| |
|---|---|
| 54S-21H-31 | |
Speaking to general audience now: People often seem skeptical when I tell them I give as much weight to a 5- or 7-depth 3-ply trunc (i.e., truncated rollout) of 1k (i.e., 1296) or even fewer trials as I do to a full rollout of 5k (i.e., 5184) trials. However, my methodology is based upon many years of observing a huge amount of early game rollout data.
My old 2k trunc is [U V5]. (Translation: the Up 24/21 play beats the Variant-up Bar/22 24/23 play by a margin of .005.) By contrast, Stick's 5k full is [U V17]. In general, which of a 1k trunc or 5k full is more reliable? It's hard to say. The first result is at the mercy of misevaluations at the end of the x-move sequences. The second is at the mercy of variance.
If the full is fewer than 5k, I lean towards the trunc. If the full is more than 5k, I start to lean towards the full. In this case the full was exactly 5k, so I weighted the trunc and full 1:1 and derived [U V11] as "my current best guess of value...," as mentioned in the third paragraph here.
Now Chuck has run an independent 20k full rollout of [U V7]. Adding Stick's [U V17] 5k gives us a combined XG full result of [U V9] 25k.
I weigh a 25k full 9:1 compared to a trunc, so my [U V5] trunc now has a much smaller effect on perceived value. (If I want to go out another decimal place, I tweak the full's .009 margin down to .0086.)
In this case the 1k trunc (margin of .005) was a better predictor of value than the 5k full (margin of .017). It didn't have to work out that way. It often doesn't. But overall the predictive value of the 1k trunc and the 5k full seem to be about equal for early game rollouts, in my experience.
So, early game rollout enthusiasts, please understand that the old truncated results I drudge up have value. (This was especially true when all we had was Snowie and much slower computers, but it's still true today.) Until a rollout with a substantial number of trials has been done for a given position, I still pay close attention to any truncated rollout data I happen to have.
Nack