I want to a produce candidate models list from Logistics.
However, Selection option (None, Forward, Backward, Stepwise, Score)
can only produce one best model only.

For example, I want require models with ROC GT 70%. How to produce the
candidate list that meets my requirements?

On Aug 16, 2:22=A0am, "litonc...@gmail.com" <litonc...@gmail.com> wrote:
> I want to a produce candidate models list from Logistics.
> However, Selection option (None, Forward, Backward, Stepwise, Score)
> can only produce one best model only.
>
> For example, I want require models with ROC GT 70%. How to produce the
> candidate list that meets my requirements?

I think you're going to need to explore using the ods tables.

Using sample data from sas documentation and the ods table statement
as follows gets you the model for each iteration.
You'll need to figure out how to get the ROC into a similar table and
then merge the results together to get all models that meet the
condition you've set.

HTH,
Reeza

ods trace on;
*The model dataset will hold the estimates;
ods table parameterestimates=3Dmodels;


  title 'Stepwise Regression on Cancer Remission Data';
   proc logistic data=3DRemission outest=3Dbetas covout;
      model remiss(event=3D'1')=3Dcell smear infil li blast temp
                   / selection=3Dstepwise
                     slentry=3D0.3
                     slstay=3D0.35
                     details
                     lackfit;
      output out=3Dpred p=3Dphat lower=3Dlcl upper=3Ducl
             predprob=3D(individual crossvalidate);
   run;
ods trace off;

On 8=E6=9C=8817=E6=97=A5, =E4=B8=8A=E5=8D=889=E6=99=8240=E5=88=86, Reeza <f=
khurs...@gmail.com> wrote:
> On Aug 16, 2:22=C2=A0am, "litonc...@gmail.com" <litonc...@gmail.com> wrot=
e:
Thank you for your replying.
I know how to use trace on to record relevant statics from each
regression
But there are lots iterations.
For examples, I have 100 variables and the best number in a mode is
between 5~15.
It means:
1. I have to run 11 times (from 5~15 variable combination) by myself.
i=3D1;
  j=3Di;
   k=3D j;
     l=3Dk;
       m=3Dl;

2.It take lots of time to iteration
5 variables out of 100, there will be (100*99*98*97*96)/
(1*2*3*4*5)=3D75,287,520 combinations
15 variables out of 100, there will be COMBIN(100,15)...a horrible
number


-----------------
I don't know if there any solution in SAS to solve the problem?

On Aug 16, 7:21=C2=A0pm, "litonc...@gmail.com" <litonc...@gmail.com> wrote:
> On 8=E6=9C=8817=E6=97=A5, =E4=B8=8A=E5=8D=889=E6=99=8240=E5=88=86, Reeza =
<fkhurs...@gmail.com> wrote:> On Aug 16, 2:22=C2=A0am, "litonc...@gmail.com=
" <litonc...@gmail.com> wrote:
>
> Thank you for your replying.
> I know how to use trace on to record relevant statics from each
> regression
> But there are lots iterations.
> For examples, I have 100 variables and the best number in a mode is
> between 5~15.
> It means:
> 1. I have to run 11 times (from 5~15 variable combination) by myself.
> i=3D1;
> =C2=A0 j=3Di;
> =C2=A0 =C2=A0k=3D j;
> =C2=A0 =C2=A0 =C2=A0l=3Dk;
> =C2=A0 =C2=A0 =C2=A0 =C2=A0m=3Dl;
>
> 2.It take lots of time to iteration
> 5 variables out of 100, there will be (100*99*98*97*96)/
> (1*2*3*4*5)=3D75,287,520 combinations
> 15 variables out of 100, there will be COMBIN(100,15)...a horrible
> number
>
> -----------------
> I don't know if there any solution in SAS to solve the problem?

I'm confused. The ods output table stores all the models tested in the
stepwise regression, that was your original question, why does the
number of variables have any effect on this?
I don't understand your question anymore.