topic Re: using neural nets and now HPNEURAL in SAS Data Science

using neural nets

rboire — Tue, 09 May 2017 14:11:16 GMT

here is some code I ran which is confusing to me based on the log errors. yet age is on the dataset

4234 proc neural data=lib2.devcorrallvar_mar30
4235 dmdbcat=lib2.devcorrallvar_mar30
4236 random=789;
4237      input age joingrouptenure/ level=interval id=memberno1;
ERROR: The data LIB2.DEVCORRALLVAR_MAR30 does not contain the variable age.
The node memberno1 is flagged as a bad node.
4238   target response / id=o level=nominal;
4239
4240       hidden 3 / id=h;
4241       prelim 5;
ERROR: Cannot construct the network.
ERROR: Cannot train this network.
4242
4243         train;
ERROR: Cannot construct the network.
ERROR: Cannot train this network.
4244
4245         score out=out outfit=fit;
ERROR: The weights and the network have not yet been initialized.
4246      score data= lib2.valcorrallvar_mar30   out=gridout;   title 'MLP with 3 Hidden Units';
ERROR: The weights and the network have not yet been initialized.
4246! run;

here is a proc print of that dataset which contains age.

Obs AGE JOINGROUPTENURE

                                    1     0            0
                                    2     0            0
                                    3     0            0
                                    4     0            0
                                    5     0            0
                                    6     0            0
                                    7     0            0
                                    8     0            0
                                    9     0            0
                                   10     0            0

Re: using neural nets

WendyCzika — Tue, 09 May 2017 14:43:19 GMT

You need to run PROC DMDB to create the input data set and catalog for PROC NEURAL (sorry this step was omitted in the post that I included in your other post). Here is an example:

*** CREATE DATA MINING DATABASE WITH PROC DMDB - REQUIRED FOR PROC NEURAL;
proc dmdb data=lib2.devcorrallvar_mar30 out=dmdbout_dca_mar30 dmdbcat=dmdbcat_dca_mar30;
var age joingrouptenure;

class response;
target response;
run;

*** RUN PROC NEURAL;
proc neural data=dmdbout_dca_mar30 dmdbcat=dmdbcat_dca_mar30 ...;

*** the rest of your code below... ;

Re: using neural nets and now HPNEURAL

rboire — Tue, 09 May 2017 17:03:43 GMT

that's great for neural nets which now works but am now trying HPNEURAL

because we can have training and testing through the partition statements and train statements.See below code

proc hpneural data=lib2.devcorrallvar_mar30;
input age joingrouptenure PAYCASHDRAWEROTHR
LTVGROUPDESC1
TENURE
etc;
target response/ level=nom;
hidden 2;
partition fraction (train=0.6);
train outmodel=model_assoc maxiter=1000;
run;

Above code is fine. Now I try running the scoring code on a validation

proc hpneural data=valid;
score model=model_assoc out=valid1 ; run;

run;

Here is my output:

The MEANS Procedure

Variable     Label                      N          Mean       Std Dev       Minimum       Maximum
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
P_response0 Predicted: response=0 25300     0.9447408     0.0310369     0.8167701     0.9708281
P_response1 Predicted: response=1 25300     0.0552592     0.0310369     0.0291719     0.1832299
ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ

There is no variable for the actual observed behaviour of response in VALID1. Can you please advise and by the way thanks for your support.

Re: using neural nets and now HPNEURAL

WendyCzika — Wed, 10 May 2017 13:07:00 GMT

If I'm understanding your issue correctly, I think you just need to add an ID statement to PROC HPNEURAL to list the variables you want to keep in the OUT= data set (response, e,g,).