how can 13/11 13/9 be soooooo terrible???

[kerosene]

Hi all, I'm confused with this analysis by GNU

My logic was:-

1) it's probably a race and we're even after this move so it goes with the luckier dice
2) i'm going to try to bear in with a slope down from 6 to 4 to avoid wastage.
3) 13/11 13/9 crosses two corners and puts them in range to bear in.

so why would GNU place that move 22nd out of 22?
.... and why would moving  3/1 be in the top 3 moves followed by 4/2 and 6/2.





I thought a very smart expert said the sloping bear in between 6 and 4 (or 3) was best

is there a better shape?

boop 

[Zorba]

Good question boop, and a simple answer too: use at least 2-ply for GnuBG. If you find that too slow for general use, at least use 2-ply whenever there's any doubt about the 0-ply analysis.

0-ply is pretty decent overall, but there's several situations where it makes misevaluations upto 0.1 equity, occasionaly more, that are corrected (for a large part at least) by using 2-ply.

I'm pretty sure the 3/1 moves will go down the list at 2-ply. What the very best move is in this position is hard to say, most likely a few moves are very close. 13/11 13/9 should be right up there I think, and could very well be the best move AFAICS.

After using 2-ply, you may also try a 3-ply for the top moves, see if there's any changes.

[dorbel]

Some points to note.
The pipcount is level after your roll, but the race isn't, because your opponent will be on roll, worth about 4pips. At this length of the race, that makes him about a 54% favourite, an advantage that will grow as the race shortens if you both roll the same numbers.
You are using 0-ply for your analysis. It is, in robot terms, rubbish. Why not take an extra five minutes (max) and use 2-ply for your match analysis and get a more accurate answer? Then your play will still be down the list but closer at about 0.016 behind the "best " play.
Unfortunately, bots are notoriously poor at sorting out races with a long way to go and a lot of choices, so once you have narrowed the field a bit, you need to do a rollout, ideally 2-ply to avoid incorporating lots of weak 0-ply moves. You could save time by truncating it, 9 rolls time might be good enough and/or stop when you get to the exact bearoff database. I expect that if you do this you will find that your play is best after all and certainly better than the crappy 13/9, 3/1.
Filling that 5pt is vital and I would play a 1 on the next roll 6/5.
..........and what Zorba said, except that I have found in the past that the odd numbered plys in Gnu do strange things, so I avoid them. Perhaps that has been sorted, I don't know.

[kerosene]

ahhh!! i had no idea the levels of analysis could be sooo different!

2 ply places my move 2nd and not doubtful .. oh, and after a wait, 3 ply places it top!!
this is going to reduce my confusion with GNU immensly

Thanks zorba

boop  

Warning - while you were typing a new reply has been posted. You may wish to review your post.

thanks dorbel ... good points - I had my first inkling about ply only last night, i think on www.bkgm.com, but hadn't looked to check what my default GNU setting was. 2 ply seems to be very fast ... onwards and upwards

boop