Supported and Unsupported PMML Features and Attributes

Using External Models With Vertica gives an overview of the features Vertica supports for working with external models. This topic provides additional details on limitations in how Vertica supports working with PMML models.

With PMML models, Vertica currently supports only:

• PMML models that do not contain a data preprocessing step.
• PMML models that encode only these model types: kmeans, linear regression, logistic regression.

Supported and Unsupported PMML Attributes

The following table details the PMML attributes that Vertica currently does and does not support:

XML-tag name	ignored attributes	supported attributes	unsupported attributes	ignored sub-tags	supported sub-tags	unsupported sub-tags
Cluster	size	id,name,	-	KohonenMap, Covariances	NUM-ARRAY	Extension, Partition
ClusteringField	-	field(required), isCenterField(true is the only supported value), compareFunction	fieldWeight, similarityScale	-	-	Extension, Comparisons
ClusteringModel	modelName	functionName(required- clustering is only valid value), algorithmName, modelClass(required- only centerBased is supported), numberOfClusters(required), isScorable (true is the only supported value)	-	ModelVerification	MiningSchema, ComparisonMeasure, ClusteringField, Cluster	Extension, Output, ModelStats,ModelExplanation, LocalTransformations, MissingValueWeights, ModelVerification
ComparisonMeasure	minimum, maximum	kind(required- only distance is supported), compareFunction	-	-	euclidean, squaredEuclidean	Extension, chebychev, cityBlock, minkowski, simpleMatching, jaccard, tanimoto, binarySimilarity
DataDictionary	-	numberOfFields	-	-	DataField	Extension, Taxonomy
DataField	displayName	name(required), optype(required), dataType(required)	taxonomy, isCyclic	-	-	Extension, Interval, Value
Header	copyright, description, modelVersion	-	-	Extension, Application, Annotation, Timestamp		-
MiningField	importance, missingValueTreatment	name(required), usageType, optype,	outliers, lowValue, highValue, missingValueReplacement, invalidValueTreatment	-	-	Extension
MiningSchema	-	-	-	-	MiningField	Extension
NumericPredictor	-	name(required), exponent, coefficient(required)	-	-	-	Extension
PMML	-	version(required), xmlns	-	MiningBuildTask	Header, DataDictionary, ClusteringModel, RegressionModel	TransformationDictionary, Extension, any unsupported model type
RegressionModel	modelName, targetFieldName, modelType	functionName(required - can be regression or classification), algorithmName, normalizationMethod, isScorable (true is the only supported value)	-	ModelVerification	MiningSchema, RegressionTable	Extension, Output, ModelStats, ModelExplanation, LocalTransformations, Targets, ModelVerification
RegressionTable	-	intercept(required), targetCategory	-	-	NumericPredictor	Extension, CategoricalPredictor, PredictorTerm