With all of the S-Plus 6 disgruntlement that's been aired, I must say that,
so far and for my applications, S-Plus 6 is *much* faster than S-Plus
2000. Now, I'll freely admit to using (and liking) the accursed GUI
graphics capability. I do not, however, use the GUI for any statistical
analysis, in that I prefer the flexibility, control and speed of the
command line. I also like the command line graphics and use each as I deem
convenient.
As a meteorologist, I'm blessed (or cursed, depending on your point of
view) with obnoxiously large data sets, often from numerical models of the
atmosphere. Right now, I'm currently dealing with a partial data set that
is roughly 200 x 37901 for a single parameter, and I have 10 to 15
parameters to examine. When full sized, it will be around 350 x
37901. While I may do admittedly mundane things with these data, even
mundane things become big jobs when the data sets are this large. For
example, in order to adjust for serial correlations when I estimate
degrees-of-freedom for a parameter at a grid point, I use acf() with only
lag-1. It may be overkill, but I'd like to know how variable the lag-1 ACF
is over the entire grid. To do this for each row of my matrix took many
hours under S-Plus 2000, and I had to use a For loop, because it would take
*days* under a more straightforward regular for{} loop due to memory
problems. Under S-Plus 6, it took a few *minutes* with a straightforward
for{} loop, and never once caused my system to thrash (Win2K, 2 x 733 MHz
P-III processors, 512 MB of DDR). I haven't tried the new SpatialStats or
Environmental Stats under S-Plus 6 yet, but I expect an equivalent
performance improvement.
There are some bugs: I uncovered one in the Migration Wizard and Support is
working on it as I type. I may be unusual in that I have _Data folder
that, in one instance, contains around 2600 objects (I know, I know, I
could have done things differently). Currently, for an on-going field
experiment (see http://www.spc.noaa.gov/exper/Spring_2001/elmore/ if you
need to waste some time), I already have over 1000 separate data objects.
Mercifully, I'm near the end of this one. But, once Insightful knows about
a problem, they *will* work on it.
In general, S-Plus 6 is *MUCH* faster than S-Plus 2000, and time is my one
non-renewable resource. S-Plus 6 does a much better job of handling memory
than did S-Plus 2000. I'm not Pollyanna, though. I don't much like the new
help interface because it, of all things, is quite a bit slower than the
help interface under S-Plus 2000. I use help a lot, so this is a
disappointment. But, if S-Plus 6 handles large data sets this much better
than S-Plus 2000, I'm happy with the trade-off.
Kim Elmore
Kim Elmore, Ph.D.
"All of weather is divided into three parts: Yes, No, and Maybe. The
greatest of these is Maybe" The original Latin appears to be garbled.
|