Thanks for the link perceval, reading!
I think the link may have messed with the page-formatting though...
A vague treatise on influence
-
RobertJasiek
- Judan
- Posts: 6273
- Joined: Tue Apr 27, 2010 8:54 pm
- GD Posts: 0
- Been thanked: 797 times
- Contact:
Re: A vague treatise on influence
Read the diagram captions! From them, you easily infer that influence is not proportional to the number of influence emitting stones.shapenaji wrote: I don't see this as an example of non-linearity of influence,
It fails to model connection, life and territory. What is it that it models at all? Proximity * #stones * sign_of_stone_colour. Not completely useless information, it suffices to roughly identify basic moyos and stone distance (regardless of whether the friendly stones are dead or alive...). That is pretty much what you can get from such a model. Too little! My model gives the same information and much more.As far as the validity of the light model.
You forget that shortest distance to dead stones does not assess particularly useful strength of influence.I think the reason why the light model is a good place to start, is that light operates on a shortest path principle.
Manhattan distance. Regardless of how light travels, my light-independent model does not have that problem of travel nature of influence and of considering light hurdles.The limitation of this approach is that it assumes continuity, rather than computing shortest paths on a grid,
It is a bad start because path (of influence particle travelling) is irrelevant. What matters are connection and life status. Simply speaking, go does not work like "Stone A sees (visible seeing along the movement of a light ray, whether straight or zig-zag) stone B and is therefore alive." but works like, e.g., "Stone A can connect to alive group B is therefore also alive.".But I think that kind of path-ing approach is a good start
-
RobertJasiek
- Judan
- Posts: 6273
- Joined: Tue Apr 27, 2010 8:54 pm
- GD Posts: 0
- Been thanked: 797 times
- Contact:
Re: A vague treatise on influence
Correct it, then consider a particular intersection's influence and tell me what such an influence map says about black / white connection / life / territory.perceval wrote:your examples Robert, the influence map can be quite easily corrected
How do you want the tweaking to be defined?by tweaking the diffused influence if the stones are alive or dead
So that such influence spreads also beyond the surrounding of one player's independently alive stones? Failure! You need more tweaking...collaborative diffusion would take care of not diffusing the influence of a stone blocked by some others.
-
perceval
- Lives in gote
- Posts: 312
- Joined: Thu Aug 05, 2010 3:35 am
- Rank: 7K KGS
- GD Posts: 0
- KGS: tictac
- Has thanked: 52 times
- Been thanked: 41 times
Re: A vague treatise on influence
i am quite mad because i lost a long post because i was not logged. 
Here is what i want ot talk about:
perceval definition of the influence function:
a function B(x,y) on the coordinate of the go board such as:
you can disagree with that definition if you want but that is what i am going to talk about, it seems an interesting object as being able to compute it for all posistion would solve the game.
As i do not have a god complex i just want to throw some ideas to compute an approximation of this function.
i agree with you that the influence obtained from a given shape should nto depend on what is on the inside (excpet for life ar death status). For example: should have the same influence than: This is what you mean by non linearity of influence (i think)
but more influence than the dead shape: My point is thqt the algo in the link above take all this into account:
What you do is to define source for the function, here sources are the intersection where there is a B or W stones.
you define the value of B at those source and the algo compute the function on the whole board by spreading from those sources.
The algo has good properties for the issues above:
it "blocks" influence from inner stone from spilling out of a closed shape.
in addition, the "easy tweak" take live and death status is to seed the algo with value other than B=1 for stones that are not alive, and to go close to zero for dead black stones.
Here is what i want ot talk about:
perceval definition of the influence function:
a function B(x,y) on the coordinate of the go board such as:
- B(x,y)=1 if the intersection if an indisputable point for B
- B(x,y)=0 if the intersection if an indisputable point for W
- The sum of B(x,y) on the whole board gives the score difference at the end of the game
you can disagree with that definition if you want but that is what i am going to talk about, it seems an interesting object as being able to compute it for all posistion would solve the game.
As i do not have a god complex i just want to throw some ideas to compute an approximation of this function.
i agree with you that the influence obtained from a given shape should nto depend on what is on the inside (excpet for life ar death status). For example: should have the same influence than: This is what you mean by non linearity of influence (i think)
but more influence than the dead shape: My point is thqt the algo in the link above take all this into account:
What you do is to define source for the function, here sources are the intersection where there is a B or W stones.
you define the value of B at those source and the algo compute the function on the whole board by spreading from those sources.
The algo has good properties for the issues above:
it "blocks" influence from inner stone from spilling out of a closed shape.
in addition, the "easy tweak" take live and death status is to seed the algo with value other than B=1 for stones that are not alive, and to go close to zero for dead black stones.
In theory, there is no difference between theory and practice. In practice, there is.
-
RobertJasiek
- Judan
- Posts: 6273
- Joined: Tue Apr 27, 2010 8:54 pm
- GD Posts: 0
- Been thanked: 797 times
- Contact:
Re: A vague treatise on influence
Whatever you define "indisputable" to be, such is too simplifying because it does not distinguish the intersection's degrees and in particular does not distinguish Black's connection status from White's connection status from Black's life status from White's life status from Black's territory potential from White's territory potential. By ignoring degrees and putting the three aspects in the same bowl, you only get a too rough approximation of the kind yes or no for "Black has non-negative and White has negative connection status there AND Black has non-negative and White has negative life status there AND Black has but White has not territory there". IOW, all that you get is a special case of my model.perceval wrote: a function B(x,y) on the coordinate of the go board such as:
- B(x,y)=1 if the intersection if an indisputable point for B
- B(x,y)=0 if the intersection if an indisputable point for W
Wrong. Influence changes dynamically (at least) until the game end! You only get the score difference at the game end if the game end is already reached. For that, simpler models than yours suffice.The sum of B(x,y) on the whole board gives the score difference at the end of the game[/list]
Who cares? Influence is relevant DURING the game and not for predicting which intersections score for whom at the game end. If you want that, then a territory definition similar to n-connected is useful: how often a player can pass while an intersection is still his territory.This can be losely interpreted as the probability that a given intersection will be B or W at the end of the game
Use my model and you get such information implicitly.but this interpretation misses the case of a dead stone with some aji that will help make some point elsewere even if its dead.
The point is not that one would need to disagree with the definition but that it provides by far too little information!you can disagree with that definition
No. Too little information.if you want but that is what i am going to talk about, it seems an interesting object as being able to compute it for all posistion would solve the game.
You are too pessimistic. As a special case of my model along the thoughts outlined above, calculation is reasonably possible for human beings. Only PCs can have greater problems because they are in greater need of reducing computational complexity related to proof-play game tree exploration.As i do not have a god complex i just want to throw some ideas to compute an approximation of this function.
No. Not the same. One has a *-alive group. The other has a 2-alive group. For practical purposes, this is almost the same. Almost but not exactly.should have the same influence than:
This is insufficient since some influence can spread from inside to outside.The algo has good properties for the issues above:
it "blocks" influence from inner stone from spilling out of a closed shape.