Re: [Bug-gnubg] gnubg.bd

[Joern Thyssen]

On Tue, Nov 05, 2002 at 07:30:24AM +1300, Joseph Heled wrote
> 
> >So the "price" for storing a zero gammon distribution is 2 bytes. I
> >think I can live with that.
> >
> >For example, the 6 pt database is 6,945,792 bytes for both bearoff
> >distribution and gammon distribution. With the compression scheme above
> >it shrinks to 1,467,498 bytes. Storing only bearoff distribution results
> >in a file of size is 1,339,230 bytes, so storing the gammon distribution
> >only costs 130kb. 
> >
> >Still, there is some overhead compared to your bb1.o at 1,126,324 bytes.
> 
> But I don't store gammons. 

Yes, but my gammonless file si still 1,339,230 bytes, However, the index
takes up 434,112 bytes, so the actual bearoff distribution is 905,110
bytes which is in line with your 1,126,324 bytes.

> The numbers look very reasonable. I only wish 
> you would have waited for me to finish the analysis and see if it is needed 
> or not :)

I've just been playing around with the code trying to understand it.

I'm not convinced of the value of the onesided bearoff databases either.
But I might as well commit my code. Anyway, we can use them for
benchmarking OSR and the race net. 

The two-sided bearoff databases are more interesting since the give
exact probabilities as well as exact cubeful money equities.

> 
> On the whole, it is good to give the user the options of deciding if she 
> wants perfect play at the expense of time and disk space. I am just not 
> sure it should be so high on your agenda at this stage.

I'm not sure anyone would bother wasting approximately 40 hours of CPU
time to generate the 13pt database, but at least now they have the
option. Maybe some benchmarks agasint the race net can convince them
that the gnubg race net at 2-ply is actually quite good compared to
Snowie's race net. 

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!