[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Could you win this game as Blue 1 out of 5000 times?

Posted By: Taper_Mike
Date: Sunday, 5 June 2016, at 6:40 p.m.

In Response To: Could you win this game as Blue 1 out of 5000 times? (benf)

To elaborate a bit on this, XG performs the analysis is it set to use. That might be 3-ply/XGR cubes, or 4-ply, or even XGR+ or XGR++. Each of these "looks ahead" a certain number of rolls. The look-ahead examines all (or at least a sample of) future possible dice rolls. At the end of each look-ahead sequence XG examines, XG uses its neural net to evaluate the position. Thus, there are many different positions are neural net evaluations that are checked. The final evaluation is a weighted average of the many neural net evaluations that were made along the way. Variance reduction can also affect the calculation.

It seems odd to us humans, but sometimes the net's evaluation of a gin position will be a little off. That's because the neural net evaluation is, itself, just a weighted sum of its various inputs and fed-back outputs. All the net does is calculate a number (or, rather, 6 numbers, one each for win rate, gammon win rate, backgammon win rate, gammon loss rate, and backgammon loss rate). The net has no understanding of whether the numbers it calculates are sensible or accurate.

Hope this helps.

Mike

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.