[R] Moving from Splus to R - advice/opinions request from amanagement perspective

Brian Koch bkoch at decisiondevelopment.com
Thu Jul 27 21:53:15 CEST 2006


 
Hi Dave.  I'll try to offer some feedback from a fellow non-stats guy
working in a business setting.  I've not directly transitioned from S+
to R, but I've previously hit roadblocks with SAS and SPSS that
compelled me to investigate the S/R languages.

I think that as you read messages on this list, you'll find that most
users don't implement R as a back end to a chain of processes: Many of
the community members are importing data into R via an ASCII or SQL
connector (R has good base functions and/or packages for both), writing
custom code for their analyses, and using the R output to [manually]
compile a report or similar deliverable.

So my biggest concern for the migration is the import/export.  Given
your evaluation of your statistics needs such that S+ might be
"over-kill" - and because your organization seems to be at a crossroads
for this information system, maybe take a good look at what is serving
S+ its data.  If it's something to the tune of SAP, MySQL, or Oracle,
then that application might have enough mathematical/statistical/data
management features to accomplish your objectives -- not to mention that
all of these applications are extensively supported by their
manufacturers (for a price, of course - hopefully competitive with
Insightful and more customer-responsive).  If you're planning to
outsource your statistics expertise, then definitely consider this
option.

The biggest differences between S+ and R deal with data storage and
variable scoping -- and this impacts import/export especially.
Transitioning your code from S+ to R will be the most intensive when
routines based on these differing features need to be re-engineered.

Another consideration for management is how you might use R packages.
There are a large number of them available, but as you'll see in reading
this list's messages, the quality varies.  I've found the R base
software to be rock-solid, and I trust its accuracy.  But I've been
reluctant to deploy packages whose workings are beyond my expertise
(i.e., a package that I couldn't debug).  If you need capabilities
outside of the base functions, then take a good hard look at which
packages are out there (CRAN is where you find R packages - link to it
from the R main site) and examine their authors and the degree stability
and/or active maintenance.  If the package was written by a member of
the R Development Core Team, or by the author of a prominent book on R,
it's more likely that package is as reliable as the base.

I can attest that maintaining custom, from-scratch R analytics as a
non-stats-guy is a challenging but rewarding activity.  It forces you to
understand the structure of your data and the interconnection of your
analysis processes.  So in the end, it might come down to preference.
If your job is to ensure the reliability of this system and vouch for
its accuracy, then perhaps R might pull you from your other
responsibilities.  However, if your job is to grow this system with your
organization's needs, then investing in R is a great choice: When you
learn R, R returns the favor and enlightens you about the data under
your care.

And don't forget that you can always just "try" R.  It might take some
bravery, but if you can take a dump/export/snapshot of some data and
process it via R, you'll learn quickly whether the transition would
require minor updates vs. major overhauls.  And here the good folks of
this list could probably help -- if you can provide a representation of
your data and a description of what operations are involved (or the
relevant S+ code, better yet), I think you'll find somebody with the
willingness to help and the expertise to advise.

There's a curious number of Fortune 500 domain names that appear in the
addresses of this list's regulars... I wonder how many of them got their
start by "trying" R when at a juncture similar to your own.

Finally, to answer your last question about consultants:  You might try
guru.com to look for freelance talent.  Also, if you can define your
project such that someone could bid, you could post an RFP to this list
(or send a general query for contract help at an hourly rate).  There
have also been postings to the list from professional training firms
that run workshops -- perhaps they could provide a customized solution
for your business.


HTH,

-Brian J. Koch
Data Manager
Decision Development Inc
(A qualitative marketing research firm.  Evanston, IL)


-----Original Message-----
From: r-help-bounces at stat.math.ethz.ch
[mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Dave
Sent: Thursday, July 27, 2006 11:45 AM
To: r-help at stat.math.ethz.ch
Subject: [R] Moving from Splus to R - advice/opinions request from
amanagement perspective

Hi,
   
  I've looked through the archives and seen several posts discussing
technical differences between R and S(plus).  It appears to me that R
can likely functionally replace Splus for my situation, but I'm more
interested in looking at the risks and benefits of moving from Splus to
R from a (project) management point of view.
   
  Background (a bit wordy, I'm afraid):
   
  - I'm not a stats guy, but rather a project manager responsible for an
internal application that utilizes Splus as a back-end analysis engine
to analyze manufacturing data at a med/large company.
   
  - I really don't know why Splus was chosen as the analysis engine for
this app - that choice was made long before I inherited the project.
   
  - The developer currently in charge of the Splus code is not a stats
guy either, but he is a very talented programmer that has been to one
Insightful course and taught himself enough to maintain the existing
code (the original authors are long gone).
   
  -  While there is quite a large quantity of existing code that is used
by our application, I don't believe that it is terribly complex, from an
applied statistics point of view.  Using Splus may be over-kill from a
functionality standpoint, but we just haven't had the time to re-write
in a more "appropriate" package - even if we knew what that more
appropriate package might be.
   
  -  With the imminent release of Splus 8, I'm feeling uncomfortable
with the risk associated with remaining on Splus 2000, which we are
currently using.  ***Comments on this point would be appreciated.***
   
  -  The developer was able to modify the existing code to run in Splus
7.  Unfortunately, the code is significantly slower than before, and
Insightful claims this is due to a change to the S language (at version
6) that was out of their control (and one reason that they bought the S
language rights).
   
  -  We have found Insightful's telephone support to be rather
unresponsive, and not very helpful, other than to recommend their
consulting services.  Obtaining consulting services from Insightful to
improve performance has proved challenging (don't ask), and if we ever
actually do receive any consulting services from them, it will no doubt
be quite expensive.
   
  (if you are still with me, thank you)
   
  I see our realistic options as:
   
  A)  Stick with Splus with the assumption that eventually, Insightful
will help us migrate our code to the latest release, and performance
will be comparable to what it is today.  The advantage of this is
working with a known entity, if not one we are very pleased with.  In
addition, if we can get the relationship to work, we can hopefully
"outsource" future statistics development and support to Insightful.  
   
  The disadvantage is that it is costing us quite a bit of $$$ to
maintain a relationship that we are not really happy with.
Incidentally, I'm not intending to bash Insightful here - it may just be
that we are not a good fit for each other.
   
  B)  Drop Splus for R with the assumption that migrating to R and
rewriting to improve performance will be little more difficult than
rewriting by ourselves in Splus.  The obvious advantage is the cost
savings, which is very significant.  In addition, based on the archives
I've read, it appears that there is a very responsive and helpful R
community.  
   
  The disadvantage is that we may be committing to maintaining
statistical expertise in-house, whereas we were hoping to be able to
outsource some or all of it (to Insightful).  In addition, we will be
leaving behind an entity that has a mailing address, phone number, and
stock symbol for one that is represented "only" by a mailing list - I'm
rather conservative and risk averse.
   
  C)  Stick with Splus and either find some consulting help outside of
Insightful or slog away internally.  I suppose that we could even end
our M&S contract with Insightful and just continue to use Splus, knowing
that we will never receive any releases newer than 8.x nor any future
support.
   
  (almost done)
   
  So, I'd like to hear opinions of whether you think that the benefits
of moving to R outweigh the risks.  Listing additional benefits and
risks that I have not identified would be appreciated.  I'm particularly
interested in hearing from anyone that has made this same S -> R
transition in a business environment.
   
  Also, does anyone have any recommendations for either Splus or R
consultants, other than Insightful?  I would prefer someone who is
willing to work remotely, rather than anyone having to travel.
   
  Thats it.
   
  Thanks for your time,
  Dave

 __________________________________________________



	[[alternative HTML version deleted]]

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.



More information about the R-help mailing list