-
Notifications
You must be signed in to change notification settings - Fork 543
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Will anyone use KataGo to attend the World AI GO Competition on 27, Sep in Fuzhou, China? #320
Comments
There have been some people in China who are collaborating with lightvector, but the language barrier has been a problem, and he requested assistance with this a while ago via discord. I do not know if anyone stepped up to help. But 10 minutes is probably a little short notice ;D |
Yep, as you've probably noticed in announcements from elsewhere, yes someone has arranged to run it in the tournament. KataGo's participation is mostly for fun. I'm not 100% confident in exactly what configuration and hardware the person or group who will operate it intend to use, and although I trust that they're interested in KataGo doing well and configuring it appropriately, the communication has not been the easiest due to the language difference. For the record, stating my guess prior to the tournament taking place, sort of in response to some of the messages I've received privately and through other channels from people hyped for KataGo to be in this tournament, some of which have seemed bizarrely-to-me highly-secretive and cutthroat about beating others and winning: I think that KataGo is unlikely to place near the top (but perhaps it will, it would be a pleasant surprise). Unlike many competitors, KataGo is open source - its code, its data, and its techniques. KataGo (I hope) has pushed the state of the art in its field of research - and doing that well and sharing freely with the community has always been the goal. I doubt it would be hard for a dedicated group of developers, particularly ones working full-time, to replicate its methods, and if any of them happened to know any further improvements and new techniques not yet published so far, it would not be hard for them to improve yet further. Undoubtedly, there are many further algorithmic and training improvements to be found, and I'd find it shocking if among many smart researchers it weren't the case that more things have been found. And of course there are undoubtedly well-funded research groups and labs out there with much larger amounts of computation for training and testing available to them, that KataGo's resources would be no match for. Tournaments like this are primarily for entertainment. They don't involve the hardware controls or the hundreds to thousands of games you'd run when actually trying to scientifically compare different programs and training methods and architectures, instead the focus is on commentary and engaging and bringing people together in a shared activity. So let's just enjoy the entertainment. Enjoy spectating the games, best of luck to everyone of course, and have fun! |
Thanks for your reply, my meaning is a consultation, not a notice. I'm happy to see the information about that KataGo may attend this competition. And in the condition of the same time and hardware, KataGo is stronger than Golaxy and FineArt based on the tests made by some amateurs and professors in China. |
@lightvector |
Did you ever contact to the AGA(American Go Association), it is one co-organizer of this tournament? They may give help to the problem of language difference. |
There seems to be some incidents on the first day, don't know what was the cause. |
I just feel that the result of this kind of tournaments without any specs on hardware/energy consumption is less meaningful. The bots win most likely due to they have more computing power. It will be cool if there are some AI Go competitions that has an up-limit on power consumption (say, <1000W) |
player win lost |
Some of these games include unfair wins and losses due to timeouts or failure to connect due to unstable network or internet at the competition site, for both Leela Zero and KataGo. Many other participants have also experienced timeouts or other issues - I think somewhere around a quarter to a third of the games in the tournament as a whole in the first two rounds were decided by these issues, rather than by the actual skill of play of the bots, and some games in later rounds too. Possibly the lags have also influenced games where it wasn't quite bad enough to cause a timeout. And there have been other discussions that make you sort of question what's going on about other details too. Even after going in expecting (as I mentioned in my initial post above) that the results in tournaments like this aren't worth trusting too much because of such a small sample size and lack of controls, definitely feeling a disappointed in this tournament due to the mess of issues above, regardless of what the outcome from here is. |
@mega-optimus @lightvector
Players need to fit the new rule by switcing AI during playing, and as a result, timeout is everywhere. |
Interview of lightvector by Yike published: https://mp.weixin.qq.com/s/XWMFvQYB8a5hBcpIHmz5DQ |
@alreadydone |
semi final matches began |
someone says two of the top four are using katago https://m.newsmth.net/article/Weiqi/648486 |
In their own statements, four in eight are using KataGo, they are ChaoRan, YiLe, izis and Tian. |
The English introduce of this competition on 2019 is here: leela-zero/leela-zero#2278
The Chinese introduce of this competition on 2020 is here: https://sports.sina.com.cn/go/2020-09-08/doc-iivhvpwy5522546.shtml
The email address of this competition is here: fuzhouworldaigo@sina.com
The deadline to enroll this competition is Sep 25 00:00 (UTC+8), there is only 10 minutes left.
I know KataGo is powerful, and I hope it will attend this competition, but I don't know if someone have enrolled.
The text was updated successfully, but these errors were encountered: