Date:         Tue, 5 Sep 2006 18:06:21 -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
Comments: To: sas-l@uga.edu
In-Reply-To:  <CA8F89971ADA9F47A6C915BA2397844201EA743B@MAILBE2.westat.com>
Content-Type: text/plain; charset="iso-8859-1"

Hi

thnak you for your reply. I don't understant how do i find 'stepwise
selection' (except logistic regression) in enterprise miner GUI. there
are few more problems I have during the model developement which are:

A. for this problem I have made a sample of 150,000 records with 80
variables and saved it on a hard disk. Now the problem is when I build
the tree then it only shows two 1 node with two leaves ( while responce
has (34,000 (0) and 80,000 (1) and some missing data). I don't
understant how to create tree for this model.

B. With the similar dataset, I also tried to build Neural network and
it doesnot shows any result.

I am wondering,  I may be failed to develope such a model. Suggest some
way to get out from this point.

cheer's
Syed

Sigurd Hermansen wrote:
> Syed:
> Sounds like a good opportunity for you to analyze what is happening and =
> find a good solution....
> =20
> I'd look first at obvious bottlenecks. Is EM trying to transfer millions =
> of rows in a table on a server to another node on a network. Network =
> transfer speeds will slow things down. Are you following the dubious =
> practice of searching very large numbers of predictors for a relevant =
> model? (If so, search the SAS-L Archives for postings on 'step-wise =
> selection'.) Moving large volumes of data takes time.
> =20
> A good 150,000 row sample of a larger database should give you good =
> estimates of all but very small proportions. The key word is 'good'. If =
> you know little about your data source and choose a sample with an =
> underrepresentation of an important subgroup, you could minimize the =
> importance of the subgroup and find a good fit for a bad model.
> =20
> EM and all other automated data mining programs may facilitate modelling =
> but, as with all statistical packages, also make it too easy to go =
> astray. Don't expect easy answers.
> =20
> Remember to train models on one sample, test models on another =
> independent sample, and, after selecting a specific model, evaluate its =
> performance on one or more samples not used in training and testing. You =
> have enough data to implement that strategy.
> =20
> Best wishes for exciting challenges and success in your academic work.
> Sig=20
>
> ________________________________
>
> From: owner-sas-l@listserv.uga.edu on behalf of fnaqvi@gmail.com
> Sent: Tue 9/5/2006 7:51 AM
> To: sas-l@uga.edu
> Subject: Enterprise miner problem
>
>
>
> Hi
>
> I am a student and working on a project. I have enterprise miner and
> working on a on a dataset (there is millions of records) which is on
> server. The problem is when i tried to connect data with in "input data
> node", the computer and SAS software become very slow. I had wait about
> more than 30 minutes to get responce.  I have taken a sample from the
> same dataset (more than 150,000 records) and saved on local drive and
> it works properly. If this problem persis, is there any solution
> without using enterprise miner GUI?
>
>
>
> Regards
>
> Syed