On May 19, 8:09 am, William Stein <wst...@gmail.com> wrote:
> On Wed, May 18, 2011 at 10:59 PM, lgautier <lgaut...@gmail.com> wrote:
>
> > On May 18, 11:21 pm, William Stein <wst...@gmail.com> wrote:
> >> On Wed, May 18, 2011 at 2:16 PM, William Stein <wst...@gmail.com> wrote:
> >> > On Wed, May 18, 2011 at 12:24 PM, kcrisman <kcris...@gmail.com> wrote:
> >> >> On the page you linked to:
> >> >> "This the documentation for what will be soon the older version of
> >> >> rpy2. Do consider the upcoming 2.1.x releases if you are starting a
> >> >> project now."
>
> >> >> If you look athttp://rpy.sourceforge.net/rpy2/doc-2.2/html/changes.html
> >> >> it also seems like in higher-numbered releases there are changes in
> >> >> this functionality. Maybe this is a known bug; can you try
>
> >> >> from rpy2.robjects.numpy2ri import numpy2ri
> >> >> ro.conversion.py2ri = numpy2ri
>
> >> >> as on the 2.2 documentation and see if that works.
>
> >> > Nope, not at all.
>
> >> >> That said, I never use it - I always use the Sage pexpect interface to
> >> >> R. But now I will bookmark this worksheet if I get the chance to give
> >> >> another R/Sage talk!
>
> >> > The Sage pexpect is slow and brittle if the size of data you need to
> >> > move into or out of R is large. The rpy2 interface is much more
> >> > robust in this regard.
> >> > But rpy2 is also somewhat weird and wacky, as I learned today.
>
> >> And, surprisingly, it is not very fast for evaluating basic
> >> expression. It's really *shockingly* slow for a C library interface.
> >> I don't know how it can be so bad:
>
> >> sage: import rpy2.robjects as robjects
> >> sage: R = robjects.r
> >> sage: print R('2 + 3') # the rpy2 cython interface (note the import!)
>
> > I don't think that Cython is currently used anywhere in rpy2.
>
> I know. I meant "C library" interface. Sorry for the confusion.
>
>
>
> >> [1] 5
> >> sage: timeit("r('2+3')")
> >> 5 loops, best of 3: 1.46 ms per loop
> >> sage: timeit("R('2+3')")
> >> 625 loops, best of 3: 686 µs per loop
> >> sage: timeit("pari('2+3')")
> >> 625 loops, best of 3: 5.64 µs per loop
>
> >> Seriously? 686 microseconds to do 2+3? Our PARI C library
> >> interface does that in 5.64 microseconds.
>
> > You may want to check your settings. Here I have with Python 2.7 /
> > R-2.13 / rpy2-2.2.0beta3 :
>
> >>>> import timeit
> >>>> t = timeit.Timer(stmt="R('1+2')", setup="import rpy2.robjects; R =
> >>>> rpy2.robjects.r")
> >>>> print "%.2f usec/pass" % (10000 * t.timeit(number=1000)/1000)
> > 1.37 usec/pass
>
> I get the same horrible timings (around 305 microseconds) on Linux as
> well, so it's not some trivial setting issue.
> From your benchmark, the problem must be either the version of R,
> Python, or rpy2 that we
> ship with Sage. All are older than the versions you're testing above.
> I'm very glad that
> there is some future version of rpy2 where it isn't so slow; thanks
> for letting me know.
Note that the rpy2.robjects layer is trading performances for
convenience, and is implemented in pure Python.
When after speed, the rpy2.rinterface layer is closer to the C-level
interface.
>>> t = timeit.Timer(stmt="res = reval(rparse('1+2'))", setup="import
>>> rpy2.rinterface as ri; ri.initr(); rparse = ri.parse; reval =
>>> ri.baseenv['eval']")
>>> print "%.2f usec/pass" % (10000 * t.timeit(number=1000)/1000)
0.17 usec/pass
> I hope somebody will update rpy2 (and/or) in Sage.
>
> -- William
--
To post to this group, send an email to sage-devel@googlegroups.com
To unsubscribe from this group, send an email to
sage-devel+unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URL: http://www.sagemath.org
