Recommend
12 
 Thumb up
 Hide
93 Posts
Prev «  1 , 2 , 3 , 4  Next »   | 

Abstract Games» Forums » General

Subject: Starweb rss

Your Tags: Add tags
Popular Tags: [View All]
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb

This is the position after white-22 (comments in the game).

Black has managed to cut off D1 and threatens to cut off the entire bottom group. The group's value is a 'mere' 3 points but it's defensive value may be more important. Under an approaching white group black's lower left has some vulnerabilities. One of them allowed me to play G9 in the first place.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Michael Howe
United States
Cromwell
Connecticut
flag msg tools
As I see it, black is likely to make a g(7) + g(2) for 31 points. White is likely to make g(6) + g(2) + g(1) for 25 points. Under n^2 scoring, black still wins 53 to 41. Under "single largest group" scoring or "fewest number of groups" scoring, black still wins. I could be wrong, obviously, since I'm not an expert Hex player. I could in fact end up being very wrong.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mhowe wrote:
As I see it, black is likely to make a g(7) + g(2) for 31 points. White is likely to make g(6) + g(2) + g(1) for 25 points. Under n^2 scoring, black still wins 53 to 41. Under "single largest group" scoring or "fewest number of groups" scoring, black still wins. I could be wrong, obviously, since I'm not an expert Hex player. I could in fact end up being very wrong.
I'm not particularly fond of the 'largest group' or 'smallest number of groups' object and triangular scoring doesn't differ all that much from square. For triangular the 'first stone counts as 1 point, the second as 2 points, the third as 3 and so on' bridge provides a nice handle, and for future seasoned players (as I surely hope there will be) it will be a non-issue.

As for the game, my strategy has proved more naive than Ed's and I actually realised too late the natural emergence of purely defensive groups that contain no corners at all. That was of course stupid, stupid, stupid, but it's the kind of thing that sometimes happens to me as a player.

The board has some convenient attacking points for such groups:


The black stone can cut off a stone at either red point. The stone that is one diagonal step away from it towards the centre can cut either way too and even easier get away with that. It is obvious that such vulnarabilities should be considered early on!

As for the game, it did indeed prove you wrong. Ed's got the hang of it, obviously, and decided to attempt a deep cut at Q12.


I'm not yet sure how to handle it. I do feel I'm running behind the facts though.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Michael Howe
United States
Cromwell
Connecticut
flag msg tools
After a lot of initial discussion about hos important the center would be, so far it has played relative little role. But I think it is important for groups to be able to reach the center as another way to connect up to defeat deep cuts. Interesting.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mhowe wrote:
After a lot of initial discussion about hos important the center would be, so far it has played relative little role. But I think it is important for groups to be able to reach the center as another way to connect up to defeat deep cuts. Interesting.
In the last diagram I next played H11 because if no local reaction would follow, then G13 could cut and escape with it (narrowly). But Ed spotted the danger. Now his solid 28 points group will likely be decisive.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
Position after white-31 (comments in the game).
For interested posters it may be clear that I have lost this one. I woke up too late to the fact that claiming corners must immediately be followed by guarding against the impact of cuts. Almost unavoidable cuts for that matter.
So after having more or less secured his main group, Ed started chopping my position to pieces with nice illustrative tactics.

Now Black's 28-points group on the right seems unassailable and even if I can tie some groups together in the centre, I can't get them all.


So I'll fight to maximise the score and see how it pans out.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
Guess what, Cameron Browne sent me a Starweb AI by return! I can't begin to say how much I appreciate it. It features a small board and a regular one. Here's a game I narrowly won:


The picture is half the actual size (linear). The AI, according to Cameron, is not very strong, but it's certainly a great learning tool. If I get Cameron's permission I'd gladly share it with you. I presume it's Mac only (but I'm not sure).
Edit: It's java so it runs on windows too.

Edit 2:
Here's a large game, I'm getting the hang of it



Edit 3:
One more to get an idea of how a game might develop. This one was played with 16 seconds thinking time per move for the AI.


Towards the end the program makes a lot of sente moves trying to cut double connections (for lack of better). A definite score can of course only be displayed if all connections are closed. Between humans a game will most likely end in a much more 'open' position, if all is clear.

Edit 4:
Here's a reply I got from Cameron that concerns everyone interested in the game:
Quote:
I’m glad you like it! Yes, feel free to pass it on to whoever you want.

The AI uses a (reasonably) new algorithm called Monte Carlo Tree Search (MCTS) that chooses moves based on thousands of random simulations. The benefit is that it doesn’t need to know what constitutes a good or bad position, this is learnt from the win/loss results of the simulations, i.e. it essentially makes those moves that lead to wins more often.

This makes the algorithm independent of the game, so I can can just reuse the same MCTS code for each new game, which saves a lot of time. And it can be surprisingly good! I actually find the Starweb app hard to beat if I give it first move.

By the way, I found the game on the small board also quite interesting, maybe a good way to introduce players to it.
So whoever wants the program can mail me at 'christian at mindsports.nl' and I'll gladly forward it.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Michael Howe
United States
Cromwell
Connecticut
flag msg tools
Christian, is it too soon to tell, or can you tell us which you think is the better game, Havannah, or Starweb?
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mhowe wrote:
Christian, is it too soon to tell, or can you tell us which you think is the better game, Havannah, or Starweb?
I'm sceptical about the weight my opinion would carry. As you know I invent all the wrong way. Where's the struggle, the modifying, the tinkering, the 'finetuning'? And then, four lines of rules! (That's about the same as the number of thumbs ups it got).

But I can tell you that (in the lightness of my opinion) it is the game *Star wanted to be! laugh
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
I've tried to forward Cameron's message (including the Starweb AI) to several posters, but they bounced (in the case of Ton van der Valk, Martin Grider and Luis) because

"This message was blocked because its content presents a potential security issue"

All Gmail accounts.

I'll try and see how I can get around that. Please let me know how you were informed at your end.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
Armed with some fresh strategic insight, I let the AI go first (there's no swap) and gave it the maximal 64 seconds per move. After 56 moves of both the board looked like this - more or less the point where humans would consider the game decided:


Black has no corners on the bottom edge because I declined in favour of move 18 (second point from the bottom of the NS diagonal) that gave cutting options in the middle of the white corners. So it isn't 9-9 but 10-8.

White has a 3-group, a 2-group and 5 isolated corners, totalling 14 points.
Black has a 4-group, a 3-group and 1 isolated corner, totalling 17 points.

The reason I upload the image is that it illustrates the importance of the centre.

1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb

I got fed up with Gmail. You can download Cameron's Starweb AI from the mindsports homepage.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Michael Howe
United States
Cromwell
Connecticut
flag msg tools
So I think an interesting question then becomes: can a human expert beat another human expert by only taking 8 corners? Do you think it would be a viable strategy between top players?
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mhowe wrote:
So I think an interesting question then becomes: can a human expert beat another human expert by only taking 8 corners? Do you think it would be a viable strategy between top players?
I do, actually. The board is a base-7 hexhex with six 15-cells concavities protruding from it. The total number of cells thus equals a regular base-9 hexhex. That makes the Starweb centre smaller relative to the total area.

The shape of the board puts six corners nearer to it so these are natural cells to take first. If you take a look at the cells that are in the middle of the edges of the base-7 centre, the open circle in this image ...


... you see that these are anchor points for cuts in three different directions. Occupying such a cell in between opponent's corners on both sides, early on, is a quite formidable weapon. As the second player it will cost you a corner, as the first player that may not even be the case (though it may cost you initiative). The cut should of course be efficient. If you cut a potential 7-group, then 4/3 is better than 5/2 and far better than 6/1. The achor stone also gives immediate access to the centre and that's where it all comes down, eventually.
So I think it's a viable strategy indeed.

Edit:
It's also noteworthy that two well known objects on a hexboard - 'creating the largest group' and 'achieving the smallest number of groups' - here both play a role as a 'handle' or an 'inclination'. Like yeah, that can't be bad. But neither guarantees a win.

I'm pleased by the fact that the interest in the game comes mainly from posters who know their way around the hex grid.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mhowe wrote:
So I think an interesting question then becomes: can a human expert beat another human expert by only taking 8 corners? Do you think it would be a viable strategy between top players?
I've put it to the test under the most favourable conditions for the AI: small board, first player (white) and the maximum possible thinking time (or 'trying time' actually) of 64 seconds. To be fair, I lost the first two games by 2 or 3 points, but I won the third one and it is quite illustrative of the idea.

Here's the position after 10 moves for both:


I declined a corner with black-18, using 16 as the anchor. AI took the corner (19) and I blocked an isolated corner with 20 (at the same time connecting 4 and 10).
This is how the strategy panned out (and mind, the AI isn't so weak under these conditions):


White still has a 4-points connection at the bottom (between 5 and 19) but that's it.
(Edit: actually it's a 3-points connection because the 1-group disappears)

Ten to eight corners but eight wins with 4 points. Viable strategy!
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Michael Howe
United States
Cromwell
Connecticut
flag msg tools
That's very encouraging. Good games support multiple strategic approaches and it looks like this is true of Starweb.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mhowe wrote:
That's very encouraging. Good games support multiple strategic approaches and it looks like this is true of Starweb.
I appreciate your trust. It is very hard for an inventor to make such a claim because of the eternal and rather unavoidable suspicion of bias, the argument being that inventors are all the same, and though some are more so than others, we're all supposed to have an agenda other than the game itself. And if not ... well that's an awkward position, I can tell you.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
As Clyring pointed out at the Arimaa Off Topic Forum, the net increase in value obtained by connecting two groups with values 'm' and 'n', equals m*n. That's very convenient.

I have further explored 'minority strategy' because it's an important indicator of the relation between strategy and tactics. I've put the AI on 64 seconds per move on the large board, and let it go first.


With 18 I declined the last corner and AI took it. Then I cut through the middle of the white position with 20. Here is how it panned out:


The AI insisted on a number of unnecessary cutting attempts in the endgame, but that actually makes the situation clearer. Black gets another 5 points by connecting stone 12, White adds another point by connecting stone 7.
So minority strategy won 23-12.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Luis Bolaños Mures
Spain
flag msg tools
designer
Avatar
mbmbmbmbmb
Here's another challenge: win without having the highest-scoring group.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
luigi87 wrote:
Here's another challenge: win without having the highest-scoring group.
Or win without having the smallest number of groups.
But I'm about to try if I can stretch 'minority strategy' one step further.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Eugene Panferov
msg tools
designer
is there an online implementation of the game?
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
silly_sad wrote:
is there an online implementation of the game?
Not yet, we're going to put it in the ArenA and the Players' Section at mindsports.nl. later on, but I'm in no hurry.
At mindsports you will need java to play, just as for Cameron's program (you can download it from the homepage).
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Eugene Panferov
msg tools
designer
christianF wrote:
silly_sad wrote:
is there an online implementation of the game?
you will need java to play

which is equal to "no, and will never be playable"
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb

I've played around with Cameron's program for a while now and I hope to show here what I actually did see right from the start, that Starweb is a strategy game, as opposed to a tactical one. It's not surprising that members who so far have shown some real interest in the game are all very familiar with the peculiarities of the hex grid. In my case the source of this intuition is that the building blocks of the game and the way they interact are very similar to Havannah.

Now instead of deviating to a popular subject like the difference between a strategy game and a tactical one, I'm going to show you strategy.

Michael Howe already pointed out the significance of 'minority strategy' in Starweb. Minority strategy means that you sacrifice a corner to get more influence in the center and to create cutting options in the opponent's position. Cameron's program certainly benefits from the smaller board, the more so if it's given maximum 'trying time', so I played a fair number of games trying to stretch minority strategy beyond its sensible limits by sacrificing two corners.
I lost a couple of games and quickly learned some of the program's shrewd cutting tactics, and how to guard against them. So here I have a very clear example of how to go about it ... against an AI. I'm sure sacrificing two corners is not a good strategy against a clever human player. For that we have to scale back a bit: sacrificing one corner has all the hallmarks of a lasting strategy.

So here we are. I tried to secure a large group (the first 3 stones of both lent themselves for that) while AI was grabbing corners. 15, 17 and 21 secured cutting points in AI's position and 19 and 23 provisionally secured the 7-group.


It's a position that clearly illustrates my plan: a 7-group counts 28 points, while two 4-groups and a 3-group make it to 26. My two plans are: secure the big one and prevent AI from connecting anything beyond the 4,4,3 division.


This is how it panned out. Note that 27 can connect in two ways and that I still have a cutting option from 49 down. But AI can close it and it doesn't matter much because I win with 2 points.
As said, don't try a 7/11 division against a strong human player, and come to think of it, not against a strong AI either.

3 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Michael Howe
United States
Cromwell
Connecticut
flag msg tools
Wow, interesting! I wonder if 52 and 53 were reversed, and then black responded in the upper center, maybe two hexes to the right of 1, would black have had a chance to connect to 8 and make a 5-group?

1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Prev «  1 , 2 , 3 , 4  Next »   |