[Computer-go] My 9x9 network

I trained this network from scratch over a few months using the KataGo engine. It is already super-human strength on 9x9, but probably wouldn't understand 19x19 opening well https://github.com/lightvector/KataGo/issues/316 ___ Computer-go mailing list

Re: [Computer-go] Training an AlphaGo Zero-like algorithm with limited hardware on 7x7 boards

hought. > > Have either of you ever benchmarked your 7x7 (or 9x9) models against GNU > Go? > > By the way, all benchmarking against GNU Go that I've reported was in > single-pass mode only (i.e., I was not running the tree search on top of > the net outputs) > > Thanks, > Cody >

Re: [Computer-go] Training an AlphaGo Zero-like algorithm with limited hardware on 7x7 boards

I trained using David Wu's code for a few months on 9x9 only and it's been superhuman after a few months. I'm not sure if anyone's interested, but I can release my network to the world. It's around the strength of KataGo, but only on 9x9. I could do a final test before releasing it into the wild

Re: [Computer-go] GCP passing on the staff ...

I saw a game where it decided connecting its group wasn't important since it could still make the cut off side sabaki later Approaches corner, decides it's ahead, lives later It doesn't care about keeping groups strong as much On Wed, Jan 30, 2019, 13:24 uurtamo They will abandon a fight to

Re: [Computer-go] Video of Aja Huang's presentation

Here it is: https://youtu.be/KoIv7oYZ8wc On 2016-07-06 12:03, Jim O'Flaherty wrote: Any chance someone has put this on Youtube for those of us who primarily consume videos on phones or tablets (where a 2.0GB is very large to store locally)? And if so, replying with a link here would be

Re: [Computer-go] Scoring Network

I don't know about different stages of the game. Intuitively, a certain area is in different stages of being settled. Certain joseki only leave 2 points up for grabs in the endgame, while other areas might stay unsettled (ko for life) for 150 moves. On 2016-06-22 22:50, Stefan Kaitschick

Re: [Computer-go] DarkForest is open-source now.

Can I run the CNN on my ATI card somehow? On 2016-06-10 10:31, Yuandong Tian wrote: Hi all, DarkForest Go engine is now public on the Github (pre-trained CNN models are also public). Hopefully it will help the community. https://github.com/facebookresearch/darkforestGo Thanks! Best,

Re: [Computer-go] AlphaGo to play against Ke Jie

https://twitter.com/demishassabis/status/739832323160563712 Sounds like news to Demis Hassabis :) On 2016-06-04 20:00, "Ingo Althöfer" wrote: Hi, during the amateur World Championships in Wuxi the new president of the international Go Association, Yang Jun'an, announced that Chinese young

[Computer-go] Try a neural network in your browser

http://playground.tensorflow.org/ this actually uses your cpu to train the network it looks like you can either make a deep learning network with simple inputs or a shallow learning network with more complex inputs to solve the spiral case (the only challenging one) the shallower networks

Re: [Computer-go] Postscript... Preface

not hard to find the main line (the actual game); it's the longest branch, which is easy to see if you have a tree viewer open on the side. i appreciate it would be hard to navigate in the dark without being able to see the tree as well On 16/04/2016, Igor Polyakov <weiqiprogramm...@gmail.com>

Re: [Computer-go] Postscript... Preface

it bothers me that the main line is not the main line On 2016-04-15 23:35, djhbrown . wrote: = Go talking software (featuring coalesced expert commentaries on Alphago's mistake in game 4 and a suggestion for an enduser-useful variation network editor student

Re: [Computer-go] Nice graph

AGA ratings (not ranks, which only go up to 7d, but ratings) go up infinitely, so some players have had above 10.0 rating. Mark Lee could have given a 7.0 three stones and still won possibly, despite them both being 7d amateur. Mark Lee also is favored against weaker professionals, too. So 10d

Re: [Computer-go] Densei-sen

Wow, Black still recovered after a joseki mistake. You'd expect joseki to come baked-in, but it didn't need them! I wonder what would happen in a two stone game. On 2016-03-23 1:42, Hiroshi Yamashita wrote: Zen won against KOBAYASHI with 3 handicaps. (;GM[1]SZ[19] PB[Zen] PW[KOBAYASHI

Re: [Computer-go] Congratulations to AlphaGo

At this point I don't doubt that the single machine version is professional strength which is enough to be used as a tool to analyze games... On 2016-03-12 16:24, Lukas van de Wiel wrote: This would be many thousands of dollars per day. A single game would be more than a thousand dollars in

Re: [Computer-go] AlphaGo won first game!

The amount of points AlphaGo was ahead is also meaningless because it started playing slow moves when it got ahead. After the big fights were finished it already knew it was going to win easily. On 2016-03-08 23:55, Marc Landgraf wrote: It was pointed out by Lee Sedol after the game and Kim

Re: [Computer-go] Mastering the Game of Go with Deep Neural Networks and Tree Search

I think it would be an awesome commercial product for strong Go players. Maybe even if the AI shows the continuations and the score estimates between different lines, it will give the player enough reasoning to understand why one move is better than the other. On 2016-02-02 8:29, Jim

Re: [Computer-go] guess AlphaGo's CGOS rating

But AlphaGo single machine is stronger than Fan Hui, since it won 8-2 in all matches combined on one machine, as far as I understand they used this version On 2016-02-02 13:33, Hiroshi Yamashita wrote: Hi, I made CGOS rating include AlphaGo. CGOSRemi's pro rating(same as

Re: [Computer-go] The Game AI Forum is back

Is there no Go section? On 2016-01-01 0:56, Rémi Coulom wrote: Hi, I had created the Game Programming Forum a few years ago. I decided to put it online again, at a new URL: http://www.game-ai-forum.org/ Maybe some of you will be interested to participate there. Rémi

Re: [Computer-go] CNN with 54% prediction on KGS 6d+ data

I doubt that the illegal moves would fall away since every professional would retake the ko... if it was legal On 2015-12-09 4:59, Michael Markefka wrote: Thank you for the feedback, everyone. Regarding the CPU-GPU roundtrips, I'm wondering whether it'd be possible to recursively apply the

Re: [Computer-go] Facebook Go AI

If you train your neural network on pro games, pros never play out ladders that end up in capture, so when a ladder situation happens and it gets played out, the running group is always safe. This is not the case always, but you'd need to specifically play out the ladder to check. On

Re: [Computer-go] How to handle triple ko efficiently?

If you're only getting 1000 table generations a second, you should look into your algorithm. You should get at least 100,000 table generations a second! On 2015-10-16 7:21, Álvaro Begué wrote: That sounds kind of obsessive. I think the probability of having a 0 or repeated numbers in the

Re: [Computer-go] Neural Network Chess Computer Abandons Brute Force For "Human" Approach

afaik monte carlo approaches in chess achieved less than 1000 FIDE level On 2015-09-14 20:09, Gonçalo Mendes Ferreira wrote: I haven't opened the links but my immediate 2 cents are: 1. chess master level is not impressive for a chess engine 2. trained ANN have diminishing returns of their error

Re: [Computer-go] KGS access problem

Try disabling https everywhere extension and accessing the unsecure site at http://www.gokgs.com/ but yes, it would be better if they fixed this On 2015-09-10 5:31, "Ingo Althöfer" wrote: Hello, I have problems to access the KGS server. My Firefox 40.0.3 (under Windows 8.1) is even not

Re: [Computer-go] Life and Death

Just stay up for like 30 hours and it will be pretty spooky at that point. 10/10 would watch again On 2015-09-03 0:44, Jim O'Flaherty wrote: I think you forgot to suggest which pharmaceuticals, legal or otherwise, to be using while watching this. Without said pharmacological assistance, that

Re: [Computer-go] Building A Computer Go AI

Check out http://senseis.xmp.net/?UCT this is a basic Monte Carlo bot, probably 25k in strength the way it works is better explained here: https://en.wikipedia.org/wiki/Monte_Carlo_tree_search but I didn't understand the article until I looked at the Sensei's library implementation first On

Re: [Computer-go] Fwd: mental imagery in Go

I'm very curious about your bot because I haven't seen how exactly it works. What algorithm does it use to generate moves? On 2015-08-01 7:08, folkert wrote: Hi, You can take a look at https://youtu.be/c3wbE1V4bTc It shows the thought proces of my Go program called Stop:

Re: [Computer-go] CGT endgame solver

I'm working on Iomrascalai with Urban Hafner. It's fairly strong at 9x9 (better than GnuGo), but needs work on larger boards. On 2015-07-14 0:22, Ingo Althöfer wrote: Hi, Igor Polyakov weiqiprogramm...@gmail.com: Even strong engines like Fuego will give wrong sequences in yose. And even

Re: [Computer-go] CGT endgame solver

Even strong engines like Fuego will give wrong sequences in yose. 1. Sometimes it will play a 1 point sente too early (right as it's available, so maybe it already played it early game), get into a ko and lose 10 points in seki or die This is because playouts always give a better result for

[Computer-go] Is CGOS 13x13 frozen for anyone else?

I haven't been able to run my bot. ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go

Re: [Computer-go] Seki in playouts

I just tried this, and included code in the playout to save groups in atari by counter-capturing enemy stones. It still chooses to connect in that situation, so I am going back to passing in playouts. On 2015-04-16 3:41, Aja Huang wrote: On Wed, Apr 15, 2015 at 12:24 PM, Igor Polyakov

[Computer-go] Seki in playouts

I have an interesting problem with my program. 1. I have self-atari detection, but self-atari is allowed for smaller groups like the two stone group because it can make a nakade. 2. Self-atari detection means that in most playouts the program will let the black player suicide, but it won't let

[Computer-go] UCB-1 tuned policy

I implemented UCB1-tuned in my basic UCB-1 go player, but it doesn't seem like it makes a difference in self-play. It seems like it's able to run 5-25% more simulations, which means it's probably exploiting deeper (and has less steps until it runs out of room to play legal moves), but I have

Re: [Computer-go] What's a good playout speed?

It doesn't matter with simple UCT, but it's a huge difference when you have more targeted playouts because you can find a subtree you really like and spend 99%+ of your time exploiting that particular line (for example, not dying in one move). I see Fuego re-use most of its existing tree on a

Re: [Computer-go] CGOS future

C++ is not simple at all. It's probably the biggest language out there with the most caveats. On 2015-04-03 0:13, folkert wrote: Please write it in something common. If you target linux I think it is unusual to target .net. Why not just c++? Proven technology, simple to write in. On Thu, Apr