Re: [Computer-go] alternative for cgos
>
> And I'm now excited about possibly reusing the baduk.io's
> Docker infrastructure for the EGC2015 Computer Go tournament I'm
> planning, where bots will run on equal hardware too - it'd make the
> logistics quite easier for many competitors.
That's great to hear! If you need any help, let me know!
> Note that while this is fine for GNUGo, it may not be for some of the
> more advanced bots:
> * Initialization time may be non-trivial. If you use Pachi with the
> patterns database (which is extremely recommended), it takes a few
> seconds to load.
> * You lose not just pondering (which might be unreasonable to do with
> a web service), but also previous game tree state that the engine might
> want to reuse across moves.
> If I was stopping engines mid-game to conserve resources, I'd do this
> only about some period of inactivity (be it 15s or 120s).
Thanks for this, Petr Baudis. These are great points which would warrant
not terminating engines mid-game. I was under the impression previously
that engines would only perform the calculations they need to when asked to
"genmove", but I can see that it would be possible (and maybe preferable)
to do some thinking while the opponent is playing.
Folkert van Heusden,
- a networked protocol so that you can also run them from your own
> computer
> - with clients that talk the usual gtp protocol
This is good feedback that sort of makes a compromise between CGOS and
Baduk.io, where if you want to run the engines on your own hardware, you
can. Definitely an interesting idea.
- make a simple json emitting page that one can poll to collect
> statistics, e.g. { "stop" : "1300" } where stop is the name of my
> program and 1300 a not very realistic (it plays very bad) elo-rating
Good suggestion! Baduk.io's web frontend is a single-page web app that just
consumes information from a JSON API already, so I'd just have to document
how that information is exposed so engine developers can consume it. Thanks!
On Thu, Jan 15, 2015 at 12:27 AM, folkert wrote:
> Very cool!
> What I would like:
> - a networked protocol so that you can also run them from your own
> computer
> - with clients that talk the usual gtp protocol
> - the program interfacing between the server and the client-engine in
> something like c/c++/python and open source so that one can more
> easily run it everywhere. also please not difficult to setup, eg no
> oauth and other stuff requiring gui's (I usually run my chess/go
> engines on systems that do not even have a video card)
> - elo is my current preferred rating as I "know it" but I could "learn"
> kyu/dan as well
> - make a simple json emitting page that one can poll to collect
> statistics, e.g. { "stop" : "1300" } where stop is the name of my
> program and 1300 a not very realistic (it plays very bad) elo-rating
>
> On Mon, Jan 12, 2015 at 05:34:44PM -0800, Chris LaRose wrote:
> > Hi,
> >
> > I'm actually working on something similar at http://baduk.io. Right now,
> > you can log in an play against a handful of bots over the web, but one
> day
> > I'd love to make it so you can add your own bots to let them compete
> > against the others. It's not quite ready for the public, but I'm working
> to
> > get something small working quickly. Unlike CGOS, the bots are all
> > hosted--they all run inside Docker (http://docker.io/) containers. The
> > Dockerfiles I've written for a few public bots are available at my github
> > repository https://github.com/baduk-io/ai-dockerfiles.
> >
> > What sorts of things would you expect from such a service? I was planning
> > on modeling baduk.io after CGOS in a lot of ways as far as the rules
> that
> > are used (area scoring, no dead stones removed, etc), and distinct
> ratings
> > for 9x9, 13x13, and 19x19 boards. What sorts of improvements do you think
> > could be made in a new service? Do you have a preference for ELO ratings
> > over kyu/dan ratings?
> >
> > Chris LaRose
> >
> > On Fri, Jan 9, 2015 at 2:47 AM, folkert wrote:
> >
> > > Hi,
> > >
> > > I have the feeling that cgos won't come back in even the distant future
> > > so I was wondering if there are any alternatives?
> > > E.g. a server that constantly lets go engines play against each other
> > > and then determines an elo rating for them.
> > >
> > >
> > > Folkert van Heusden
> > >
> > > --
> > > Afraid of irssi? Scared of bitchx? Does xchat gives you bad shivers?
> > > In all these cases take a look at http://www.vanheusden.com/fi/ maybe
> > > even try it or use it for all your day-to-day IRC conversations!
> > > ---
> > > Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com
> > > ___
> > > Computer-go mailing list
> > > Computer-go@computer-go.org
> > > http://computer-go.org/mailman/listinfo/computer-go
>
> > ___
> > Computer-go mailing list
> > Compute
Re: [Computer-go] What's a good playout speed?
Hi all, I am working on a successor to Valkyria written in C++. I agree with Petr here. I am still tweaking the basic engine and I have still lot to fix. With a sloppy implementation initially I have 20 kpps for 9x9 and 4.5 kpps for 19x19. When the basic stuff is running I will build feature detection directly into the engine first based on my experiences from Valkyria. It will not be fast when board features are calculated but a lot of features necessary for heavy playouts will come for almost free. Magnus Persson On 2015-01-14 13:32, Petr Baudis wrote: On Wed, Jan 14, 2015 at 01:26:44PM +0100, Urban Hafner wrote: If I did it again, after picking some really low-hanging optimization fruit I'd rther focus on implementing playout heuristics, only then coming back to optimize the board code when it is clear what all information I need to compute the heuristics efficiently (list of up to N libs, maybe pre-computed pattern hashes, etc.). ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] CGOS back online
On Jan 14, 2015, at 7:30 AM, folkert wrote: > I've connected a couple of programs but nothing happens. > They login and that's about it. Same here. http://cgos.boardspace.net/9x9/standings.html was updated last about 2 years ago. Christoph ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] CGOS back online
Hi, I've connected a couple of programs but nothing happens. They login and that's about it. On Wed, Jan 14, 2015 at 04:14:31AM -0500, Joshua Shriver wrote: > CGOS is back on boardspace.net and running. Please feel free to try > and connect like you use to and let me know. I'll be monitoring it. > It's just as it was when it was taken down after Mr. Daileys passing. > > I'm grateful for Dave Dyer for hosting it in the past, and for > keeping a clean copy of the system as Don originally wrote it. > > Originally I wanted to host it myself but the code is rather delicate > and I never could get it running properly. I'll still be maintaining > it and keep track of it's ups and downs but for all intent and > purposes it "just worked" back then so that's why I asked to use > boardspace again to get things rolling again. > > Thank you for your understanding, > -Josh > ___ > Computer-go mailing list > Computer-go@computer-go.org > http://computer-go.org/mailman/listinfo/computer-go Folkert van Heusden -- MultiTail är ett flexibel redskap för att följa en eller flera logfiler, utföra kommandon, filtrera, färglägga, sammanfoga, o.s.v... -- Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] alternative for cgos
Very cool!
What I would like:
- a networked protocol so that you can also run them from your own
computer
- with clients that talk the usual gtp protocol
- the program interfacing between the server and the client-engine in
something like c/c++/python and open source so that one can more
easily run it everywhere. also please not difficult to setup, eg no
oauth and other stuff requiring gui's (I usually run my chess/go
engines on systems that do not even have a video card)
- elo is my current preferred rating as I "know it" but I could "learn"
kyu/dan as well
- make a simple json emitting page that one can poll to collect
statistics, e.g. { "stop" : "1300" } where stop is the name of my
program and 1300 a not very realistic (it plays very bad) elo-rating
On Mon, Jan 12, 2015 at 05:34:44PM -0800, Chris LaRose wrote:
> Hi,
>
> I'm actually working on something similar at http://baduk.io. Right now,
> you can log in an play against a handful of bots over the web, but one day
> I'd love to make it so you can add your own bots to let them compete
> against the others. It's not quite ready for the public, but I'm working to
> get something small working quickly. Unlike CGOS, the bots are all
> hosted--they all run inside Docker (http://docker.io/) containers. The
> Dockerfiles I've written for a few public bots are available at my github
> repository https://github.com/baduk-io/ai-dockerfiles.
>
> What sorts of things would you expect from such a service? I was planning
> on modeling baduk.io after CGOS in a lot of ways as far as the rules that
> are used (area scoring, no dead stones removed, etc), and distinct ratings
> for 9x9, 13x13, and 19x19 boards. What sorts of improvements do you think
> could be made in a new service? Do you have a preference for ELO ratings
> over kyu/dan ratings?
>
> Chris LaRose
>
> On Fri, Jan 9, 2015 at 2:47 AM, folkert wrote:
>
> > Hi,
> >
> > I have the feeling that cgos won't come back in even the distant future
> > so I was wondering if there are any alternatives?
> > E.g. a server that constantly lets go engines play against each other
> > and then determines an elo rating for them.
> >
> >
> > Folkert van Heusden
> >
> > --
> > Afraid of irssi? Scared of bitchx? Does xchat gives you bad shivers?
> > In all these cases take a look at http://www.vanheusden.com/fi/ maybe
> > even try it or use it for all your day-to-day IRC conversations!
> > ---
> > Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com
> > ___
> > 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
Folkert van Heusden
--
MultiTail is een flexibele tool voor het volgen van logfiles en
uitvoer van commando's. Filteren, van kleur voorzien, mergen,
'diff-view', etc. http://www.vanheusden.com/multitail/
--
Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com
___
Computer-go mailing list
Computer-go@computer-go.org
http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] CGOS back online
BTW, which GTP commands are required for a bot to play on CGOS? The basic commands are probably not quite enough, right? At least time_left and time_settings should be implemented if I remember it correctly. Urban On Wed, Jan 14, 2015 at 10:43 AM, Urban Hafner wrote: > Great! Thank you Joshua. For me personally that will be a great help as > testing if my new bot works is so much easier on CGOS than during a KGS > tournament. :) > > Urban > > On Wed, Jan 14, 2015 at 10:14 AM, Joshua Shriver > wrote: > >> CGOS is back on boardspace.net and running. Please feel free to try >> and connect like you use to and let me know. I'll be monitoring it. >> It's just as it was when it was taken down after Mr. Daileys passing. >> >> I'm grateful for Dave Dyer for hosting it in the past, and for >> keeping a clean copy of the system as Don originally wrote it. >> >> Originally I wanted to host it myself but the code is rather delicate >> and I never could get it running properly. I'll still be maintaining >> it and keep track of it's ups and downs but for all intent and >> purposes it "just worked" back then so that's why I asked to use >> boardspace again to get things rolling again. >> >> Thank you for your understanding, >> -Josh >> ___ >> Computer-go mailing list >> Computer-go@computer-go.org >> http://computer-go.org/mailman/listinfo/computer-go > > > -- Blog: http://bettong.net/ Twitter: https://twitter.com/ujh Homepage: http://www.urbanhafner.com/ ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] What's a good playout speed?
On Wed, Jan 14, 2015 at 1:32 PM, Petr Baudis wrote: > > Note that the mistake I did with Pachi is that I worked hard to optimize > the Go board (playout) code with minimal functionality to perform > a playout correct by the rules, only then discovering how much other > information I need to keep to add some useful heuristics to the playout, > and maybe the hotspots change then. > > If I did it again, after picking some really low-hanging optimization > fruit I'd rther focus on implementing playout heuristics, only then > coming back to optimize the board code when it is clear what all > information I need to compute the heuristics efficiently (list of up to > N libs, maybe pre-computed pattern hashes, etc.). > That's a good point Petr! But as I started with 50pps on 9x9 (yes, really) I actually needed to do some optimisations as this made the bot completely unusable. There are still a few low hanging fruit (or learning opportunities for me), but it's probably time to finally start implementing UCT. BTW, the board.c from pachi is very readable and it (together with Board.java from the current Orego) serves me well as an inspiration on how to design the board. Urban ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] What's a good playout speed?
On Wed, Jan 14, 2015 at 01:26:44PM +0100, Urban Hafner wrote: > Hey Jason, > > thanks. Rust is probably not very well optimised, yet. It hasn't even > reached 1.0 and there are still many language changes happening. But I > attribute the slowness to the fact that I'm not used to writing code in > compiled languages (I'm a web developer). :P But learning how to write > performant code is half the fun and I guess I still have a bit of work to > do here. Note that the mistake I did with Pachi is that I worked hard to optimize the Go board (playout) code with minimal functionality to perform a playout correct by the rules, only then discovering how much other information I need to keep to add some useful heuristics to the playout, and maybe the hotspots change then. If I did it again, after picking some really low-hanging optimization fruit I'd rther focus on implementing playout heuristics, only then coming back to optimize the board code when it is clear what all information I need to compute the heuristics efficiently (list of up to N libs, maybe pre-computed pattern hashes, etc.). -- Petr Baudis If you do not work on an important problem, it's unlikely you'll do important work. -- R. Hamming http://www.cs.virginia.edu/~robins/YouAndYourResearch.html ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] What's a good playout speed?
Hey Jason, thanks. Rust is probably not very well optimised, yet. It hasn't even reached 1.0 and there are still many language changes happening. But I attribute the slowness to the fact that I'm not used to writing code in compiled languages (I'm a web developer). :P But learning how to write performant code is half the fun and I guess I still have a bit of work to do here. Urban On Wed, Jan 14, 2015 at 1:11 PM, Jason House wrote: > Housebot was probably on the low end with 10kpps on 9x9. Libego was > probably the highest with 100kpps. I attribute some of the difference to > compiler maturity (D vs. C++). I don't know how rust will perform. > On Jan 14, 2015 3:14 AM, "Urban Hafner" wrote: > >> Hey everyone, >> >> I'm currently in the early stages of writing my own Go engine and right >> now I'm trying to make my playouts reasonably fast. I've come a long way in >> the past few days. Probably not because the payouts are really fast right >> now, but because they were just so slow before. :) Right now I'm at >> ~2000pps on 9x9 and ~1000pps on 19x19. This is for playouts with simple ko >> and suicide checks and no concurrency. Now I wonder if this is fast enough >> to even start thinking about implementing a UCT/MCTS player and also if >> there's something I'm missing with the playouts, e.g. is the suicide check >> necessary? >> >> Thanks for your input! >> >> Urban >> >> ___ >> 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 > ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] alternative for cgos
On Wed, Jan 14, 2015 at 09:25:28AM +0100, Marc Landgraf wrote: > I'm not convinced about that concept, tbh. > People put a lot of work to optimize their bots, include GPU usage and > figure out, how to use Pondering the best way. And then you want those > programmers to remove those features and put work into making their bots > run on your system, just to "level the playing field". I don't think you're being that fair to Chris. He has his own goals for a service he wants to provide, which isn't primarly a CGOS replacement, and is perfectly fine for the purposes he wants. And it's fine to suggest that maybe something like this could work as a CGOS substitute, even though in the discussion it's turning out that probably that's not the case. (Still, I'd love to read about the results in a mini-tournament between the bots, if possible multiple tounrmanets with different resource limits. And I'm now excited about possibly reusing the baduk.io's Docker infrastructure for the EGC2015 Computer Go tournament I'm planning, where bots will run on equal hardware too - it'd make the logistics quite easier for many competitors.) > 2015-01-14 3:04 GMT+01:00 Chris LaRose : > > Also, bots that aren't currently playing a game can be > > terminated and won't consume resources. Starting and stopping containers is > > so fast that I can afford to only start bots immediately after its opponent > > plays, request a single move, and terminate it. Note that while this is fine for GNUGo, it may not be for some of the more advanced bots: * Initialization time may be non-trivial. If you use Pachi with the patterns database (which is extremely recommended), it takes a few seconds to load. * You lose not just pondering (which might be unreasonable to do with a web service), but also previous game tree state that the engine might want to reuse across moves. If I was stopping engines mid-game to conserve resources, I'd do this only about some period of inactivity (be it 15s or 120s). -- Petr Baudis If you do not work on an important problem, it's unlikely you'll do important work. -- R. Hamming http://www.cs.virginia.edu/~robins/YouAndYourResearch.html ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] What's a good playout speed?
Housebot was probably on the low end with 10kpps on 9x9. Libego was probably the highest with 100kpps. I attribute some of the difference to compiler maturity (D vs. C++). I don't know how rust will perform. On Jan 14, 2015 3:14 AM, "Urban Hafner" wrote: > Hey everyone, > > I'm currently in the early stages of writing my own Go engine and right > now I'm trying to make my playouts reasonably fast. I've come a long way in > the past few days. Probably not because the payouts are really fast right > now, but because they were just so slow before. :) Right now I'm at > ~2000pps on 9x9 and ~1000pps on 19x19. This is for playouts with simple ko > and suicide checks and no concurrency. Now I wonder if this is fast enough > to even start thinking about implementing a UCT/MCTS player and also if > there's something I'm missing with the playouts, e.g. is the suicide check > necessary? > > Thanks for your input! > > Urban > > ___ > 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] CGOS back online
Great! Thank you Joshua. For me personally that will be a great help as testing if my new bot works is so much easier on CGOS than during a KGS tournament. :) Urban On Wed, Jan 14, 2015 at 10:14 AM, Joshua Shriver wrote: > CGOS is back on boardspace.net and running. Please feel free to try > and connect like you use to and let me know. I'll be monitoring it. > It's just as it was when it was taken down after Mr. Daileys passing. > > I'm grateful for Dave Dyer for hosting it in the past, and for > keeping a clean copy of the system as Don originally wrote it. > > Originally I wanted to host it myself but the code is rather delicate > and I never could get it running properly. I'll still be maintaining > it and keep track of it's ups and downs but for all intent and > purposes it "just worked" back then so that's why I asked to use > boardspace again to get things rolling again. > > Thank you for your understanding, > -Josh > ___ > 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
[Computer-go] CGOS back online
CGOS is back on boardspace.net and running. Please feel free to try and connect like you use to and let me know. I'll be monitoring it. It's just as it was when it was taken down after Mr. Daileys passing. I'm grateful for Dave Dyer for hosting it in the past, and for keeping a clean copy of the system as Don originally wrote it. Originally I wanted to host it myself but the code is rather delicate and I never could get it running properly. I'll still be maintaining it and keep track of it's ups and downs but for all intent and purposes it "just worked" back then so that's why I asked to use boardspace again to get things rolling again. Thank you for your understanding, -Josh ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
[Computer-go] Teaching Deep Convolutional Neural Networks to Play Game X
HI, did someone think about the question for which other games these Deep Convolutional NN may be helpful? For instance, from the portfolio of turnbased server littlegolem.net ? Ingo. ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] alternative for cgos
I'm not convinced about that concept, tbh. People put a lot of work to optimize their bots, include GPU usage and figure out, how to use Pondering the best way. And then you want those programmers to remove those features and put work into making their bots run on your system, just to "level the playing field". 2015-01-14 3:04 GMT+01:00 Chris LaRose : > Thanks everyone for all the feedback! Sorry I hijacked the thread! > > There are definitely some big pros and cons to a hosted, containerized > environment for Go bots. I've replied to some comments below: > > Won’t hosting limit your usability? With cgos I can build and immediately >> test on cgos on my development machine. With your service, how do I get my >> new executable to run? > > > You're absolutely right. The workflow for developers will have some > additional overhead. The nice thing, though, is that Docker is very good at > providing a way for an application to run virtually anywhere in a > consistent fashion. That is, if you can get your bot running in a Docker > container on your own machine, then it's pretty likely that it'll run on > Baduk.io's Docker host without a hitch. > > If my engine uses a GPU or is a multinode cluster, how does that run on >> your docker service? > > > Right now, I don't have plans for supporting such bots. But as Urban > Hafner replied, it puts everyone on a level playing field--all bots have > access to the same exact resources. As some have noted, maybe that means > Baduk.io wouldn't serve the same purpose as CGOS. That's fine by me. > > At first, I want Baduk.io to be a place where players, especially > beginners can get a few games in against bots. Because the bots are hosted, > I'm not limited in the number of simultaneous games I can play against one > bot. As far as I understand it, it GNU Go is playing against five different > people on KGS, that means that there are five different instances of GNU Go > running on people's machines someone in the world. If someone else wants to > play, they can't. Also, bots that aren't currently playing a game can be > terminated and won't consume resources. Starting and stopping containers is > so fast that I can afford to only start bots immediately after its opponent > plays, request a single move, and terminate it. > > Anyway, Baduk.io is largely only a proof-of-concept right now--I'll have > to post a message to the list when I get a little further working on it. > > On Wed, Jan 14, 2015 at 8:33 AM, Joshua Shriver > wrote: > >> If you can send me a binary that would be greatly appreciated. Trying >> to build some anchors now. >> >> -Josh >> >> On Tue, Jan 13, 2015 at 5:11 PM, Hideki Kato >> wrote: >> > Shilver, >> > >> > I'll be able to run FatMan1, the anchor for 9x9, on my site, if >> > necessary. Or, it's also possible to send you its binary and password >> > so that you can run it on your site. >> > >> > Hideki >> > >> > Joshua Shriver: > xwb5...@mail.gmail.com>: >> >>I'll try and get CGOS back online before this weekend. Technically it >> >>should be running now, but there were several issues. In order to use >> >>it now you take the cgos client for your architecture and you have >> >>to specify cgos.computergo.org manually since the binaries are >> >>hardcoded to the old boardspace address. I've had some troubles >> >>unbundling the binaries and rebuilding the executables with TCL. >> >> >> >>Rankings are also an issue as well which is something I'll have to >> >>change in the code to make sure anchors and their predefined ELO >> >>ratings are used. >> >> >> >>Will try and make a better write-up on how to connect. Hopefully this >> >>weekend I should have the anchors running 24/7 and some people can try >> >>connecting. >> >> >> >>I'll flush the old data and in terms of games and we'll start with a >> >>fresh slate. Though all the data even from years past are still >> >>available though for historic reasons and for anyone who wants the >> >>SGF's. >> >> >> >>-Josh >> >> >> >>On Fri, Jan 9, 2015 at 5:47 AM, folkert wrote: >> >>> Hi, >> >>> >> >>> I have the feeling that cgos won't come back in even the distant >> future >> >>> so I was wondering if there are any alternatives? >> >>> E.g. a server that constantly lets go engines play against each other >> >>> and then determines an elo rating for them. >> >>> >> >>> >> >>> Folkert van Heusden >> >>> >> >>> -- >> >>> Afraid of irssi? Scared of bitchx? Does xchat gives you bad shivers? >> >>> In all these cases take a look at http://www.vanheusden.com/fi/ maybe >> >>> even try it or use it for all your day-to-day IRC conversations! >> >>> >> --- >> >>> Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com >> >>> ___ >> >>> Computer-go mailing list >> >>> Computer-go@computer-go.org >> >>> http://computer-go.org/mailman/listinfo/computer-go >> >>_
[Computer-go] What's a good playout speed?
Hey everyone, I'm currently in the early stages of writing my own Go engine and right now I'm trying to make my playouts reasonably fast. I've come a long way in the past few days. Probably not because the payouts are really fast right now, but because they were just so slow before. :) Right now I'm at ~2000pps on 9x9 and ~1000pps on 19x19. This is for playouts with simple ko and suicide checks and no concurrency. Now I wonder if this is fast enough to even start thinking about implementing a UCT/MCTS player and also if there's something I'm missing with the playouts, e.g. is the suicide check necessary? Thanks for your input! Urban ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
Re: [Computer-go] alternative for cgos
> Starting and stopping containers is so fast that I can afford to only start > bots immediately after its opponent plays, request a single move, and > terminate it. Thats interesting. And impressive. ___ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
