[Bug-gnubg] GNU multi-ply analysis

[Albert Silver]

I had two questions regarding GNU's analysis. The first is probably
obvious, but it made me wonder all the same: as I understand it, GNU has
a heuristic it uses to define which net to use for a certain position.
Of course when analyzing the first plies, it is easy, but what about the
look-aheads? If the positions definition changes a ply or to ahead, does
it use a different net in those future plies? It seems obvious it
probably does, but since this sort of problem has existed with chess
programs (i.e. some plies ahead the middlegame was now an endgame)
nothing is for granted.

My other question was what exactly are the characteristics that it uses
to define which net will be used? In other words, when does it use the
Crashed Net (the Race Net seems rather obvious) and when does it not? I
remember someone, possible Neil, showing a backgame position of some
kind and complaining about GNU's handling of it. I also remember Oystein
adding that this position was in fact being handled by the Contact Net
and not the Crashed Net.

                                        Albert