[Bug-gnubg] NN for DMP, GS, GG etc.

[Robert-Jan Veldhuizen]

At 11:43 9/30/2002 -0400, Joern wrote:

> A related idea would be to include the gammon (and backgammon) price as
> inputs to the neural net, and train the net on a number of different
> gammon prices.

Would that really work? My intuition says no... but who knows what the NN 
is capable of... any other ideas on this?

It'd be interesting to have seperate NN for various gammon prices. I guess 
GG and GS might show the greatest differences, however, DMP is probably the 
most interesting to consider. An improvement in DMP play (where there's 
100% MWC at stake!) probably has the biggest overall effect on match play 
strength, and DMP (also post-Crawford doubled, etc.) happens quite often.


--
Robert-Jan (Zorba on FIBS)