri Blass
Posts: 10267
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: LC0 finally?

Post by Uri Blass »

Rebel wrote: Sun Jun 24, 2018 11:04 am Well, there was a bug, hence the restart.

As said -
It will likely be weaker to start but will hopefully grow stronger quickly. Thanks to scs-ben!
This seems to be exactly the case.

Patience, patience and hoping for no new bugs
what about having correct rating graph by calculating the rating in the graph not based on self play but based on games against known opponents?
yanquis1972
Posts: 1766
Joined: Wed Jun 03, 2009 12:14 am

Re: LC0 finally?

Post by yanquis1972 »

that's going on in the background & not-so-in-the-background...

http://lczero.org/?full_elo=1 click elo estimates

https://docs.google.com/spreadsheets/d/ ... li=1#gid=0
User avatar
Leto
Posts: 2071
Joined: Thu May 04, 2006 3:40 am
Location: Dune

Re: LC0 finally?

Post by Leto »

yanquis1972 wrote: Sun Jun 24, 2018 1:42 pm that's going on in the background & not-so-in-the-background...

http://lczero.org/?full_elo=1 click elo estimates

https://docs.google.com/spreadsheets/d/ ... li=1#gid=0
I wish I understood that second spreadsheet, looks so confusing to me. This one I can understand:
https://docs.google.com/spreadsheets/d/ ... edit#gid=0
yanquis1972
Posts: 1766
Joined: Wed Jun 03, 2009 12:14 am

Re: LC0 finally?

Post by yanquis1972 »

true, i linked the second bc i know it's official, but that one's simpler.