Testing autoscore algorithms

[Feijoa]

My goal is to automatically determine which stones are dead from a couple of AI estimates that give you dead/live/unsettled. The idea that I took from Vsotvep (though it was different from what he had intended) was to resolve the unsettled ones with an estimate of the ownership of “unsettled area”:

@Vsotvep

Chains of unsettled area (open intersections + unsettled stones) are considered as a whole.

(I think he meant to also include the dead stones in that definition.) I guess I prefer to say “region” since “area” already has a particular meaning in Go.

@gennan

• Is a “boundary” the set of stones of either color touching the area under scrutiny, regardless of their status (dead/alive/unsettled)?

The boundary can only have live stones, since the region expands to include all others. Look at the eggplant region from one of the links above for an example:

@gennan

• Is a “boundary” the set of stones of either color touching the area under scrutiny, regardless of their status (dead/alive/unsettled)? … In that case, wouldn’t a boundary usually consist only of living stones of one color (attacking the unsettled group inside the area)? If that is true, most unsettled stones would be declared dead, right?

I said “not counting stones touching only unsettled stones”. I don’t know about “most”, but in the one above the unsettled group dies, since the whole eggplant boundary consists of black stones, specifically the the seven highlighted here:

(The white stone above is initially part of the boundary, but it gets promoted to unsettled as described. And yes, it’s just direct touching.)

But consider this situation (which actually had a bug that I fixed just now ):

Since we don’t count ones that only touch unsettled stones, there are NO live boundary stones at all. With no black or white stones to count, the region is considered dame and all unsettled stones live.

@gennan

• Does “being in dame” mean unsettled stones touching (or inside of) an area that has living stones of both colors in its boundary?

I call a region dame if it has both colors in its boundary or an empty boundary (like the rectangular six above).

@gennan

I thought you’re counting the black/white stones in a boundary as a tie breaker to determine the status of unsettled stones, but you don’t mention this now (or perhaps I missed it). Is it still the case? If not, why count those stones?

After your earlier comments I decided that the tie-breaker idea was too arbitrary. So all that v3 cares about is whether the “count” is positive. We only assign a winner to the region if the count is positive for one color and zero for the other. Otherwise, everything lives.

I hope that clears it up a bit.

Now, is it like a different ruleset? Since I’m proposing to just use this for life/death and then follow the normal rules of go for doing the actual score counting, I think it gets us closer to the traditional rules than what we have now. And I hope that there aren’t any properly finished games that the algorithm will mark incorrectly. I do think that there are things about Internet Go that are just necessarily going to make it different than real life games, and I’d like to work on making it different in the right way.