| |
BGonline.org Forums
Disadvantages of Seed 0
Posted By: Taper_Mike In Response To: ALT *RO* 61P-31P-33 (Dmitriy Obukhov)
Date: Saturday, 18 February 2012, at 9:05 a.m.
There is nothing particularly wrong with using the seed 0 — rollouts with that seed are just as valid as those with any other seed — but there is a good reason not to use it. It has to do with combining rollouts from different sources. In order to combine two rollouts, their seeds must be different. Even then, the process is problematic, because two rollouts are not independent unless the entire sequence of random dice rolls used in one does not overlap any part of the sequence used in the other. Specifically, if the two random dice sequences overlap in such a way that rolls are repeated starting from the first turn of a game, then the trials from that point on are not independent. In practice, backgammon players tend to ignore this difficulty, and combine any rollouts that have a different seed.
Those who know XG1 better than I do can correct me, but I believe XG1 suppresses the display of the dice seed used in a rollout whenever the seed is 0. I know for certain that is how XG2 works. As well, XG defaults to using the seed 0 for all rollouts. That is how it is setup when you initially install it.
This thread contains a case in point. There are three rollouts of the position 61P-31P-33. Stick’s uses the seed 74138057, but both Paul and Dmitriy post rollouts where the seed is suppressed. I believe both of them have used the seed 0. As Nack points out, the fact that their rollouts match almost exactly tends to support the thesis. If I am right, then the data from Stick’s and Paul’s rollouts can be combined (they use different seeds), while that of Dmitriy should be held out (it uses the same seed as Paul’s).
Now you can see why there is a disadvantage to the seed 0. If we all were to use it, then we could never combine any of our rollouts.
You can change XG’s seed options by opening the Options and Setting dialog box. Click Options->Settings, and then click the Rollouts tab. At the bottom of the tab, put a check mark in the check box labeled “Generate a new seed randomly each time the program starts.”
I have read suggestions that there is an advantage in using the same seed for all rollouts. The idea is that rollouts of different positions can be compared more readily when they use the same seed. For instance, the contrasting positions 61P-31P-33 and 65R-61P-31P-33 that are discussed in this thread might be more accurately compared when their rollouts use the same seed. The rollouts posted in this thread have 24/18(2) as best in 65R-61P-31P-33, but that move is only third best in 61P-31P-33. Is the difference due simply to variance? Some suggest that when both rollouts use the seem seed, variance is lessened. I am not so certain myself. Some of the mathematicians reading this may wish to comment.
Even if you do believe that all your rollouts should use the same seed, that seed should not be 0! You can have your cake, and eat it, too. Choose some other constant seed. Timothy Chow often uses 2. Stick used 1296 for a while. I don’t care for either of these choices. My preference is for a larger, random value. Another roller (Miran?) used 1296 once or twice, and his rollouts could not be combined with Stick’s. Since then both Stick and Miran have switched to “Generate a new seed randomly each time the program starts.” Some of the rollers using 0 recently include several giants who post regularly to this site.
And don’t forget me. I am just as guilty as anyone else. Before I had all this explained to me, I used seed 0 for the longest rollout I have ever made, the 11-day, 4-ply rollout of 63S-32-s for Nack’s T4 project.
Whoops.
Mike
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.