Re: [Bug-gnubg] Doubts about the new pruning net

[Joseph Heled]

Robert-Jan Veldhuizen wrote:

>  >I run the benchmark overnight and I get for 13294 moves
>  >Pruning 124 differences to full 2ply, error 0.00726047 vs. 0.00725554
>  >Reduced33% 914 differences error 0.00755261
>
> I don't know what these figures mean. What benchmark are you running? 
> One position rolled out many trials, or many different positions 
> evaluated? The real problems I'm getting are with rollouts.
>
> Are the differences you report for checker play, cube or both? Is it 
> money play, or also matchplay?

The benchmark is the standard benchmark we all rolled together. made of 
positions, where top moves are rolled out for one dice

This is a later number,
                               full-2ply error per move    2p-prune  33%-reduced

154,1177/17342 (0.009 0.068)           0.00714507        0.00715006 0.00746464
^     ^    |   (as percentages)
|     |    |
Number|of t|imes 2ply-pruning disagrees with full 2-ply
      |    |
      Numbe|r of times 2ply-33% reduced disagrees with full 2-ply
           |
           Number of positions checked

As you can see pruning error differ from full by 0.000005, and reduced by 
0.00032, almost two orders of magnitude.

-Joseph