Re: [Bug-gnubg] Re: Train bots for DMP, GG, and GS play ?

[Joern Thyssen]

On Tue, Oct 01, 2002 at 01:56:49AM +1200, address@hidden wrote
> This idea is floating around for some time now. I think Zare hinted at some 
> general improvments for all scores in his new bot, but I might be wrong. I 
> will look at generating a DMP net at some stage. 

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. 

Pros:
- we use the same neural net so we do not have discontinuities at
  borderline match scores
- we can use it for every gammon price

Cons:
- it'll be many times more expensive to train the neural net, since we
  need rollouts as many different match scores (i.e., match scores)
- we use the same neural net (it need to accomodate more
  information and may consequently play worse?!)

Well, just a few thoughts...

Jørn

-- 
Joern Thyssen, PhD
Vendsysselgade 3, 3., DK-9000 Aalborg, Denmark
+45 9813 2791 (private) / +45 2818 0183 (mobile) / +45 9633 7036 (work)
Note: new mobile number!