Nath-project

User avatar
Hagbard
Posts: 707
Joined: Sat Mar 14, 2015 6:07 pm
Location: Copenhagen
Contact:

Nath-project

Post by Hagbard » Sat Oct 21, 2017 8:26 pm

Nath has started a project to investigate deeper into the mysteries of tafl games.

For use by the project is introduced the option "No Timing + No Rating".

Project discussion has been going on in the small forum for a while,
here's a copy of the discussion:
10-20/15:40 nath: Here's the last part:
On top of that there is commentary at all about the question when one should focus more on a single corner and when try focus on multiple ones at the same time or phasing it from the black side of things when to limit the king mobility and when try to secure specific corners/edges of the board.
Tbh this is a question where top players disagree on a lot, but none the less there are common ideas that shape their though process even though judgment about the importance drastically varies.

Even just explaining the most basic concepts in a manner where it's easy to grasp, would take far more than the triple amount of characters the two guides you mentioned currently have.
I'd gladly contribute, I don't feel that I am good at getting examples that demonstrate the concept in a way such it's easy to grasp...

10-20/15:37 nath: @altti: Let's talk about the interesting theory part! :)
Afaik there are a ton of concepts without publicly available guides about.
Regarding the sources you just mentioned: The most obvious thing to be is
that there is no introduction to edge win forts at all - neither how look,
nor how they are build correctly, no guide how to force weakness upon a
special edge, nor specific examples how to stop white from building one.
Second there is no comparison of different defensive patterns listing
advantages and disadvantages of the single pattern and ideas how to
disrupt the individual building series of the pattern. Even if the
destination pattern is the same there are a ton of different way to get
there which call for different counterplay from white to exploit different
weaknesses.

I don't bother so much that there is no list of common opening sequences,
I much rather mind that there is no explanation of common opening ideas.

Besides from that there a few strategic rules that seem to be used by a
few (but even not all) of the player you'd call *strong* (I don't intent
to argue about the word in this response), mainly methods for measuring
the importance of second and third rows, but as well the concept of sparse
rows.

10-20/15:36 nath: @altti: Sorry, I was very busy recently.
To me it's dull to play a game where you know that the exact move doesn't matter anyways unless somebody makes a huge mistake. If there's a big set of best moves (besides the counters of obvious short term wins like blocking a one move escape from a king) obviously don't effect the outcome of the game, it seems dull to me. That itself isn't connected to winning or loosing in itself.
I think it's public knowledge that players who care about winning tend to play out even such positions quite much, since they care about the question who wins (a big likelihood doesn't have to materialize if the other player doesn't care enough and hence has different priority), not just how interesting the position is as a analytical question.
I didn't get bored of winning since I was never interested in the winning in the first place. I don't think the theoretical question is of philosophical nature, but you think I personally never really wasn't interested in anything else.
Altogether I am sorry, I tried to make you understand me and made you read that long text...

10-17/12:38 Hagbard: Mange tak!!!

10-17/12:37 OdinHimself: Thank you Høvding! Lad Odin fylde dine net med sild!

10-17/12:30 Hagbard: correct

10-17/12:18 OdinHimself: @Hagbard: So, games marked as No Timing are also Not Rated, correct?

10-17/07:18 Hagbard: @OdinHimself: yes No Timing appears an important option.

10-17/00:34 altti: if there are a series of moves which have not been analyzed I will eat my words.

10-16/00:10 altti: you seem to be very good at making the proper moves required to result in victory. it's not dull to play from a challenged position. is it dull to play from a superior position? are you so bored with winning that you need to seek some philosophical positions on the tafl board? I think we already have a statistical analysis of all the games played up to last year. it was quite enlightening, as are crusts' and cyningstans overview of strategy. there you may find your answers.

10-15/22:59 nath: @Plantagenet: I never said it is. I said that I'd rather expect it to be in favor of black from my subjective perspective and the way my games play out.
Yet this might be related to personal tendencies we both share. My gut feeling tells me otherwise, but I don't yet have the data to back it up.

I don't think this gut feeling (even if it's common along the top players), is quite enough to request a change of rules. If it would be a given that if someone sits down at a table, any white player had less than 5 % of winning, things would stand differently. But looking at recent games of you, we aren't there yet. Just to name one particular issue that should have been addressed first in my opinion, as things stand we haven't fully embraced the edge fort as natural part of opening strategy yet. We are not measuring the impact this shape has on the game, if we are just dismissing it as a mere endgame scenario. Pressure isn't just related to third row tactics, that can be suppressed - at least judging from your performance. The conclusions we arrived back in 2013 were similar, yet we weren't even completely sure about tactical corner assaults.

I think we should further understand the nature of this game before jump to changing the rules that drastically. What you suggested is a completely different game. I don't want to stop anybody from trying that out, but I don't think the results reveal anything particular about the Copenhagen rulese

10-15/22:34 nath: @altti: I can play the psyche game - it's quite common at tournaments. Yet this isn't pushing the playstyle of the community forward.
I think the main difference is that I don't personify the winning nor the move. It's not relevant for me who makes the move nor whether he wins with that. It comes down to the concept of a strong position that any decent player can win regardless of the opponent strength nor psyche. I don't care if somebody wins more by executing these positions in a more stable way.
I don't care about executing dull positions (which is a part of victory - looking at your playstyle you care about that a lot), I rather care about interesting positions and devote my time trying to understand them.
If you care about victory, you care a lot about stability and a special player. I care about pushing the understanding of the community which shouldn't necessarily increase the chance of victory for a particular player, since I want free access to resources by all players.

10-15/14:32 OdinHimself: Hagbard, thank you for non-rated/non-limted games, I think this will help a lot! Now many strong players who care about their rating and free time could play against each other without concerns. This is going to be the beginning of the new tafl age!

10-15/10:51 Plantagenet: I indeed think Hnefatafl is unbalanced to white and you said yourself the same thing two years ago.

10-14/21:58 altti: are we talking about making the best possible move that will result in an ultimate victory? then we are talking about winning. if you make that move quickly it can disturb your opponent. (though waiting for an opponents move can set one to second thoughts. that reveals ones own doubts.) this is a psyche game, as is chess.
as far as the caliber of opponents - challenge those you wish to play. there are other sites to test as well.

10-14/20:30 nath: @Plantagenet: What do you think is unbalacned about the game? What problem are you trying to solve? I think everybody is trying to solve different problems. I am trying to simplify rules, since I have no idea about which side is having the upper hand.

@Steiger: I doubt rules are very interesting. But I'd very much like to write something about strategy and tactics. I have some ideas what to write about, but I always fail at getting decent examples for the principles I introduce to demonstrate the effects. If somebody wants to do a joined projects regarding a book about strategies, I'd be very happy...just contact me in that case.

10-14/20:25 nath: I don't want a player to stay on top, since he INVESTED so much, but I much rather like him to stay because he INVESTS so much. Especially openings seem much like a collaborative process to me than the brilliant mind of a single player.

10-14/20:25 nath: @edmond-dantes: I also want to question whether chess actually ended. Not long ago the current chess world champion (Carlsen) hold the three popular world champion titles (normal (2+ hours), fast (15+ min), lighting (5 min)). I watched a lot of his games and sometimes with commentary. It's hilarious what other grand masters say about his opening style. Quite often the commentator mentions "this is a very passive move". His positions after the openings are quite frequently bad (often even worse to what I play at my bad level). But the opponent can't read to far ahead, since he just plays anything no matter whether it's called strong or not. His midgame play is something else. He just sits down and plays what he likes. The game is more alive than ever.

I notice we have a completely different view regarding how things should be.
You seem think that winning is the relevant thing, the great thing one plays for. You further suggest that winning should be granted to those who put in the biggest amount of hard work.
I am different, since I don't play for winning. I also think being able to win should be regarded as a gift and not as sweat. I don't mind sweating training with my gift, but I don't take skill for granted just if I invest some work.
At the same time I want to gift the community with my work and not myself. I want to encourage other players to keep working to push the game and the community forward. I don't want a player to stay o

10-14/20:24 nath: @edmond-dantes: I never minded any of these. The rating at this site isn't the best, but that's not the issue at hand.
Please note that the current "solution" is not at all what I suggested, I just think it's better than what we currently have.
Besides - Hnefatafl has simpler rules than chess, but the game itself is more complicated. Humanity has solved a lot of games that are supposed to be more difficult (even though chess is simpler). Hnefatafl is special in two ways: It's very hard to determine who has the upper hand. I don't know know a player who's able to see that in "normal" positions. Games like Chess, Shogi or Xianqui have the simple metric "just cound how has move material". That doesn't apply to Hnefatafl. It's also impossible to compare the progress black and white made, since the game asymmetric nature. You are underestimating the complexity of Hnefatafl.
The complexity in openings in chess is much older than engines - that's not at all related.
It#s true that chess has a notable amount of theory, but that didn't kill the game at all. It's true that there are some experts out there who win the games in endgames a lot. But winning the game in the middlegame ist still possible (and happens) at any skillevel. Winning in the opening on contrary happens quite rarely even if one player doesn't know the opening. Sine players who don't know the opening doesn't follow the lines...
I also

10-14/20:11 nath: @altti: The correspondence chess players aren't strong in 2 hour games. Besides I don't see how that is related to any earlier discussion we had. Non the less, I like the idea of more games that played in a live manner like edmond-dantes does a lot.

The point I made regarding the time limit is, that cutting the limit down to 1.25 days just help players that focus their entire life on the game, since I doubt that any of our players spent more than 11 hours on a single move. It's just helping players with no live and/or few open games.

10-14/17:53 altti: strong chess players are strong under time constraints as well.

10-14/15:06 Plantagenet: I did offer new rules for Hnefatafl but noone wanted to apply them because, in their view, it would complicate the game while they didn't account for the rules' complexity of other games like Xiang Qi. Since, I think I make these ones more simple: Copenhagen rules + King may make a short orthogonal jump over one of his own men from or to restricted squares + King is captured against a cormer square by one opponent + Black can't cross over the throne.
I still don't know if it's enough for counterbalancing the game.

10-14/10:11 edmond-dantes: @nath: when we want a game to have a competitive character there is a must-have time factor and rating there is no way

10-14/10:07 edmond-dantes: @nath keep in mind that the game does not have many initial variations as a whole. The complexity is far less than chess, the game is much simpler. The engine will kill the game, because the game will end. You see in chess how the top chess players play 20-30 moves theory, and the game is won or lost only at the end. Developing a new starting line takes a lot of time and effort on the discoverer but then everyone is already using the news on ready. I am against keeping the game database as a new player for two months can see winning sewn lines and beat the best players who have made hard work

10-14/07:24 Steiger: Write a new book about tafl rules and strategies ;-)

10-14/04:21 nath: @Kratzer: At one point in time I thought about making an own website, but I didn't want to split our small community. I don't mind putting effort into organizing an tournament, but if our community splits we are even off worse. I don't mind competitive players - they might even help we with that dream moving forward. Despite that I never was and never will be a competitive player myself. I rather study interesting moves.

@OdinHimself: I can't promise that I will be able to give you that new meaning. But I am confident if we work on it, we will find it along the way.
I'll try to get accustomed again, I played less than 4 games in the last 3.5 years. Then I'll try to push forward further. If anybody has specific ideas about how to get some decent materials about strategies out or how help players learning curve in general, how to try out new moves or is interested in building an engine let me know.

10-14/00:20 nath: The character limit is quite hard to foresee...I always have to type half of messages again and again...

It's quite the opposite. I care about learning and understanding Hnefatafl. I want to learn from other players. I want to learn and push the community with my knowledge. So please surprise me!

@edmond-dantes: Regarding the titles: I fought against Schachus. I found in him what I was looking (and am looking for again): a player that is stronger than I am. I don't hope to archive the great level of white play he showed back then, yet he was a good teacher regarding the white pieces to me.
I know to this very day that he was stronger back then than I am now (or will be this year). What's the point in holding a title if you only hold it because the stronger players aren't interested?
Both Schachus and me were very annoyed about the short times. That pushes hard and consequently sabotages high quality play. You notice that Schachus lost less than a hand full of games at this site before the new time limit came and all his games timed out.

10-14/00:01 nath: @edmond-dantes: I don't think quite got what I wanted to say. I never claimed that I's be strong - I am not. I never complained about not having any players that are able to challenge me at this website. That might have been true for 2012 and 2015/16, but that doesn't mean I was the strongest player (since I was not).

I am not saying I understand even the most basic things about this game, since I am not. I'm not even complaining about the fact that nobody does. I am only wording my concerns about developing play. I want to develop beyond a level of weak players (the level I currently have).

I am a rather weak chess player, but yet I am a by far stronger chess player than Henfatafl player. The fact that a new player is able to compete in the world championships final after two months rather proves my point - there is no school building up this game.

I am puzzled about that fact that openings look quite what we developed back in 2013. Even some of the mistakes we fixed up during the winter 2013/2014 weren't adopted. I don't think this a problem in it self - it's just another symptom for the state this community is in.

People play for entirely different reasons. If one doesn't want to become a strong player that's a perfectly fine way to be. Yet I care about the development of strategies and tactics. I was always looking to develop a game and a community at the same time.
That doesn't mean that I care to win, quite to opposit

10-13/17:34 OdinHimself: Thank you Kratzer, thank you Hagbard, thank you Edmond-dantes. Titles mean a lot. I'd lie if I say titles and ratings are not important. Probably this is the reason previous tafl champions to avoid defending their titles and stay "Undisputed". But this is something I understand and accept. I am ready to lose all if this is the price for upgrading Hnefatafl and me as a player. I am eager to find out if there is another higher level of playing. To discover the unknown if there is something hidden from us. If nath could help us to find a new meaning of tafl, we all should stay beside him and give our effort, even if it turns into a perennial quest. All for one and one for all is what I profess!

10-13/16:30 Kratzer: @ OdinHimself: Well spoken, tafl brother!

10-13/15:50 edmond-dantes: @ nath:I do not understand how you are interested in the development of the game you do not play. Progress has when the strong players play with the weak so the weak are developing, the competition is developing and the game develops and popularizes (example how OdinHimself did with me and the result is visible participation in the final, only for two months play).You are not right that everyone is weak and you will soon be convinced ...

10-13/15:40 edmond-dantes: yes, we need to thank Hagbard for the effort it puts for the site. But in the future you can really think about the shape of the championship. Now I think it is not in the best format

10-13/15:01 Hagbard: Of course. There's the "No timing" option for project games. It could easily be extended into No timing + No rating.

10-13/14:51 OdinHimself: This has turned into a Nordic Thing, so I will speak as Hovding infront of the meeting. I don't like the things I love to be ridiculed. Hnefatafl is an important part of my life. Our community is important for me, I respect every single tafl player and became friend with many of you. I respect everyone's opinion. If there is a way for developing our game I want to help finding it. No need of harsh words. Let's work as team, as tribe, as tafl brothers. I am asking Hagbard if he could create new distinct working group, a project, with non-rated and non-limited games, where volunteers could play against each other only with educational purpose. Main goal - improving our beloved game. Skal!

10-13/12:45 altti: sorry I'm so slow on the pick up. seems to me you wish only to play a single game against the single winner of the rest of us so as not to be bothered by the tedium. if you want to play... play. if you want to win.... play.

10-13/09:41 Kratzer: @ nath: I find your propositions for new tournament rules very interesting. Maybe that should be discussed further in the forum?
Besides, I would not go as far as to say "fun tournament", "doesn't even care to learn", "no strong players". Sorry, but how can you know that if you do not participate? Maybe some of us would have surprised you. Not me though ;)

10-13/09:00 Kratzer: @ nath: Thanks for your input. You know, this a small web community with a lot of fluctuation. People come, people go. Why not create a blog or website of your own? To make this professional one would have to collect and centralise as much information about strategy as possible in one place and attract the masterminds of the game. Right now, it's all over the web, with crust, mmagari, Mr. Nielsen's website and a lot more. You tell us that your time is very limited – well, I think that applies to most of us. I am not and will probably never be one of the "strong players" you mention, but that does not mean I am not interested in their insights and lore

10-13/03:23 altti: this is a relatively new organization, wanting in bureaucracy or even participants. It has not yet grown into a system such as the Chess Federation, which functions the way you describe tournament play. The host is one man, the rest of us are visitors on his site. This is an autocracy. I'm sure he welcomes the input but there are no bodies to handle the work to organize beyond what he already does. this is my personal assumption.

10-13/01:00 nath: Since I got cut off, I will post that paragraph again:
I don't want to strip the public of their fun tournament. At the same time for me strong games are much more interesting to watch, so I want to encourage players to play them as much as possible. If one use an open tournament just for qualification purposes, one could get creative how to assign Valhöll spots. For instance one could assign the looser of the last championship duel a fixed place, take the first four players of the tournament that didn't participate in last years duel and add in three players with the highest ranking that are not yet qualified.

That would add much more meaning to carry a title since you have it until you loose in a match for your title. It would also make for more games between skilled players, since it doesn't limit the ability at all to play skilled games, but also adding a lot more focus and empathy to the challenger and the orld champions game - not only because they carry a lot of weight, but also because they don't interfere with any other important games, so the players can devote all their thoughts to these games.

10-13/00:59 nath: @edmond-dantes: I can answer why I don't participate anymore. I stopped playing games with weird time limits, that put a harsh constraint on my daily schedule. Besides practically all of the games feel rushed in an unreasonable manner to me. I care about top level games, not about a title.
I don't see what I should defend. I am not the strongest player of the world and even if I would win that feels empty to me, since that is only because the really strong player don't participate anymore.

Besides defending doesn't quite cut it. There isn't a match for the title. There is just some abstract tournament that is good for 2-6 strong games tops...and a ton of games where the other side doesn't even care to learn. If we would really want to produce top level games, we would held an open qualifying tournament for the Valhöll group, while the world champion isn't qualified for that. The winner of the Valhöll group gets the right to challenge the current world champion - so you only get to become the new one if you actually beat the old one - then there would be something to defend. This way it's only an empty statement. I suggested that back in 2012/13, but nobody was listening.

I don't want to strip the public of their fun tournament, but at the same time for me strong games are much more interesting to watch, so I want to encourage players to play them as much as possible. If one use an open tournament just for qualification purposes, one

10-11/23:47 nath: I don't think no timing at all is the right answer, it's just about finding the right balance for the thing one is trying to archive.

10-11/14:41 OdinHimself: Sign me in!

10-11/14:30 Hagbard: A project group with no timing?

10-11/01:44 nath: In the unlikely event, anyone would be interested on my advice or thoughts on a specific position or a certain part of an (already ended) game, just sent me a notice, I'd gladly try to comment.

10-11/01:34 nath: The real reason I stopped playing here was because I noticed that I wasn't able to deliver halfway decent games with these new tight time limits. I just can't log in every day unless I make very drastic changes to my weekly schedule. I stopped playing after basically half my games timed out for several weeks. I don't care for the rating points I lost (they don't mean anything to me), but from time to time it even destroyed an interesting position.

If I could play games with longer time limit, I'd probably go for it. I'm interested in high quality games, not fast ones.
There are players much stronger than I am. I'm sorry for bringing up the title part at all - I am not interested in titles, but in developing the game and the player strength. But I can't do that alone. Unless we have a notable part of the community that actually wants that (that doesn't mean that they shouldn't care about sport or titles, but it means that they care about the development of strong play), it's in vain.

10-11/00:24 animals: @nath: Those "even worse Hnefatafl-players" could probably do with the benefit of regular competition against high ranking players like yourself. The "World Championship" can only be as good as it's participants.

10-10/22:33 nath: I'd also be particular interested in an AI project for Hnefatafl. I tried a few years ago and failed, but I am always eager to help, provide feedback or join forces with a person, who's eager to produce a strong Hnefatafl Engine.
I'd be also ready to set a small incentive for an AI that truly challenges me even after a few matches.

Contact me if you are interested in any of these.

10-10/22:26 nath: chess balance would be better if you include players that are labeled *strong* according to our Hnefatafl metric. I am a weak chess player, but I am still a lot better at chess and yet I was granted the title of grandmaster (and even worse Hnefatafl-players were).
Correlation doesn't imply causation.

Regarding understanding Hnefatafl we aren't in child stages. We are still in infant stages.
I won't demand something unreasonable as not calling grandmaster (even though that title doesn't mean require understanding the basic concepts), but I am rater asking whether I am the only one that cares about pushing the strategy (and maybe tactics) side of the game. I'd be thrilled about more players being able to challenge me with the white pieces, but I'd have to lie if should name about five people who managed to threaten me the slightest bit with that color and most of these players are the same regarding that. Yet your statistics show an advantage for the white color.
Grandmasters are weak and yet your statistics aren't reliable for even that set of weak players anymore. Don't you want to look behind the scenes and grasp the true meaning of certain openings, moves and strategys?

10-10/11:50 Adam: @Nath, We are all waiting eagerly if you have any new suggestions to all these interesting points. The first two are ongoing and the rate of development depends on the caliber of players active on the site. Regarding the last, I think most agree to allow the long term and ever improving statistics be the judge of game balance. Any of the games considered roughly balanced by the statistics have a balance that is fine enough that new tactics or response to tactics can send the needle back and forth between black and white winning. I think the finest tafl balance has even beaten chess balance, though that may have changed since I last checked.

10-09/21:25 nath: Is there anybody still trying to improve openings? Is there anybody still striving to get a grasp of this game? Is there anybody still looking for the answer whether black or white has the upper hand?

nath
Posts: 65
Joined: Mon Oct 09, 2017 9:34 pm

Re: Nath-project

Post by nath » Sat Oct 28, 2017 6:18 pm

Thank you for pasting it here. Sorry for the late reply - I wasn't much at home recently.

First of all: I don't feel like this *should* be my project. I know well enough that I have neither the time nor the skill to pull that off my own. I am ready to move on if people join me, but unless I get practical help, this will fail.

Second, I don't think this timing is the best possible thing, non the less it's better what we had previously - thank you for the quick action Hagbard!

Besides I love to be proven wrong (since I can learn that way). I'd still appreciate feedback from people who are interested to scale it up. I don't mind die trying, but I can't move much on my own.

Nath

Brench
Posts: 19
Joined: Sat Feb 11, 2017 7:14 am

Re: Nath-project

Post by Brench » Mon Nov 13, 2017 8:14 pm

Nath: I am very interested in any project that involves developing a strong AI player.

If I were to participate, though, I'd probably want to try and do it using particular techniques, particular recurrent neural networks, and not using brute-force tree searches, as the former is my area of expertise. Would such an approach be acceptable, or do you have a particular strategy in mind?

nath
Posts: 65
Joined: Mon Oct 09, 2017 9:34 pm

Re: Nath-project

Post by nath » Wed Nov 15, 2017 1:45 am

I'd never want to limit the ways we explore the game in any way. I don't have the gift of perfect foresight.

I came to think in the past three years that NNs might even be better, since they might be used tell us something qualitative (not just in a quantitative "comparing" sense) about the game, which we don't understand the slightest bit to this point, in contrast to being limited to testing out knowledge we already have. I never seriously tried to apply NNs to Hnefatfl, but I think it's a great idea.

If I'd just be about tree models Jay aka Fishbreath would already have solved it. A naive tree approach without further intelligence works great on a 7x7 board, but sucks using big boards.

Despite a fact that I'd like an AI to help analyzing mistakes, I am even more interested in getting an AI to actually help to explore completely new concepts.

Let's do this!

nath
Posts: 65
Joined: Mon Oct 09, 2017 9:34 pm

Re: Nath-project

Post by nath » Wed Nov 22, 2017 12:02 am

Brench - are you still in? I'd love to hear from you!

Fishbreath
Posts: 24
Joined: Sun Apr 24, 2016 12:13 am

Re: Nath-project

Post by Fishbreath » Tue Jan 30, 2018 10:32 pm

nath wrote:
Wed Nov 15, 2017 1:45 am
If I'd just be about tree models Jay aka Fishbreath would already have solved it. A naive tree approach without further intelligence works great on a 7x7 board, but sucks using big boards.
I'm extremely late to the party here, but I should point out that OpenTafl's tree search is between 100 and 1000 times slower than state-of-the-art chess engines. (OpenTafl does about 60,000 nodes per second on fancy hardware in non-Copenhagen rulesets; Stockfish 8 can do 9 million on consumer hardware and did 80 million in its recent games against AlphaZero.) There's a lot of room left for tree search to teach us things about the game, but OpenTafl's built-in AI is in part handicapped by Java—less than one might think, though—and in part by the particular board representation chosen.

I do want to revisit a neural network/machine-learning tafl AI, but I recently moved and haven't had much time for it lately. I think it is a very promising approach, however, because in a sense it embeds knowledge about tree exploration in each neuron's coefficients, rather than having to search the tree for that knowledge at every move. On the other hand, a given neural network will only ever be able to play a single tafl variation with any reasonable guarantee at success, and to become the best chess engine, AlphaZero spent at least 400,000 teraflop-hours of computing time, and possibly twice that. (On a single AMD Vega, that's at least 3.8 years of computation time.) I don't have those kinds of resources.

nath
Posts: 65
Joined: Mon Oct 09, 2017 9:34 pm

Re: Nath-project

Post by nath » Mon Nov 19, 2018 1:22 pm

If anybody thinks about revisiting this. Recent results of Leela Chess Zero show that a lot of this can be done with less resources. I think that these neronal stuff work much better with Hnefatafl than with chess, because we have an incredibly simple way of representing the features of the position that tell much more about the position than we have in chess. It's much harder to hand craft an eval function (the thing I always failed at), but the representation is much more natural than it's done with chess.

If anybody wants to work on this, I'd gladly help. In particular I have a wacky but fast movegen sitting at my desk. The prototype I built a few years back gets to a speed of nearly a quarter of a million nodes per second in my old hardware. The tree is much bigger than it is in chess, so I highly doubt visiting 9 kk nodes is the way to go. I can totally view the deep learning thing as a working one, yet I don't have the time to do that on my own.

If we could get this to work it might spark my interested for the game again, since it could enable totally new ways of thinking regarding the game. This might be just my personal opinion:
While some new opening variants are tried by different players all the time, I see no radical new ideas established anymore. I think even if the engine can't defeat humans, due some specific problems, it could still contribute ideas.

unhandyandy
Posts: 39
Joined: Fri Dec 28, 2018 6:39 pm

Re: Nath-project

Post by unhandyandy » Sat Jan 12, 2019 7:43 pm

"I have a wacky but fast movegen sitting at my desk. "

Are you a hardware guy? Is the movegen specifically designed for hnefatafl? That would be Awesome.

nath
Posts: 65
Joined: Mon Oct 09, 2017 9:34 pm

Re: Nath-project

Post by nath » Mon Jan 14, 2019 7:21 am

I'm not a hardware guy. I hate hardware. Hardware mostly fails without decent debugging options and quite often because of time-races...

That being said, I have at least some basic understanding how computers work. My movegen (written in C) isn't incredibly well optimized, but reasonable fast.

Yes the movegen is specifically for 11x11 hnefatafl (king armed, corners/empty throne hostile). I wrote it a few years back and have two or three things things that definitely cost performance, but I never got around to fix that, because I never managed to do anything reasonable with that anyways.

If you are interested in a project, please message me about what you need. If there's a need for it, I can picture myself optimizing it further.

unhandyandy
Posts: 39
Joined: Fri Dec 28, 2018 6:39 pm

Re: Nath-project

Post by unhandyandy » Sat Jan 19, 2019 8:19 pm

That sounds like fun, unfortunately I'm just a hobbyist programmer, I've avoided C like the plague. I'm working on a javascript webapp to play brandubh. To compensate for the slowness of the platform (<2000 nodes per minute) I'm trying to write a decent position evaluator. I think I may be able to get it to play above the beginner level, we'll see.

Post Reply