display confidence of win metric
Posted: Fri Jan 18, 2019 3:23 am
The standard presentation these days for move evaluation seems to be G,p were G is Goodness-Winrate-Metric and p is the performed playout numbers. Is there already a way to obtain the standard deviation (or some related property relating to confidence) from the little machine's brain? I assume that a really verbose debug mode could let you recalculate such a value, but do any robots already produce such a value?
I naively assume that it is meaningful and possible to be able to produce such a value, but trust that somebody will rabidly correct me if not. I ask because of the current question of whether 0 and 2 are distinguishable.
I naively assume that it is meaningful and possible to be able to produce such a value, but trust that somebody will rabidly correct me if not. I ask because of the current question of whether 0 and 2 are distinguishable.