RE: [Bug-gnubg] Questions on pruning net

[Ian Shaw]

> -----Original Message-----
> From: Joseph Heled [mailto:address@hidden 
> Sent: 23 November 2004 21:24
> 
> Ian Shaw wrote:
> > 1. When the pruning nets have finally selected a move, is that move 
> > re-evaluated using the full nets at the relevant play, to give the 
> > best possible evaluation of the equity?
> > 
> I think most of you still miss the point of the prune nets. 
> It just pre-filters the options in a move selection *inside a 
> higher ply*. If the move is on it's top 10 list, the error is 
> 0. If not, the error is the equity difference between the top 
> move on the list and the top *0 ply* move.

So the pruning nets pick the pest 10 moves, then these 10 moves are evaluated 
using the main net as before.

> > 2. How strong are the pruning nets in their own right? How do they 
> > compare with 0-ply?

[snip interesting details] Do you equivalent numbers for 1-ply and higher?
> 
> So, prune error rate per move is ~ three time higher than 0 ply.
> 
... but about 10 times worse on cube decisions. I don't really understand how 
this can be, since you are still looking at play selections to evaluate the 
overall cube equity.

> "I don't understand those numbers", I hear you saying very 
> quietly. So In money play, prune give around  0.50960 ppg to 0ply.
> In 7pt matches, 0ply wins approximately 80% of the time.

Half a point per game is a lot, as is 80% of 7-point matches. So the pruning 
net on its  own is quite weak. (This is an observation, not a criticism.)

> > 3. Might it be useful to use the pruning nets as a playing option 
> > lower than "expert", rather than introduce noise? 

I think this might be worth doing. It will be good to give people a weaker 
opponent to play, but I doubt its usefulness for rollouts. (Hmm, unless we want 
to roll out expert vs intermediate etc).

Øystein's post also made me wonder whether pubeval can be listed in the playing 
options. (I know you can set it up in the players menu, but it is unlikely that 
the average users knows what it is.)

I've set up an overnight session of gnubg vs pubeval. I'll analyse it the 
morning to see how it rates.

-- Ian