Quote:
Originally Posted by TheRealFatboy
that's what i was thinking, early blunders can be overcame with good play/dice. late blunders lets say when bearing off with a 44 moving checkers from the 6/2 5/1 instead of pealing off 4 in a no contact situation is a huge blunder that is unlikely to ever be offset by good anything, i suppose 22 & 11 rolled would help as the higher points might be stripped causing the force pealing of checkers.
i realize thats a extreme example but it serves a point, late blunders are more damaging,
how is all that calculated? I dont have the math background to answer this, but i'm curious
Try setting up a couple positions and looking at the sizes of the errors.
There's not really a math thing going on. The fact that early errors "can" be overcome with good play is already embedded into the equity calculation. The equity values for the early positions and late positions are simply numbers that reflect the values the bots think those positions are worth.
Edit: A little bit more detail... just a little.
The equity of any given position is determined by the bot's neural network. Just imagine it's a black box that magically knows how often you will win based on the position. There's a lot of calculation going on based on computer learning techniques and such, but that's not important.
The bot then determines the best play based on the possible positions you can reach from a given roll. Your error is the difference between the equity of your chosen play and the equity of the best play.