Re: [gnugo-devel]

[bump]

I wrote:

> I've been running the regressions with the patch stephane_3_22.1b
> but changing COSMIC_GNUGO to 0. Is this supposed to give the
> same results as the standard CVS? It doesn't ...

Stephane wrote:

> Ah ? It's strange, because the first 3 FAILS seem to be more
> related to the absence of break-in/block-off than to my patch.
> 
> Are you sure you recompiled with the break-in code enabled by
> default when you changed COSMIC_GNUGO to 0 to make the regression ?
> Did you run ./configure --enable-experimental-break-in ?
> 
> I'm pretty sure that the *only* change against the current CVS that is
> not in a COSMIC_GNUGO conditional flag is the EXPLICIT_LOOP_UNROLLING
> which I unadvertantly set to 0 for testing purpose.

You are right, I was inadvertantly running the regressions
in a directory with experimental break-in disabled. After
correcting that, the patch has no effect on regressions
up through lazarus.tst.

I'll finish the run but it seems your assertion is probably
true. In that case maybe we should make a version of the
patch in which cosmic is a configure option disabled by
default.

However ...

> So I tried to tame it down by making it use the cosmic values only
> every other moves... (that's the reason of the ((movenum / 2) % 2)
> test in the patch below).
>
>  And, to my great surprise, it seems to have worked somehow !
>
> In a sense, alternating between the "slightly unreasonable" and
> the "slightly conservative" evaluation functions uses the strong
> points of both :-)

... this ``pair go'' approach will certainly complicate regression 
testing of the patch. (http://www.pairgo.or.jp/home.htm)

Dan