Following the recent discussion on differences between average PR of various XG's settings here: http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=196934, and having some spare CPU time available, I've made a small test.
I've chosen 8 various settings and measured average PR & speed over the sample of thirty 13-point matches, both sides, for an EP of 30 x 13 x 2 = 780 pts. You can find the results below.
Software: XG2 (ver. 2.10.199.2658)
CPU: i7-4790 @ 3,60 Ghz (4 cores x 2 threads), calibrated at 5 900 000 pos/s
Common features:
- bear off database generated for 15 checkers over 10 points
- use opening book (ver. 3.70)
- 2nd pass use all nods (different, error, outplay, blunder)
- resign errors counting
Table 1. Average PR and CPU time
| Run No. | Settings | Average PR | CPU time | |||
|---|---|---|---|---|---|---|
| PR | vs run 1 | time [min] | vs run 1 | |||
| 1 | 1-pass XGR++ Gigantic | 5.15 | 1922 | |||
| 2 | 1-pass XGR+ Gigantic | 5.20 | + 1.0% | 176 | 9.2% | (about 11 times faster) |
| 3 | 1-pass XGR+ Large | 5.18 | + 0.6% | 79 | 4.1% | (about 25 times faster) |
| 4 | 2-pass 4ply / XGR++ Gigantic | 5,05 | - 1.9% | 272 | 14.1% | (about 7 times faster) |
| 5 | 2-pass 4ply / XGR+ Gigantic | 5.03 | - 2.3% | 68 | 3.5% | (about 30 times faster) |
| 6 | 1-pass 4ply Gigantic | 5.20 | + 1.0% | 49 | 2.5% | (about 40 times faster) |
| 7 | 1-pass XGR Gigantic | 5.44 | + 5.6% | 21 | 1.1% | (about 100 times faster) |
| 8 | 1-pass 3ply Gigantic | 5.02 | - 2.5% | 3 | 0.16% | (about 600 times faster) |
The following diagrams show the single entry deviations (single player in a single match, for a total of 60 entries) between given settings and run #1 settings, grouped into ranges of 1% span.
