[Primeusers] Release 1.0 of PrimeGSR
Lars Arvestad
arve at csc.kth.se
Mon Feb 16 23:04:13 CET 2009
This (0.5 vs 1.0) is probably a bug. We have consistently used 1.0 for
the root, so if you do too, you will be safe.
No, we do not have parallelized the program. It is on the todo-list.
There are many opportunities for parallelization, we just need someone
to do it.
Yes, we are planning to extend the manual! :-) I have not had time to do
much about it though. You are welcome to fire off questions however.
Lars
Filipe Garrett skrev:
> Yes, I was referring to the same tree but having the root at 0.5 instead of 1.
> This tree only works in prime GSR with a "-c" option of, at least, 500 and this greatly increases
> time and memory usage. I suppose it can be because when scaling the tree some branches are just too
> small. I just thought it was strange....
>
> By the way, is it possible to execute primeGSR in parallel (threads, pvm, mpi, ....)?
> Are you planning to add a chapter to the manual to explain the MCMC output?
>
> thanks a lot for your help and time,
> FG
>
>
> Lars Arvestad wrote:
>
>> I am not sure what you mean by "branch lengths roughly half of the
>> branches of the 1.0 root tree".
>>
>> Do you mean the root is at 0.5?
>>
>> The number after -c affects the run time quadratically, but the actual
>> branchlengths should not affect memory.
>>
>> Lars
>>
>>
>>
>> Filipe Garrett wrote:
>>
>>> Scaling the tree to 1.0 at the root apparently works (still running) with both 90 ("-c 200") and 375
>>> ("-c 400") sequences; although with 375 sequences needs too much memory (around 11Gb).
>>>
>>> Changing the "-c" option greatly influences memory usage and speed. I've also tried a scaled tree
>>> with branch lengths roughly half of the branches of the 1.0 root tree, and it only works with "-c"
>>> equal to at least 500. Is it normal that a reduction in branch lengths to half leads to such an
>>> increase in memory usage and time (from 700Mb to 4Gb and approx. 10x slower)??
>>>
>>
>
>
More information about the Primeusers
mailing list