Re: owl improvements (Re: [gnugo-devel] Owl tuning)

[bump]

I now have timing for Nando's patch removing only A1402. It took
11029 seconds on an otherwise unloaded machine. So those three
patterns may cost almost 1.5 % on the entire regressions.

> One thing is clear in my mind: the patterns for which I proposed a removal
> need to be (at least) a lot more constrained. For instance with A107, it
> seems to me that hoping to kill in following position might prove very
> difficult if O is lacking external support
> 
> ....X
> X.*.X
> ....X
> .....
> -----
> 
> But GG is currently gonna try it whether it has support or not, because it
> will probably find that the * stone is not tactically killable.

My point was that if * does not work there may be little hope of 
killing, so * should be tried. This does not contradict what you
said, but the case for removal of this pattern would then have to
be based on performance.

> In any case, thank you all for the comments and for spending some time on
> this topic.

I propose that we take the patch first only killing one pattern,
A1402, then to treat the removal the other three patterns as a
separate patch. Unless someone objects I'll do this tonight.

On a different subject, GNU Go has a tendency to play this move:

.....
.O...
...X.
..*..
.....
-----

On the second line, that is unorthodox but playable. However it
leads to a position where the owl code sometimes recommends this
attack:

.....
.O...
...X.
..O.*
.....
-----

That is usually a mistake.

Dan