Re: [Computer-go] MiniGo open sourced
On 30-01-18 20:59, Álvaro Begué wrote: > Chrilly Donninger's quote was probably mostly true in the 90s, but > it's now obsolete. That intellectual protectionism was motivated by > the potential economic profit of having a strong engine. It probably > slowed down computer chess for decades, until the advent of strong > open-source programs. Paradoxically, when the economic incentive to > create strong engines was removed, we saw an explosion in strength. There still seems to be an economic incentive to improve [1] strong engines and try to sell them. It should be noted that until Stockfish came along, open source computer chess engines were a graveyard where every strong enough engine just got cloned or plagiarized and real enduring cooperation was essentially nonexistent. You just had 10 non-cooperating forks (some closed source, and some allegedly commercial ones) that added <-20 ... >+100 Elo. There had been open source engines as early as GNUChess (or probably earlier...), and very strong ones like Fruit. I don't know for sure what allowed Stockfish to (mostly) escape the same fate. Right now I would say fishtest is a huge factor, but it might've been doing fine before that. [1] I originally wrote "create" here but that might not be correct. -- GCP ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] MiniGo open sourced
Chrilly Donninger's quote was probably mostly true in the 90s, but it's now obsolete. That intellectual protectionism was motivated by the potential economic profit of having a strong engine. It probably slowed down computer chess for decades, until the advent of strong open-source programs. Paradoxically, when the economic incentive to create strong engines was removed, we saw an explosion in strength. Álvaro. On Tue, Jan 30, 2018 at 2:14 PM, "Ingo Althöfer" <3-hirn-ver...@gmx.de> wrote: > Hi, > > GCP wrote: > > ... > > > Of course, in the end, strength is the best way to tell that your > > > implementation is correct :) > > > > In other words, do not take "correct" as necessarily meaning "matching > > the published research". > > Chrilly Donnninger, one of the computer chess gurus in the 1990's and > the early 200x's (project Hydra) had an expressed opinion: > "Those who know, do not publish. > And those who publish do not know." > He himself violated this rule in the early 1990's when he published > a price-winning paper on how to implement null-move search correctly. > > Ingo. > ___ > Computer-go mailing list > Computer-go@computer-go.org > http://computer-go.org/mailman/listinfo/computer-go > ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] MiniGo open sourced
Hi, GCP wrote: > ... > > Of course, in the end, strength is the best way to tell that your > > implementation is correct :) > > In other words, do not take "correct" as necessarily meaning "matching > the published research". Chrilly Donnninger, one of the computer chess gurus in the 1990's and the early 200x's (project Hydra) had an expressed opinion: "Those who know, do not publish. And those who publish do not know." He himself violated this rule in the early 1990's when he published a price-winning paper on how to implement null-move search correctly. Ingo. ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] MiniGo open sourced
On 30-01-18 02:50, Brian Lee wrote: > We're not aiming for a top-level Go AI; we're merely aiming for a > correct, very readable implementation of the AlphaGoZero algorithm I had a look around to see how you resolved what I'd consider the ambiguities in the original paper: https://github.com/gcp/leela-zero/issues/785 > Of course, in the end, strength is the best way to tell that your > implementation is correct :) In other words, do not take "correct" as necessarily meaning "matching the published research". -- GCP ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] MiniGo open sourced
Dear Brian, thank you for your posting and for publishing the MiniGo code. > I'm happy to announce MiniGo is now open source. > https://github.com/tensorflow/minigo > > We're ... aiming for a correct, very readable implementation > of the AlphaGoZero algorithm and demonstration of Google > Cloud / Kubernetes / TensorFlow. ... Hopefully, others will use your code also for attacking other games with simple rules, like Clobber, ConHex (by Michail Antonow), or Yavalath (by Cameron Browne/Ludohex). > A few Googlers, including myself and Andrew Jackson, have > been working on this, but we're otherwise completely > unaffiliated with DeepMind and the AlphaGo project. May you tell us, in which Google lab you are working? Thanks again for your contribution! Ingo. ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
