Basic analysis says that you shouldn't play 13/6, leaving a bad 61 next time, or 8/5, 8/4, removing a potential landing point for the checkers on the mid-point.
After that you decide to shuffle the checkers in your home board. The question is how? Logic says that there shouldn't be much difference between the various plays and many would choose Neilkaz's 5/2, 5/1. Indeed that is Snowie's choice on rollout.
The problem is that XG thinks that 6/3, 5/1 is the standout play as evidenced by the rollout information below. Can that play really be that much better? The players who contributed to the Facebook discussion decided that this is an XG bug. I have sent it to Xavier and asked him to run it through XG3 but, as noted by others, he appears to be on radio silence. This, to my mind, is clear evidence that XG is not by any means a finished product and there is room for development/improvement.
The last twist in the tail is that I did a rollout on gnubg and it also had 6/3, 5/1 as the best play by a huge margin with all other plays virtually double blunders.
So, do the bots know something we don't, or have XG and gnubg got the same programming fault?
Discuss!
Chris
                         |
| ||||
| XGID=--ACBBBbB----C-a-aadbbb---:0:0:1:43:0:0:3:0:10 | |||||
| Blue to play 43 | |||||
| 1. | Rollout1 | 6/3 5/1 | eq: +0.536 | |||
| ||||||
| 2. | Rollout1 | 5/1 4/1 | eq: +0.474 (-0.061) | |||
| ||||||
| 3. | Rollout1 | 5/2 5/1 | eq: +0.472 (-0.064) | |||
| ||||||
| 4. | Rollout1 | 6/2 4/1 | eq: +0.467 (-0.069) | |||
| ||||||
| 5. | Rollout1 | 8/5 8/4 | eq: +0.466 (-0.070) | |||
| ||||||
| 6. | Rollout1 | 13/6 | eq: +0.463 (-0.073) | |||
| ||||||
| 7. | Rollout1 | 6/2 5/2 | eq: +0.458 (-0.078) | |||
| ||||||
| 8. | Rollout1 | 6/3 6/2 | eq: +0.439 (-0.097) | |||
| ||||||
| 1 1296 Games rolled with Variance Reduction. Dice Seed: 5 Moves: 3-ply, cube decisions: XG Roller | ||||||
eXtreme Gammon Version: 2.10