Re: [gnugo-devel] another semeai patch

[SP lee]

> SP Lee wrote:
> > I'm using cygwin and it compiles and runs ok.
>
> It may or may not work depending on the platform. I.e. it's not portable.
>
> > However, when I changed it
> > into strcmp like
> > if (strcmp(owl_moves[k].name, "eat lunch") == 0 && ......
> > I got a run time error although it compiled ok. So at this moment I
don't
> > have other solution. Any suggestion to avoid string comparison is
welcome.
> >
> > I do have added a constraints to this patch to eliminate the unexpected
> > fails. After this new patch there is only one pass at semeai:58 and no
other
> > breakage.
>
> Well, this is not the right place to filter out the move. It should be
> done already when the lunch eating moves are first generated. I'd also
> like to have some reasonably robust criteria for when eating a lunch
> is a bad idea in a semeai.
>
> /Gunnar
>

I have revised this patch again. Now the modification goes inside the
function estimate_lunch_eye_value. I think it's reasonable to assume the
max_eye a lunch of size 2 can make is 1. I'm not sure what was behind the
original idea of max_eye = 2.

$ cvs diff -u owl.c
Index: owl.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/owl.c,v
retrieving revision 1.188
diff -u -r1.188 owl.c
--- owl.c       14 Jan 2004 22:21:27 -0000      1.188
+++ owl.c       20 Jan 2004 04:10:55 -0000
@@ -5344,7 +5344,7 @@
     else {
       *min = 0;
       *probable = 1;
-      *max = 2;
+      *max = 1;
     }
   }
   else if (size == 1) {

The regression breakage is not much.

nngs:490 FAIL; correct J18 got G18. Although J18 is better, now gnugo
chooses G18 because it also can owl_attack H18 (before this patch G18 can't
do that). There is no obvious error in the reasoning.

trevorc:1080 PASS
semeai:58 PASS

strategy4:167 FAIL correct D4 got E5. The E5 is indeed a bad move, but it
seems to be not directly linked to the patch above.

SP Lee