Game Plan Selection 2 – Attack or Prime?

Game Plan Selection 2 – Attack or Prime?
by Mary Hickey – July 2006

Red’s choice in the position below is whether to attack the loose checker in his homeboard or to form a five-point prime and split his back men.

 

 

 

Problem 1

Here, the correct theme is “attack a single back checker”. This is often the case, since a blocking structure is most effective against two or more checkers. This is mostly because they aren’t constantly threatening to escape completely, the way a single checker is, especially when it’s already at the edge of the block as it is here. Another reason is that two men back means the opponent has only 13 checkers to use to counter-prime or attack you elsewhere.

The priming approach can be effective as a prelude to an eventual attack, but not when he’s at the edge threatening to escape, and you have no builders to attack him if he doesn’t. Meanwhile, he can use many of his non-escaping rolls to strengthen his front position, making your later attack more dangerous for you. Right now, while he has no board, getting attacked is dangerous for him. So, smack him while you can!

The bots had trouble with this problem, with Snowie barely getting it right only on 3-ply, and JellyFish almost getting there but still falling short on Level 7. The rollout below shows that attacking is clearly best, by .062 money equity.

Techie-talk Warning: Non-techies please top off your coffee mugs, or else skip ahead two paragraphs. Snowie 3, using a huge search space and 100% speed, got this problem wrong at first, by .076 on 1-ply and .112 on 2-ply. By 3-ply it had corrected itself, choosing the right play by .009. This was a problem where simplifying the evaluation by using a lower “3-ply speed” led to the wrong answer. When I tried this with a huge search space but only 20% speed, instead of the 100% I always use, it chose the wrong play by .032. If you run matches from your favorite backgammon game server through Snowie, I recommend you use 3-ply with 100% speed on the first pass, even though it will take a little longer than with, say, 33% speed. It doesn’t take all that long, unless your computer is truly a dinosaur, and the extra accuracy is well worth it IMO.

JellyFish 3.0 also picked the wrong play here, but by a steadily declining margin: .091/.065/.028 from Level 5 through Level 7. For those of you who are still fans of the Fishy, stay alert for this pattern. It means a rollout is mandatory even if the “final answer” on JellyFish’s Level 7 nominal 3-ply (which is said to be actually what we call 2-ply in Snowie-speak and plays more like a Snowie 2-and-a-fraction) is still a substantial error, such as .05 or .06. The same is true when JellyFish shows a pattern such as .090/-.110/.050. It’s not hard to see how in either case, further ply might progress to a reversal of the Level 7 answer, and the rollout will frequently do just that.

The actual problem came from a match I played on FIBS on January 12, 2000.

Problem 1 with rollout: