Get identical results for parallel and sequential?

sfSapply(c(1,1),rnorm)

rnorm(2)

No, it's certainly not sufficient to simply make sure the same random
number stream is used.

Depending on the manner in which work is partitioned to processes,
things can occur in an order different from that one gets in a
sequential version.

In typical cases, this will affect only roundoff error, and the parallel
and sequential results should be close to each other.

But they could be quite different.  Say for instance you are doing
variable selection in regression.  Different partitions of the data
could give different sets of predictor variables being chosen, etc.

So, if your results are quite different, the first guess would be that
you simply have a programming error, but there could be inherent
algorithmic issues too.

Norm Matloff


Norm


----- Forwarded message from Libo Sun <libosun at rams.colostate.edu> -----

Date: Sat, 11 Feb 2012 16:39:52 -0700
From: Libo Sun <libosun at rams.colostate.edu>
To: Dirk Eddelbuettel <edd at debian.org>
Cc: r-sig-hpc at r-project.org
Subject: Re: [R-sig-hpc] Get identical results for parallel and sequential?

How to do that? By using sfClusterSetupRNGstream or sfClusterSetupRNG, it
can reproduces the same parallel result, but it cannot give me the same
result as sequential mode.

Libo

On Sat, Feb 11, 2012 at 4:26 PM, Dirk Eddelbuettel <edd at debian.org> wrote:

On 11 February 2012 at 15:27, Libo Sun wrote:
| Hi All,
|
| I have a question about R parallel computing by using snowfall.
|
| How can I set the seeds on parallel workers to get the same result as
| sequential mode?
|
| For example:
|
| > sfSapply(c(1,1),rnorm)
| [1]  1.823082 -2.222052
| > rnorm(2)
| [1] -0.5179967 -1.0807196
|
| How to get the identical result?

By making sure the RNG streams are initialised identically on all nodes.

Dirk

--
"Outside of a dog, a book is a man's best friend. Inside of a dog, it is
too
dark to read." -- Groucho Marx

        [[alternative HTML version deleted]]