About WendyCzika

WendyCzika · ‎05-25-2018

You can find the formula for the BIC on page 4 of this paper: Building Bayesian Network Classifiers Using the HPBNET Procedure Note that it will always be negative for a discrete probability distribution, and the higher the BIC, the better the model. Hope that helps.

WendyCzika · ‎05-25-2018

One thing you could try is the TwoStage model node. You need to define 2 target variables for that: 1 as a binary (bought or not), and the other represents the number of units for those that bought (missing for those that didn't). Then you can do a Sequential model using Filter=Non-Events (need to define the event for the binary target as those that did not buy).

WendyCzika · ‎05-25-2018

I think your only options are to manually change the Role of that variable back to Input in the Variable Role selector on the Groupings tab or in the New Role column on the Variables tab of the Interactive editor, or you can either turn off variable selection or modify the IV cutoff then re-run the node with Use Frozen Groupings=Yes.

WendyCzika · ‎05-18-2018

If you are just running the Regression and Scorecard nodes after the Interactive Grouping node with defaults, then one thing you might want to check are the inputs used in the Regression node - by default, it is using both the GRP_ and the WOE_ variables for each input whereas the Scorecard node is only using the WOE_ variables. When I set Use=No for the GRP_ variables in the variables editor for the Regression node, then my assessment results matched.

WendyCzika · ‎05-17-2018

Since it looks like EM doesn't calculate that measure, you could do the following in a SAS Code node after the Model Comparison node (not sure if this is the best way, but I think it should work): data mae; set &em_import_data end=last; /* you can use another partition here: &em_import_validate or &em_import_test */ retain sae nobs 0; sae + abs(R_target); /* R_target is the residual. Put the name of your target in for target */ nobs + 1; if last then do; mae = sae / nobs; output; end; run; proc print data=mae; run;

WendyCzika · ‎05-15-2018

I think the replies in this post might help you: Clustering in SAS Miner: Number of clusters determination, input data, and results interpretation

WendyCzika · ‎05-15-2018

It's: emloopnumber Hope that helps! Wendy

WendyCzika · ‎05-09-2018

Do you have other results from the DMREG procedure in your Output window?

WendyCzika · ‎05-09-2018

Sorry, I wasn't including the bias terms in the number of parameter estimates. You would have one of those for each hidden and output node as well.

WendyCzika · ‎05-09-2018

By default, the Neural Network node uses 3 hidden units/neurons in the hidden layer, but you can change this by clicking on the ellipsis next to the "Network" property and setting the Number of Hidden Units. If there are 10 hidden units, you would have [(# of inputs) * 10 + (# of targets) * 10] parameter estimates (1 estimate or weight for each link from input to hidden nodes, and hidden to output nodes).

WendyCzika · ‎05-09-2018

Yes, hidden units and hidden neurons are just different terms for the same thing. The Neural Network node uses 1 hidden layer. You can use the HP Neural node to include up to 10 hidden layers.

WendyCzika · ‎05-02-2018

I'm sorry to hear the group processing is so slow - not sure why that would be. But you should be able to fairly easily compare your PROC REG model using a SAS Code node in EM and feed that into a Model Comparison node to compare with your other EM models. Just be sure to include this line in your PROC REG call so it generates the needed flow score code to enable the assessment: code file="&EM_FILE_EMFLOWSCORECODE"; And set the Tool Type property for the SAS Code node to "Model". Then you can connect that to a Model Comparison node along with your other EM models. This is some sample code I used in SAS Code node (assuming you have an interval target): proc sort data=&em_import_data out=temp; by <Your-by-var>; run; proc reg data=temp; model %em_target = %em_interval_input; by <Your-by-var>; code file="&EM_FILE_EMFLOWSCORECODE"; run;

WendyCzika · ‎05-02-2018

I'm not sure that can be done within the group processing framework unfortunately...I think you would need to set up a separate flow for each stratum in order to do this, which I know is not ideal. But will let you know if we come up with an better way to do this.

WendyCzika · ‎04-26-2018

That should work... does your target variable have the role of Target in your Input Data node? Either way, you can try putting a Metadata node after the Transform Variables node and be sure your transformed target has role of Target there, then use your Regression node.

WendyCzika · ‎04-24-2018

If you are using the Principal Components node from the Modify tab, then in order to include class inputs, you would need to dummy encode them using the Transform Variables node first as you said. In the HP Principal Components node however, there is a property for it to do the class level encoding for you. In both nodes, you can set the Eigenvalue Source to Correlation to standardize the inputs. And also in both, any observations with missing values for any of the inputs are excluded, so you would need to first run the Impute node to impute missing values.

Online Status	Offline
Date Last Visited	‎08-07-2025 02:41 PM