                         |
| ||||
| XGID=--a--aE-CA--eD---c-e-A--A-:0:0:1:32:0:0:0:0:10 | |||||
| Blue to play 32 | |||||
| 1. | Rollout1 | 24/21 9/7 | eq: -0.0235 | |||
| ||||||
| 2. | Rollout1 | 13/11 8/5* | eq: -0.0305 (-0.0070) | |||
| ||||||
| 3. | Rollout2 | 9/7 8/5* | eq: -0.0421 (-0.0186) | |||
| ||||||
| 4. | Rollout3 | 24/21 6/4 | eq: -0.0651 (-0.0416) | |||
| ||||||
| 5. | Rollout3 | 24/21 8/6 | eq: -0.0654 (-0.0419) | |||
| ||||||
| 6. | Rollout3 | 24/22 8/5* | eq: -0.0686 (-0.0451) | |||
| ||||||
| 1 62208 Games rolled with Variance Reduction. Dice Seed: 19348483 Moves: 3-ply, cube decisions: XG Roller 2 31104 Games rolled with Variance Reduction. Dice Seed: 19348483 Moves: 3-ply, cube decisions: XG Roller 3 5184 Games rolled with Variance Reduction. Dice Seed: 19348483 Moves: 3-ply, cube decisions: XG Roller | ||||||
Rollout by Taper_Mike |
eXtreme Gammon Version: 2.10
| Bot | Seed | Equity | Nacbracs | Link |
|---|---|---|---|---|
| GnuBg 1.02 2-ply Trunc | 710040130 | 17.636 | n[B H17 h25 c37 E41] '/0.360t | Taper_Mike |
| XG 2.10 3-ply, Cubes: XGR | 19348483 | 23.5 | n[B H7 h19 E42 c42 c45] "<=62*31 5 | Taper_Mike (above) |
| XG 2.10 XGR++ Eval | [XGR++ seed] | 15.0 | n[B H22 h25 E42 c44] "&e | Taper_Mike |
| Bot | Seed | Equity | Nacbracs | Link |
|---|---|---|---|---|
| XG 2.10 3-ply, Cubes: XGR | 49217571 | 16 | 55 [B H12 h27 E48 c50] "<=10*1 | Keene Marin |
Key: B = Both [24/21, 9/7], C = Cross [24/22, 8/5*], c = cross [24/21, 8/6], E = Each [24/21, 6/4], H = Hit [13/11, 8/5*], h = hit [9/7, 8/5*]
Nactation Tutorial Explanation of nacbracs Nacbracs for cube actions
One of the best arguments against playing practice matches with a bot is that you never get experience with the goofy situations that regular folks create. This position is a case in point.
In general, however, I have found that the real problem is cube actions. Against a bot, you rarely face a bad double. In nonvolatile circumstances, most of the cubes a bot gives will be takes. The exceptions occur when there has been some sort of joker or anti-joker on the previous turn that upset the smooth transition into the doubling window. If you dont include training matches against humans in your study, you will never get experience dealing with wrong doubles.
With checker play, the situation is different. Sure, there are some crazy plays like the one here that a bot might never make. It is my experience, however, that almost any bizarre checker arrangement can arise against a bot. If you think you have a counterexample, just ask Nack Ballard to find a well-played retro-sequence for it. He usually can. I would not be surprised if he could come up one for this position. The bottom line is that practicing against a bot gives you almost all the training you will ever need in how to play the checkers.
The unlimited games rollout above is somewhat closer than the one Keene made at score. The difference may be due to variance as much as score effects.
The opening 43Z is nactated in the Z (Reverse split) family because the smaller die is played on the far side. See my post Clarifying S (Split), Z (Reverse split), and B (Both) for the details.
Mike