Re: [structure-group] Random Seed?

[Vikram Chhatre]

Thats interesting.  I dont know the relationship between seed number variation and outcomes of respective projects.  However, I believe you can also arbitrarily assign seeds to new runs.  Setting RANDOMIZE=1 will automatically do this.  

Or you can specify seed number in the command with the '-D' switch.

V

On Wed, Jan 23, 2013 at 5:01 PM, Mattias Johansson <m.l.jo...@gmail.com> wrote:

I'm running Structure in batch on our local cluster. Because I'm starting multiple jobs at essentially the same time on different processors, the "random" starting seed ends up the same number (or very close). Does anyone know how much difference this makes to the outcome? I've built in a small pause, but that only adds seconds, so the numbers are still within a few hundred (between 1358979758 and 1358979923 for 110 runs, for example). It's not very random, anyway...
-MLJ

--
You received this message because you are subscribed to the Google Groups "structure-software" group.
To view this discussion on the web visit https://groups.google.com/d/msg/structure-software/-/ZEMik1sgvZ4J.
To post to this group, send email to structure...@googlegroups.com.
To unsubscribe from this group, send email to structure-softw...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/structure-software?hl=en.

[Julie Hebert]

Mattias,

I ran into the same problem when I was using our local cluster. I ended up setting up batch files setting the random number in each mainparam file (by generating random numbers myself). I didn't do it at the time because I hadn't learned R yet, but I think it could be relatively easy to create the files in R if you need to. I would try Vikram's suggestions first, and if that doesn't work, try setting up a few of your own files. Note: as long as the numbers are different from one another, it shouldn't be a problem. My problem was that they all started at the same time, and therefore, within a given K, all of the results were the same. It was really obvious this was the case, so if you're getting variation, you should probably be fine.

Julie