(1)
First of all, I don't understand why you're calling this the "cubeless win %"
It wasn't so much that I "called it" that. Rather, when I divided the 880 net points won into 1296 trials, the result -- which I expected would be equal to the rollout's reported cubeful equity, +0.5760 -- was instead equal to the rollout's reported cubeless win percentage, +0.6799. I don't understand why.
(2)
because you ran the rollout with variance reduction. Your calculation ... If you turn variance reduction off, then I would expect GNU to report 0.679 as the cubeful equity
So it does!
Here's another rollout, this one with variance reduction off. Notice that the numbers are wrong (assuming they were right in the other rollout), and the SDs are so large 1296 trials isn't even enough to determine with confidence whether it's a double or not.
The score (after 0 games) is: White 0, Blue 0
Move number 5: Blue on roll, cube decision?
| White | 30 |
|---|---|
| |
| Blue | 30 |
Position ID: 2zYAANi2AQAAAA Match ID: cAkAAAAAAAAA | |
• Blue doubles
Alert: double decision marked doubtful
| Cube decision | |||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Rollout cubeless equity | +0.3601 | ||||||||||||||||||||||||||||||||||||||||||||||||||||
| Cubeful equities: | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| 1. | Double, take | +0.5849 | |||||||||||||||||||||||||||||||||||||||||||||||||||
| 2. | Double, pass | +1.0000 | +0.4151 | ||||||||||||||||||||||||||||||||||||||||||||||||||
| 3. | No double | +0.5702 | -0.0147 | ||||||||||||||||||||||||||||||||||||||||||||||||||
| Proper cube action: | Double, take | ||||||||||||||||||||||||||||||||||||||||||||||||||||
| Rollout details | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Full cubeful rollout without var.redn. | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| 1296 games, Mersenne Twister dice gen. with seed 845469288 and quasi-random dice | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Player 0: | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Play: supremo 2-ply cubeful prune [world class] | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| keep the first 0 0-ply moves and up to 16 more moves within equity 0.32 | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Skip pruning for 1-ply moves. | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Cube: 0-ply cubeful [expert] | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Player 1: | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Play: supremo 2-ply cubeful prune [world class] | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| keep the first 0 0-ply moves and up to 16 more moves within equity 0.32 | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Skip pruning for 1-ply moves. | |||||||||||||||||||||||||||||||||||||||||||||||||||||
| Cube: 2-ply cubeful prune [world class] | |||||||||||||||||||||||||||||||||||||||||||||||||||||
For this rollout, the data from the "view statistics" tab says, for "Player White holds 2-cube:
Blue wins 807 2-cubes, 41 4-cubes and five 8-cubes, while losing 360 2-cubes, 81 4-cubes, and 2 8-cubes.
807 + 41 + 5+ 360 + 81 + 2 = 1296.
(2*807+4*41+8*5)- (2*360+4*81+8*2) = 758
758/1296 = 0.5848 ... which matches the reported cubeful equity.
[Having one side play 0-ply cube was a left-over setting. This doesn't effect the "view statistics" utility. If both sides play 2-ply cube, the results of a 1296-trial no variance reduction no database truncation rollout is similar]
(3)
Finally, the discussion started off as a discussion about Snowie rollouts
True, and I don't use Snowie, so I'll bow out ... I butted in because Ray asked about how to interpret the game breakdown in relation to the equity(ies), which, as far as I know, works the same way with Snowie and GnuBG.