Strange, it's usually the other way around.  By default (with the Basic Advisor), any numeric variable should be set to Interval.  If you use the Advanced Advisor, then if the number of levels for a numeric variable is less than 20 (by default), then it is set as Nominal, so not sure how a variable with that many levels is being set as Nominal (unless it is being read in as character instead of numeric).   If nothing else, you can use a Metadata node (from Utility tab) to change the level of the variable that you want to be interval. ... View more

If you use the HP BN Classifier node in Enterprise Miner, you will automatically get a graphical representation of the results.   Or in SAS code you can use the following code to use the SAS Graph Template Language for plotting the network  -- this code will be included in the online proceedings for SAS Global Forum 2017 for this presentation on the HPBNET procedure: https://event.crowdcompass.com/sgf2017-sessions/activity/kKxLYK6nMm (Building Bayesian Network Classifiers Using the HPBNET Procedure).  Hope it helps!     /* After running PROC HPBNET, to help visualize the final model, you can view the selected Bayesian network structure by using the following macro createBNCdiagram, which takes the target variable and the output network data as arguments. */ %macro createBNCdiagram(target=Class, outnetwork=net); data outstruct; set &outnetwork; if strip(upcase(_TYPE_)) eq 'STRUCTURE' then output; keep _nodeid_ _childnode_ _parentnode_; run; data networklink; set outstruct; linkid = _N_; label linkid ="Link ID"; run; proc sql; create table work._node1 as select distinct _CHILDNODE_ as node from networklink; create table work._node2 as select distinct _PARENTNODE_ as node from networklink; quit; proc sql; create table work._node as select node from work._node1 UNION select node from work._node2; quit; data bnc_networknode; length NodeType $32.; set work._node; if strip(upcase(node)) eq strip(upcase("&target")) then do; NodeType = "TARGET"; NodeColor=2; end; else do; NodeType = "INPUT"; NodeColor = 1; end; label NodeType ="Node Type" ; label NodeColor ="Node Color" ; run; data parents(rename=(_parentnode_ = _node_)) children(rename=(_childnode_ = _node_)) links; length _parentnode_ _childnode_ $ 32; set networklink; keep _parentnode_ _childnode_ ; run; *get list of all unique nodes; data nodes; set parents children; run; proc sort data=nodes; by _node_; run; data nodes; set nodes; by _node_; if first._node_; _Parentnode_ = _node_; _childnode_ = ""; run; /*merge node color and type */ data nodes; merge nodes bnc_ networknode (rename=(node=_node_ nodeColor=_nodeColor_ nodeType=_nodeType_)); by _node_; run; /*sort color values to ensure a consistent color mapping across networks */ /*note that the color mapping is HTML style dependent though */ proc sort data=nodes; by _nodeType_; run; *combine nodes and links; * need outsummaryall for model report; data bnc_networksummary(drop=_shape_ _nodecolor_ _nodepriority_ _shape_ _nodeID_ _nodetype_ _linkdirection_) bnc_networksummaryall; length _parentnode_ _childnode_ $ 32; set nodes links; drop _node_; if _childnode_ EQ "" thendo; _nodeID_ = _parentnode_; _nodepriority_ = 1; _shape_= "OVAL"; end; else do; _linkdirection_ = "TO"; output bnc_networksummary; end; output bnc_networksummaryall; label _linkdirection_="Link Direction"; run; proc datasets lib=work nolist nowarn; delete _node _node1 _node2 nodes links parents children; run; quit; proc template; define statgraph bpath; begingraph / DesignHeight=720 DesignWidth=720; entrytitle "Bayesian Network Diagram"; layout region; pathdiagram fromid=_parentnode_ toid=_childnode_ / arrangement=GRIP nodeid=_nodeid_ nodetitle=_nodeID_ nodeshape=_shape_ nodepriority=_nodepriority_ linkdirection=_linkdirection_ nodeColorGroup=_NodeColor_ textSizeMin = 10 ; endlayout; endgraph; end; run; ods graphics; proc sgrender data=bnc_networksummaryall template=bpath; run; %mend; %createBNCdiagram;     ... View more

What you have for your test data set is what Enterprise Miner considers a "score" data set.  To get predictions for that data set, you can connect both the Input Data node for that data set (with the Role property set to Score) and the modeling node that you want to use for your predictions (that uses your training data) to a Score node, as in the attached screenshot of a sample flow.  The Score node will apply the score code from the model to your score data set that doesn't contain the target.  Hope that helps!   ... View more

No, the only Enterprise Miner nodes that support generation of PMML are: Association, Cluster, Regression, Dmine Regression, Decision Tree, Neural Network, and AutoNeural.  It can't be generated for an imported model like that. ... View more

SAS Enterprise Miner High Performance (HP) data mining nodes have been specially developed to take advantage of parallel processing in both multithreaded single-machine mode and distributed multiple-machine mode. This is useful with large data sets and data stores.  Note that these nodes aren't necessarily a full replacement for their non-HP counterparts, and that there are some nodes that only exist in HPDM, for example: HP Bayesian Network Classifier, HP Forest, and HP GLM.  Here is information about the individual procedures:  http://go.documentation.sas.com/?docsetId=emhpprcref&docsetVersion=14.2&docsetTarget=titlepage.htm&locale=en And see the Enterprise Miner Reference Help for more informationa about the individual HPDM nodes. ... View more

For your first question, when using the "Basic Advisor" in Enterprise Miner, it is just assigning numeric variables to Interval and character variables to Nominal.  With the "Advanced Advisor", the assignment for numeric variables is based on the number of levels the user specifies (20 by default, so numeric variables with < 20 levels would be set to Nominal and >=20 would be Interval). ... View more

You can try using a smaller value for the Leaf Fraction property to possibly improve your model.  There are also different options for handling missing values, so try different options for the Missing Values property.  If you have much more than 20K observations in your training data, you can also try increasing the Node Sample property. ... View more