Date: Wed, 6 Sep 2006 17:15:53 -0700
Reply-To: fnaqvi@GMAIL.COM
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: fnaqvi@GMAIL.COM
Organization: http://groups.google.com
Subject: Re: Enterprise miner problem
In-Reply-To: <BAY123-F27C5EA30FF846C75B08FBDDE310@phx.gbl>
Content-Type: text/plain; charset="iso-8859-1"
Hi
Thanks you all for a kind discussions. I haven't use STEPWISE selection
model. i have found the other ways.
Now come to the point, How do I calculate Gini for the tree and neural
network? (while tree has been developed using CHAID). I need to
calculate Gini for my training set, and 15 other test set, can I
automate this process in EM or manually?
Second, How do I clculate area under the ROC ( generated from
Enterprise miner)?
cheers
Syed
toby dunn wrote:
> David ,
>
>
> I never said that there were not better approaches. In fact I would
> advocate strongly not to use StepWise Reg. On the other hand many places
> refuse to change until there is sufficient forces to make them change. So a
> bad modeling practice just like bad coding practices or bad programs get
> carried forward until such time as there is enough force to invoke change.
> Ussually in business this comes down to money or conforming to state or
> federal regulations.
>
>
>
>
> Toby Dunn
>
> When everything is coming at you all at once, your in the wrong lane.
>
> A truly happy person is someone who can smile and enjoy the scenery on a
> detour.
>
>
>
>
>
> From: David L Cassell <davidlcassell@MSN.COM>
> Reply-To: David L Cassell <davidlcassell@MSN.COM>
> To: SAS-L@LISTSERV.UGA.EDU
> Subject: Re: Enterprise miner problem
> Date: Wed, 6 Sep 2006 14:21:44 -0700
>
> tobydunn@HOTMAIL.COM replied to Sig:
> >>I have to wonder who still believes strongly enough in step-wise methods
> >>to create demand for an easy method to invoke them? As an analogy for
> >>step-wise model selection methods, imagine a program that selects
> >>medical treatments for patients. Say it computes a cost-minimization
> >>criterion for ranking treatment plans, and that many medical specialists
> >>have reported that the program often selects inferior treatment plans.
> >>Would you use the program to select a treatment plan for your disease?
>
> >
> >Sig ,
> >
> >Why, it is the only known model that has been proven to even work some of
> >the times. I remember in economics some one ( I think Freeman ) says you
> >dont stop using a model until you have a better one. If in fact that model
> >does perform poorly enough then the model and its assumptions should be in
> >fact checked. If the business using said model loses enough mony or gets
> >wrapped up in lawsuites due to the model then it will change or the
> >business
> >will go, even slowly, out of business. So I guess what I am saying is that
> >it does not matter whether they believe in stepwise selection but rather do
> >they believe in the model created by stepwise methods enough to keep using
> >it. If they do till have faith in it then it hangs around, if not then
> >they
> >redo the model hopefully with someone who knows what they are doing this
> >time.
>
> BZZZTT! Wrong!
>
> But thank you for playing! We have some lovely parting gifts
> including a home version of our game...
>
> :-) :-)
>
> >Why, it is the only known model that has been proven to even work some of
> >the times.
>
> That is just not true.
>
> It was sort of true back in the 70' and 80's when we lacked
> computing power to do more complex approaches. But even then,
> the formulas were not statistically valid and the intermediate steps
> all needed to be checked to avoid Bad Things.
>
> Now, even if you are in a situation where you need such a model
> building approach, there are LASSO, LARS, ... And there are other
> model building approaches!
>
>
> Sig's point is correct. It's a bad thing to do, but too many people
> have yet to figure out that they should do something else. Remember
> Cassell's Paradigm Law: "It takes a minimum of 15 years to effect
> a paradigm shift in any field." So, as long as Old Guys sit around in
> classrooms and say in a quaking voice, "Back in my day, we ALL used
> stepwise selection. And an abacus..."
>
> David, who hiked to school through five feet of snow, uphill, each way...
> --
> David L. Cassell
> mathematical statistician
> Design Pathways
> 3115 NW Norwood Pl.
> Corvallis OR 97330
>
> _________________________________________________________________
> Check the weather nationwide with MSN Search: Try it now!
> http://search.msn.com/results.aspx?q=weather&FORM=WLMTAG
| 
