Re: [gnugo-devel] 9x9 opening stats against humans

[Douglas Ridgway]

On Mon, 23 Aug 2004, Gunnar Farnebäck wrote:

> Arend wrote:
> > Douglas wrote:
> > > 5-3 is currently excluded by the "never play
> > > unpopular moves" rule in the fuseki code. Perhaps it makes sense to get
> > > rid of that and throw 5-3 in with the reasonable set. Below is a patch
> > > which does this.
> > 
> > I think in general we still need to drop these unpopular moves. The
> > reason is simply that the database as is way too bad IMNSHO, there are
> > some really bad moves in there.
> 
> Agreed, there were too many oddball games included for all moves to be
> useful.

I suppose that I have to also agree that unpopular moves deserve extra
scrutiny, but not putting them in the db at all seems better than skipping
them during pattern matching. If nothing else, there's no need to waste
time and memory on a pattern which is never used. (Of course, that's not 
what my patch did...)

> > I think the
> > database should be rebuilt using a better set of games (i.e. played 
> > y higher-ranking players).
> 
> That would be best, at least if we can collect sufficiently many high
> quality games.

I have a large set of games collected by Ashley Feniello that I hope to do
this with. It includes NNGS games 1995-05/2004, and so may be a superset
of the games Mans used to compile the existing db. It has about 480k games
total, including 38k 9x9 games. I haven't done anything with it yet,
though.

> > Of course it would make _a lot_ of sense to combine this rebuilding with
> > your analysis.
> > 
> > But for 9x9 openings I agree we could do away with it
> 
> I think a simpler and more robust solution is to disable that test for
> the first two moves. Those patterns should be few and easy to manually
> check for strange moves.

Here's a replacement patch for fuseki.c, but I haven't gone looking for 
the strange moves yet.

doug.

Index: engine/fuseki.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/fuseki.c,v
retrieving revision 1.22
diff -u -r1.22 fuseki.c
--- engine/fuseki.c     24 Jan 2004 04:04:56 -0000      1.22
+++ engine/fuseki.c     23 Aug 2004 21:54:19 -0000
@@ -300,11 +300,12 @@
     return 0;

   /* Choose randomly with respect to relative weights for matched moves. 
*/
-  /* Do not choose moves with less value than 20% of the best move */
+  /* Do not choose moves with less value than 20% of the best move
+     if there are more than two moves on board */
   best_fuseki_value = fuseki_value[0];
   q = gg_rand() % fuseki_total_value;
   for (k = 0; k < num_fuseki_moves; k++) {
-    if (fuseki_value[k] < (best_fuseki_value / 5))
+    if (stones_on_board(BLACK | WHITE) > 2 && fuseki_value[k] < 
(best_fuseki_vlue / 5))
       break;
     q -= fuseki_value[k];
     if (q < 0)