[gnugo-devel] Re: 3.3.6 vs 3.2 results

[Teun Burgers]

address@hidden wrote:

> On the other hand Trevor thinks gnugo vs. gnugo tuning is useful. He looks for
> games with big losses. (You said there were two such games.) Those games will
> probably contain one or two extremely bad errors which might be worth
> isolating.

I think this is an effective approach. See
these two blunders in game test084 for instance

# black omits an important connection
loadsgf test084.sgf 135
1 gg_genmove black
#? 1 [H8|H9]*

# interesting blunder!
loadsgf test084.sgf 175
2 gg_genmove black
#? 2 [H12]*