[gnugo-devel] Improved stabilisation in get_next_move_from_list

[Teun Burgers]

Arend wrote:

> here are two possible stabilization goals for get_next_move_from_list:
> Stability regards to
> 1) orientation
> 2) new pattern matches appearing or disappearing on the list.
> 
> The problem I see with your patch, Teun, is that it gives up 2)

Arend, I don't get this point. Why does comparing bdist(move1) <
bdist(move2)
instead of comparing move1 < move2 give this up? The only difference I
see is that
bdist maps some moves on the same distance.

> and doesn't achieve 1).

I don't claim it does, the patch improves it. Either additional
stabilisation criteria are needed or there is another source of
orientation
dependence. This can either be a bug or a fundamental property of the
owl code.

> I think that 2) is more important (because it reduces owl tuning
> noise);

> while reaching 1) would be nice, I don't think getting closer to
> 1) without actually reaching it is so interesting (because only then could
> an orientation inconsistency be considered a bug that is worth tracking
> down).

Reaching 1) can only be achieved in steps, such as the one I propose
in my patch. In the same way the orientation depencendies in reading.tst
have been removed step by step. In the end btw these all turned out to
be
due to bugs.

And I do think 1) is important. Orientations dependencies in the owl
tests
propagate into e.g. genmove type tests. I am convinced that orientation
independence in the owl tests will make tuning more effective.

> Possible suggestions:
> 1. Make your patch dependent on some #define's.
> 2. Add the move > top_move test back as a fourth condition

I am in favour of avoiding orientation dependent tests. Can't we use
something such as accuratelib(move, color, 0 ,NULL) as additional
criterion?

> 3. We could just keep your patch around and use it whenever testing for
> stability with regards to orientation.


Teun