Re: [gnugo-devel] another semeai patch

[bump]

I wrote:

> After such a change there is the following breakage:
> 
> ./regress.sh . trevora.tst 
> 370 unexpected PASS!
> ./regress.sh . nngs.tst 
> 820 unexpected FAIL: Correct 'J13|L9', got 'D14'
> ./regress.sh . semeai.tst 
> 59 unexpected PASS!
> 60 unexpected FAIL: Correct '0 0 PASS', got '1 1 H5'
> ./regress.sh . strategy4.tst 
> 155 unexpected FAIL: Correct 'D18', got 'E16'
> ./regress.sh . ninestones.tst 
> 560 unexpected PASS!
> ./regress.sh . 9x9.tst 
> 30 unexpected PASS!
> 120 unexpected FAIL: Correct 'F1', got 'B1'

After doubling the semeai nodes to 1000, we get:

./regress.sh . trevora.tst 
370 unexpected PASS!
./regress.sh . semeai.tst 
59 unexpected PASS!
60 unexpected FAIL: Correct '0 0 PASS', got '1 1 H5'
./regress.sh . strategy4.tst 
155 unexpected FAIL: Correct 'D18', got 'E16'
./regress.sh . nngs2.tst 
150 unexpected PASS!
460 unexpected FAIL: Correct 'J4|K2|K3|K4|K5|L2', got 'N15'
./regress.sh . ninestones.tst 
40 unexpected PASS!
560 unexpected PASS!
./regress.sh . 9x9.tst 
30 unexpected PASS!
120 unexpected FAIL: Correct 'F1', got 'B1'
290 unexpected PASS!

(This is still with Paul's patch modified by omitting the test.)

We get a net gain of 3 passes. Note that one of the changes
is nngs:820, where the ``FAIL'' might actually be a good move.

I suppose we could increase the number of semeai nodes but
then we could also think about implementing a persistent
cache for semeai. We could just use the union of the owl
active areas for the two dragons, assuming the owl code
has run on both dragons.

Dan