|
| ||||||||||||||||
It might be a good idea if you got the 4-ply eval of the original position before deciding whether you want to expend the CPU time to roll out your proposed variant position.
Here's what you know:
42P-52S-66B-55 (left-hand position)
[????] 4-ply eval
[A P55] 1k
51S-52S-21@-66B-55 (right-hand variant)
[A P1] 4-ply eval
[????] ??k
By replacing the question marks in the first bracket (with the eval), you can better predict the second question-marked bracket. Eval biases tend to be somewhat consistent. If there is a bias, it admittedly might be related to Blue's blots in the left-hand position, but it might very well not, or it might be just a partial effect.
Even if you are going to roll out the variant position regardless, you may be surprised how much you can learn about eval biases from the comparisons, and as a consequence of spotting the weaknesses/quirks of GnuBG (or other bot), you'll get better at predicting rollout results.
Unfortunately, the first rollout is only 1k trials -- there's a lot of volatility in the margin. And if you do the same for the variant position, you're dealing with two unreliable margins. That reduces how much you can glean from results or result pairs. Fortunately, you have the option of extending rollouts now or later, as you feel is indicated.
Nack