Hacker News Reader

Comment by mrec | Hacker News Reader

It's an interesting game, mostly because unlike most modern games it's massively asymmetrical. Goals, strategies and attitudes are very different for the two players, and despite the numerical lopsidedness in pieces it's much easier to win as the defender, and a slow cautious style is much more important for the attacker.

See Fetlar Hnefatafl for a modern codification of the (very vague) historical rules:

http://tafl.cyningstan.com/page/88/fetlar-hnefatafl

View on HN

I actually wrote a toy app recently following the Fetlar rules as a React exercise:

https://htafl.herokuapp.com/

Not sure how it renders across machines but it looks ok on my Mac.

Yeah, I did the same (VanillaJS) a couple of years back, and started thinking about implementing a computer opponent, but it's really not clear what the evaluation function should be.

You need only codify the victory and loss conditions to build an AI using standard Reinforcement learning algorithms. Feel free to message me if you want more detail on how this would work. We should be able to hammer something out in less than a day.

Thanks for the pointer, but I'm afraid I have zero knowledge of ML and wouldn't be any help.

I have to say I'm a bit sceptical that this approach can work purely based on win/loss conditions - an attacker victory in particular can often take hundreds of moves in Tafl games.

Agreed. The only thing I could think to do would be look for pieces to capture in any given turn, but that would never get it to an end game condition. It's beyond my current capabilities to make a proper computer opponent, but it would be a good exercise.

> look for pieces to capture in any given turn

Not as a general rule. If you're playing as the attacker, you don't always want to take defender pieces even if they're "free" - they're more useful boxing their own king in.

Yea, it's not even a good strategy really. Just the only thing that I felt like I could code that would be a step up from just making a random legal move.

If you could add a rooms feature and online game like http://tafl.herokuapp.com/ then I think you'd see some real use of that app.

The one I linked looks much worse than your board.

I would like to flesh it out at some point. Not sure how to go about doing the back end though. I'm currently taking a Node course so maybe after that I'll give it a shot.

Or release the source. I'm good at backend but only a hobbyist with node. Perhaps I could do something with your ready-made frontend code.

Sounds a bit like Thud in which each player has to take a turn playing each side. See https://wiki.lspace.org/mediawiki/Thud.

It does sound a bit like Thud, doesn't it? From your link:

Thud is a board game based on the earlier dwarfish game of Hnaflbaflsniflwhifltafl. It is played mainly by Dwarfs and Trolls... The game itself echos the game played by Vikings and Anglo-Saxons "Hnafletafle"- which literally means "Kings Table". Hnafletafle echos go beyond just the name. Like 'Thud' it is unusual in that it is an 'asymmetrical' game...

In that Thud is a parody of it.

Note that Thud! is a real-world version of the game, that is never precisely described in the books. Er, as far as I remember :)

I've played it a few times with friends and I have to agree the defenders have an easier time (although it certainly feels more intimidating to play the defender). I think defenders heavily benefit from most people's familiarity with chess. It's very easy to distract someone with pieces they can capture, then sneak your king out the side.

Reply on news.ycombinator.com