Date:         Wed, 31 Jan 2001 21:58:00 +0800
Reply-To:     Calvin Huang-CEO of UITRA <wong_ahpo@HKICABLE.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Calvin Huang-CEO of UITRA <wong_ahpo@HKICABLE.COM>
Organization: Another Netscape Collabra Server User
Subject:      Re: assessing scored data in EM: comparison of predicted and
              actual target

Hi,
I think what you what is the 'confusion matrix' in result of the model
nodes.

simply go to the result of model node,
then go to the 'output' tab,
look at the very last part,
then you will find the condfusion matrix in this form

                                  TABLE OF TARGET BY OUTPUT

                             TARGET     OUTPUT

                             Frequency|
                             Percent  |
                             Row Pct  |
                             Col Pct  |0       |1       |  Total
                             ---------+--------+--------+
                             0        |    102 |     45 |    147
                                      |  34.00 |  15.00 |  49.00
                                      |  69.39 |  30.61 |
                                      |  76.12 |  27.11 |
                             ---------+--------+--------+
                             1        |     32 |    121 |    153
                                      |  10.67 |  40.33 |  51.00
                                      |  20.92 |  79.08 |
                                      |  23.88 |  72.89 |
                             ---------+--------+--------+
                             Total         134      166      300
                                         44.67    55.33   100.00

my result is extracted from the html report generated by 'Reporter' Node and
is for validation data.

You may see a similar one in the 'Output' tab, but the matrix is for the
training data.

Calvin Huang
Data Miner

<anna418@my-deja.com> wrote in message news:9593kp$821$1@nnrp1.deja.com...
> Hi,
>
> could anyone help me with this problem:
>
> I'm working with the Enterprise Miner and built a neural network, tree
> and logistic regression model. The models were built with a training
> and validation data set; the "Test" field in the partition node was set
> to 0, as the training data is oversampled.
> The original data was then scored in the score node.
>
> My question:
>
> The results can be assessed in the assessment node. But how can I
> assess the results of the scored data set? It is not possible to put an
> assessment node after the score node. I would like to have the
> comparison of the actual and predicted values in a table or chart
> (alpha and beta mistakes).
>
> Thanks very much!
>
>
>
> Sent via Deja.com
> http://www.deja.com/