Date: Wed, 7 Jan 2009 09:46:06 -0800
Reply-To: stringplayer_2@yahoo.com
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dale McLerran <stringplayer_2@YAHOO.COM>
Subject: Re: R
In-Reply-To: <002e01c970bc$b08182d0$11848870$@com>
Content-Type: text/plain; charset=us-ascii
I have long thought SAS was giving up major ground to other
languages because of the inability to easily integrate matrix
manipulations with other pieces of their analytic modules.
The procedure/data step boundary makes SAS a relatively simple
language to learn, but reduces the extensibility of SAS. I
have lamented that state for some time, and have posted on that
more than once.
Apparently, SAS has been listening to that argument (or, likely,
they have had the same observation themselves.) SAS has
released a couple of improvements that provide some integration
of matrix processing with procedural code. These include:
1) IML Workshop which allows one to invoke any procedure as
well as data step code directly from IML (where IML is SAS's
matrix language application), and 2) the new FCMP procedure
contains some matrix manipulation modules which can be invoked
during execution of procedures which allow data-step-type
programming statements (e.g., NLMIXED, GENMOD, CALIS). The
FCMP procedure itself allows one to write and save functions
which can be invoked from data step code or from those
procedures which allow data-step-type programming statements.
There are certainly limitations to the functionality offered
by these improvements. IML Workshop is only available for
a Windows platform. Also, in order to take full advantage of
IML workshop, one must learn some new syntax. Further, IML
is a separate product requiring separate license. All those
serve to limit the usefulness of IML Workshop.
The matrix manipulations offered with FCMP are also limited.
The matrix code that SAS has made available through FCMP is
limited to matrix addition (and subtraction), multiplication,
inversion of a full rank matrix, computing the determinant of
a matrix, and a few other simple operations. There is no
functionality offered for computing eigenvalues and eigenvectors
or for returning a generalized inverse of a matrix. (However,
see my post from 12/21/08 where I use the FCMP procedure to
write and store a function which returns a generalized inverse:
http://listserv.uga.edu/cgi-bin/wa?A2=ind0812C&L=sas-l&P=R35830 )
Languages like R which are built with matrix manipulations
available everywhere and which do not have the data step boundary
do have a leg up on SAS for implementation of cutting edge
statistical applications. But SAS does have some capability
in this area that many are not aware of. SAS does have the
advantage of being much simpler to use for most standard
applications.
Dale
---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@NO_SPAMfhcrc.org
Ph: (206) 667-2926
Fax: (206) 667-5977
---------------------------------------
--- On Wed, 1/7/09, Murphy <goladin@GMAIL.COM> wrote:
> From: Murphy <goladin@GMAIL.COM>
> Subject: Re: R
> To: SAS-L@LISTSERV.UGA.EDU
> Date: Wednesday, January 7, 2009, 3:39 AM
> Hi Peter,
>
> I agree with on those aspects. One of the nicest thing
> about R is its extremely beautiful graphics which captivate
> more audiences of mine then a couple of SAS numbers. The
> other thing about R is the wide array of statistical models
> which are very useful in a variety of fields. If SAS did not
> introduce GLMSELECT procedure, my favorite LASSO can only be
> used on R and not SAS. I hope that SAS will start doing in
> depth research in converting cutting edge statistics into
> usable form for the SAS users. While it is possible for some
> of us to program those models, it is rather inefficient
> without the use of matrix.
>
> Two of my greatest unhappiness with SAS are the lack of
> decision tree models and neural network models in base SAS.
> I understand that SAS enterprise miner provides these 2
> types of model but the software cost a bombshell for normal
> people to use. Warren Sarle has programmed the neural
> network macro in SAS before but it is now not updated to the
> current SAS version with all the new whistle and bells. I
> have yet to see a functional decision tree SAS macro online.
> I am hoping that some experts will take the time to create
> all these macros and put them online for SAS users to use.
>
> In terms of documentations, I realize that it is just a
> matter of opinion. From what I have tried, I cannot see
> whether the quality of documentation is very much different.
> They usually have nice explanations to guide us poor souls.
> But I must admit that there are occasions where I cannot
> understand instructions from both SAS and R. While the
> problem of poor documentation is less in SAS, R is
> respectable in terms of its documentation quality given its
> open source status.
>
> Regards,
> Murphy Choy
> Certified Base SAS Programmer for SAS V9
> Certified Advanced SAS Programmer for SAS V9
> Datashaping Certified SAS Professional
>
> -----Original Message-----
> From: Peter Flom
> [mailto:peterflomconsulting@mindspring.com]
> Sent: Wednesday, January 07, 2009 7:14 PM
> To: Alex Murphy; SAS-L@LISTSERV.UGA.EDU
> Subject: Re: R
>
> Alex Murphy <goladin@GMAIL.COM> wrote
> >As far as I know, most statisticians in colleges now
> use R more often than
> >SAS for only 2 reasons.
> >
> > - Free license
> > - Matrix programming
> >
> >I have asked my alma mater about the lack of SAS in use
> for research. Most
> >of my old professors pointed to the fact that some of
> the most complicated
> >models now requires the use of matrix manipulation to
> be effectively coded.
> >However, almost all of them pointed out the fact that I
> could implement
> >almost every single model in SAS if I am given all the
> modules that SAS has.
> >My good friends in Finance also indicated that SAS is
> used more than R in
> >normal statistical analysis and data management. All of
> them have indicated
> >the fact that R is extremely powerful in data mining
> and advanced statistics
> >but difficult to be used in heavy data management work.
> >
>
> I use both SAS and R. I like SAS better, I find it much
> more user friendly. That said, there are other ways R is a
> better tool than SAS, some are:
>
> Cutting edge statistics
> Open source code
> Better default graphics, and graphs that are more easily
> modified (although with 9.2 SAS is moving in the right
> direction)
>
> Reasons to prefer SAS (beyond what's been mentioned)
>
> Better documentation
> Tech support
>
>
> Peter
>
>
> Peter L. Flom, PhD
> Statistical Consultant
> www DOT peterflom DOT com
| 
