[Bug-gnubg] Measuring performance levels 2

[Moore, Dave]

> 
> Why would 2ply play worse? Why would it get gammoned for cristsake? 
> When I play it via my script I never see a gammon. Then I 
> realized my script
> uses my own best move function, and the rollout uses GNUbg one. 
> Now I, for 2ply, filter only using 0ply. GNU filters using 
> 1ply as well on the
> way. My bet is that 1ply throws out the good moves, and 2ply 
> is forced to play
> horror moves. If this is the case I think GNUbg should follow 
> suit on what I am
> doing for a long time now.

Just a lurker throwing in my two cents, but this makes sense to me.  If
evaluating the position from the other side of the board (1-ply) causes
discontinuities due to the various nets & evaulations, then a 2-ply
evaluation should not trust any intermediate (1-ply) results for pruning
moves.

It seems to me that any of the odd ply evaluations/analyses are hard to
interpret until there is a way of normalizing the views from both sides of
the board.

Regards,
Dave