| 1 |
Arimaa / Say Hello / Hello from Fyndegil |
Jan 24th, 2021, 12:50pm |
| Started by Fyndegil | Last post by Fyndegil |
| Hello. I discovered Arimaa and played a few games several years ago, but didn't get into it very much and my involvement in it started to fade away. Then a couple of months ago I heard some colleagues talk about The Queen's Gambit, which reminded me of chess, which reminded me of Arimaa, and I decided I would like to get better at it and introduce myself to the community. |
 Reply  Quote  Notify of replies
|
| 2 |
Arimaa / General Discussion / Re: Bug in Arimaa Game Client? |
Jan 17th, 2021, 11:27pm |
| Started by dpalmer | Last post by omar |
Looks like we didn't have anyone taking up my challenge to find the bug in the code. I went ahead and fixed it. If your client version (shown in the top left) is 0.7.25b then it should now be fixed. If you need to load the newest version type Ctrl-Shift-R in the game client window.
To see the code changes for the fix, open the game client, type Ctrl-U to view the source. Click on the link for arimaa.js and search for 20210117. It was basically adding one letter 'g' 
I also fixed the sounds not working. Just needed to update the soundManager version.
|
|
Reply Quote Notify of replies
|
| 4 |
Arimaa / Off Topic Discussion / Re: Christian Freeling on inventing games (part 2) |
Jan 10th, 2021, 4:00am |
| Started by christianF | Last post by christianF |
So I've not been here since May 2020. Storisende, my intended last game, is from April 2018 and I managed almost two years to not invent another game. Quite commendable I would say, but alas, all good intentions must pass.
I may be excused for Chess+ (Nov. 2019), a cooperative effort with Nick Bentley and not really an invention.
But in July 2020 a casual remark by Stephen Tavener (of AiAi) led to a Loca, a Draughts funny based on a hybrid king/man piece. It's a capture roller coaster towards and endgame. Collateral damage of course, but playable with generic material, so why not. But it was the start of a new wave and before the end of the year there was:
Qascade
King's Castle
XiaGo
DropZone
WedgeLock and
Migong, a cooperative effort with Luis Bolaños Mures.
It went so fast that none of these have an app yet, but they all will have one shortly.
The first one wil be of Cannons & Bullets, my first 2021 game.
It's a mixed bag but it seems I'm still alive  .
|
|
Reply Quote Notify of replies
|
| 9 |
Arimaa / Bot Development / Re: Sharp 2015 available for download |
Nov 7th, 2020, 7:42am |
| Started by lightvector | Last post by aaaa |
| It doesn't appear that Sharp is useful as an analysis tool when in AEI mode, as it will not show its thinking while it's searching. It would be nice if everything it would have shown in "analyze" mode would then still show up as "info" messages. That way, Sharp could be communicated with in a standard way by other programs, like GUIs. |
|
Reply Quote Notify of replies
|
| 10 |
Arimaa / General Discussion / Re: Adapting WHR to Material Handicaps |
Oct 27th, 2020, 7:36pm |
| Started by bestaludulo | Last post by bestaludulo |
I guess it's good to know that the code doesn't really add much that the paper doesn't. I have read the paper, I suppose I will reread the paper and see if I can understand more this time around.
That's essentially what they did on [OGS](online-go.com): they have glicko-2 in the background, and found a best-fit curve to relate the glicko-2 ratings with kyu-dan rankings. The inelegance of this bothers me somewhat, but if it's stupid and it works, it's not stupid.
I did knock around in my head the idea that the function mapping handicaps to WHR adjustment would have to be constrained in some way; I'm glad to see you thinking the same thing. The OGS formula is of the form `a - {[ln(WHR/b)] * c}`. Perhaps if we assume that this form is correct, varying a, b, and c would be feasible? But now you're trying to find a point in 3d space. Or a point in a 3d solid in 4d space, since WHR also gets an axis. I suppose we can drop `a` since it's only there to give kyu-dan numbers, so that's only two variables plus WHR (probably the midpoint between the two players WHR ratings). Then one is finding a minimum on the surface for different values of `a` and `b`. Is there any reason to assume that there would be only one minimum, and that there would exist some computationally efficient algorithm for finding it quickly? If the surface had some bumps in it, then guessing a minima outside of the bump could result in estimates approaching infinity for a and b, going ever further away from the actual minimum... Of course, this assumes we know this surface, which we don't. Wait, but don't we? If, instead of treating the expression as only one side of an equation, we take it as is and solve for WHR, we have `f(b, c, r) = b * e^(r/c)` where r is the rating adjustment to estimate the effective strength after handicap of the weaker player, I think. But this formulation just makes it clear that it is actually a point in 4d space. But it can (I assume) be solved as is, but my brain is starting to misfire trying to grok what good that does. It's like the feeling when you're reading out a move and you're at the point where you can't quite read further without losing track of the moves you've already read.
I guess the end goal is that each handicap has a function `f(x, y)` such that if you took some cross-section defined by some constant `y` value, `g(x)` would be the probability distribution of the given handicap being worth different `z` rating points, if the midpoint rating of the two players is `y`. Thus it's only a surface in 3d space. Now I'm getting two contradictory answers via two different lines of reasoning: this bodes well for me being anywhere near correct. Ahh. The previous reasoning was also trying to solve for two constants, whereas this reasoning implicitly assumes that any constants are already solved for.
Regarding finding the relative strengths of 800+ handicaps simultaneously, what about only allowing the system to recommend handicaps with a standard deviation less than or equal to a semi-arbitrarily chosen number. At first, when asked for a recommended handicap between two given players of a certain rating, the system would return some message such as "No appropriate handicaps found.". The players might then decide to play a handicap they feel is about right and enter it into the system. After a bit of this, the standard deviations of the popular handicaps will fall below the threshold and the system will begin recommending them. Perhaps the threshold could also be eventually raised if the system gets enough data eventually.
I do agree that the vast majority of handicaps aren't really worth bothering with; but it feels arbitrary to just unilaterally choose some to offer. If I wasn't scared by how big a number 864^2 was, I'd like to expand it to both the weaker and stronger player sacrificing 0 or more pieces, so that handicaps like "M for a H" could be used. But 746,496 is, ummm, a big number. The only way it could be plausible is with a very large number of games to train the system, or with a very good way to seed the relative strengths of the handicaps. |
|
Reply Quote Notify of replies
|
Return to the board index.
|
Home
Help
Search
Members
Login
Register