[gnugo-devel] Tuning matters

[Daniel Bump]

The mailing list may be down, so I'm cc'ing Gunnar.

I'm adding my patch to the CVS. I'll leave Gunnar
to add the bugfix he found yesterday in response to
the game on the Polish server.

Yesterday I put up a tuning patch. It broke three tests.

After the patch, GNU Go plays L17 in lazarus:16. Locally
L16 is better and maybe this should be fixed by a
replacement pattern. In this test, Trevor's html view
shows the pointless move at L2 valued 15.  This needs
to be investigated.

It breaks nngs:1050 after which W wants to play M5.
The patch increases the strategic defense of a dragon
which is alive but with uncertain status. (Owl nodes
exhausted.) M5 is not the best move to defend this
dragon because after B:N4, W has to defend the bottom
or die, so M5 gets cut off.

In global.tst:9, Black prefers F2 to the correct F4.

These changes seem acceptable. After the patch there
are unexpected PASSes at strategy:27, trevord:1050
as well as nngs1 tests 4, 5, 6, 8, 21, 26, 29. The
tuning is against nngs1, of course. Test 8 needs
the change in uncertain defense reasons.

There are also unexpected passes in reading:168,
atari_atari.tst:15 and connection:72 and a fail
in global:38 but these are due to Gunnar's komaster
patch.

Dan