Re: DM: Boosting in Data Mining Products

[Ronny Kohavi]

> " Silicon Graphics is the first major vendor to offer a
> commercial implementation of boosting"

[See http://mineset.sgi.com for full press release]

Dan Steinberg wrote:
Dan> Just wanted to point out that Salford Systems's decision tree
Dan> CARTŪ (tm) has included Bootstrap Aggregation (bagging) since
Dan> September 1995 and ARCing (Leo Breiman's version of boosting
Dan> --Adaptive Resampling and Combining) since April 1997 (on UNIX
Dan> platforms).

Let me clarify this issue.  The press release said "major vendor."  To
our knowledge, no major vendor has Boosting.  If you are interested in
a recent IDC report showing major vendors, look at the following web
page:
    http://www.idc.com/telb/madmrch/sld019.htm
which lists data minding vendors that IDC estimates have over 2%
mindshare each.  These include: IBM, SAS, Silicon Graphics, SPSS,
Pilot, and Thinking Machines.  All other vendors have combined share
of about 12%.  

We are aware that several existing academic and commercial tools have
an implmentation of Boosting, including for example C5.0, which is
also used in ISL's Clementine as Colin mentioned.  We ourselves have
released MLC++ last year with Boosting implemented (in source code for
research purposes).  See http://www.sgi.com/Technology/mlc for
details.

--

   Ronny Kohavi (ronnyk@sgi.com, http://robotics.stanford.edu/~ronnyk)
   Engineering Manager, MineSet.
   Maximize the value of your data with data mining and visualization.