On Nov 29, 2009, at 3:40 PM, Guillaume Chapron wrote:
You might try to write the results into a file and read it back in
R. This is probably safer anyway.
This was my initial approach, but I expect a lot of overheads
reading an array that can be as large as 100,000,000*10,000*100.
Coding it should take at most, an hour, I would try it, you might be
surprised. Usually, I always code a working version then optimize
it. More than once, I have been surprised when the problem areas
turned out to be completely different than I originally thought and I
avoided a lot of additional complexity trying to solve a problem that
did not exist. If the simulation runs a few hours, spending a few
more minutes reading and writing the results might be inconsequential.
But did you mean that the Cocoa app should save its output as an R
workspace? Is it possible to convert and save a C array into its
equivalent in an R workspace?
You might follow this, http://developer.r-project.org/embedded.html to
embed R into the Cocoa app. I have no experience with this but it may
lead you to something else. You might also look at https://stat.ethz.ch/pipermail/r-sig-mac/2005-March/001706.html
Its possible to use interprocess communications to have R request the
data from the Cocoa App. However, there are a number of problems with
this. For example, the cocoa app has to be running. When R is
requesting the data it will be locked up waiting for the data. You
need to read state back and forth to know when to do what. You will
need to write an R extension in C, much the like the database
extensions. This seems like a lot of complexity to me. In essence,
your Cocoa GUI app would act like a database provider.
If R processing is segmenting the array at all, then you might be able
to use a common database from both the Cocoa app and R. A database
just writes to a file with a lot more overhead, so if its not
segmenting the array then you are probably better off reading a flat
file as it all needs to be in memory at one time.
While searching more on the web, I came across a post of someone
wanting to export a C++ array to R, and Simon Urbanek replied "the
short answer is no, you can't, because the memory has to be
allocated by R". But R.app is Applescript enabled, so I should be
able to tell R to create an array, and maybe later assign this
object to the simulation output.
I would guess that writing to a file and importing into R would be
much faster than using Applescript.
Why not run the original dyn library simulation as another user,
then the current user can do whatever they want with R? To do this
enable "fast user switching" in the "Accounts" preference. Now you
can leave a program running as one user and log in as another user
to do something else. Of course, you will need to create a 2nd
user account.
That would work, but that would not be very practicable, for
example, I would need as many accounts as simulations I want to run
at the same time.
Then why not run R from the command line, you can have as many R
instances running as you have machine (memory and CPUs), then use your
original library version.
_______________________________________________
R-SIG-Mac mailing list
R-SIG-Mac@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-mac
