Re: [Bug-gnubg] Bot comparison on BGonline - preliminary results

[Jonathan Kinsey]

Jonathan Kinsey wrote:
> Michael Petch wrote:
>> I am wondering if there's any connection to the weird cube decisions Neil
>> Robins discussed previously:
>> http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=47213
>>
>> I still think there is a bug even in the scenario above (even if not
>> related to the Depreli positions).
>
> I've had a quick look and if the bear-off database isn't used then the 3+4ply
> results both give double take. This might be entirely coincidental but I'd
> expect the bear-off values to be more accurate than the race values?

I did a further test where the code used the logic for race evaluations but
actually got the results from the bearoff database (rather than the NN) and this
also gave more correct answers (compared to using the actual bearoff logic).

Maybe someone can explain the differences in race vs bearoff evaluations? As I
would think (on a logical level), that my test above should give the same values
as using the bearoff database directly (as all the evaluation are exact).

Jon

---

New Windows 7: Find the right PC for you. Learn more.