ELF OpenGo: An Open Reimplementation of AlphaZero

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
clumma
Posts: 177
Joined: Fri Oct 10, 2014 8:05 pm
Location: Berkeley, CA

ELF OpenGo: An Open Reimplementation of AlphaZero

Post by clumma » Wed Feb 13, 2019 5:04 pm

What is Leela doing wrong?

https://arxiv.org/abs/1902.04522

-Carl

Steppenwolf
Posts: 33
Joined: Thu Jan 31, 2019 3:54 pm
Full name: Sven Steppenwolf

Re: ELF OpenGo: An Open Reimplementation of AlphaZero

Post by Steppenwolf » Wed Feb 13, 2019 6:19 pm

Great, there is alreday a binary available: https://facebook.ai/developers/tools/elf-opengo
Waiting for Porting this ELF to chess...

User avatar
Guenther
Posts: 2935
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Full name: Guenther Simon
Contact:

Re: ELF OpenGo: An Open Reimplementation of AlphaZero

Post by Guenther » Wed Feb 13, 2019 6:29 pm

Steppenwolf wrote:
Wed Feb 13, 2019 6:19 pm
Great, there is alreday a binary available: https://facebook.ai/developers/tools/elf-opengo
Waiting for Porting this ELF to chess...
This is even more interesting for the programmers section
https://github.com/pytorch/ELF
Current foe list count : [92 - still rising]
http://rwbc-chess.de/chronology.htm

Steppenwolf
Posts: 33
Joined: Thu Jan 31, 2019 3:54 pm
Full name: Sven Steppenwolf

Re: ELF OpenGo: An Open Reimplementation of AlphaZero

Post by Steppenwolf » Wed Feb 13, 2019 6:45 pm

Now are Volunteers needed for ELF OpenChess!

I just found: https://github.com/pytorch/ELF/issues/8

Daniel Shawul
Posts: 3707
Joined: Tue Mar 14, 2006 10:34 am
Location: Ethiopia
Contact:

Re: ELF OpenGo: An Open Reimplementation of AlphaZero

Post by Daniel Shawul » Thu Feb 14, 2019 4:03 am

Notes I took from glancing at the paper

a) CPUCT = 1.5

b) Virtual loss = 1

c) Ladders ( tactics in GO) is hard to learn

d) Batch normalization moment staleness. Some technical issue i don't fully understand but for which they have provided a plugin in torch

e) Value head only, which is something I used to do, gives a weak engine. They accidentally found this out when they fixed policy weight to 1/362
by mistake. The full quote:
Dominating value gradients We performed an unintentional
ablation study in which we set the cross entropy coefficient
to 1/362 during backpropogation. This change will
train the value network much faster than the policy network.
We observe that ELF OpenGo can still achieve a strength
of around amateur dan level. Further progress is extremely
slow, likely due to the minimal gradient from policy network.
This suggests that any MCTS augmented with only a
value heuristic has a relatively low skill ceiling in Go.
f) Game resignation during selfplay training is important. It will focus the net to learn the opening/middlegame (most important parts of the game) faster

vijadhav321
Posts: 1
Joined: Wed Feb 13, 2019 10:05 am
Full name: Vishal Jadhav

Re: ELF OpenGo: An Open Reimplementation of AlphaZero

Post by vijadhav321 » Thu Feb 14, 2019 9:41 am

Great information

tocnaza
Posts: 2
Joined: Sun Mar 03, 2019 6:13 pm
Full name: juanca lina

Re: ELF OpenGo: An Open Reimplementation of AlphaZero

Post by tocnaza » Mon Mar 04, 2019 7:57 am

Nice post!

Post Reply