Changes from PMML 2.1
PMML3.0 Menu

Home


PMML Notice and License

Changes


Conformance

General Structure

Header

Data
Dictionary


Mining
Schema


Transformations

Statistics

Taxomony

Targets

Output

Functions

Built-in Functions

Model Composition

Model Verification


Association Rules

Cluster
Models


General
Regression


Naive
Bayes


Neural
Network


Regression

Ruleset

Sequences

Text Models

Trees

Vector Machine

PMML 3.0 - Changes from PMML 2.1

Changes which apply to all models, e.g., adding the LocalTransformations element to all models, are listed only once in the respective document (in this case Transformations), not in each document where it applies. Also, only essential changes are listed. E.g., further explanations which were added are not listed.
Hence, if a document is listed with 'no change', this does not imply that both, the .html and the XSD, have not been changed.

AssociationRules

  • allow at most one DerivedField in the TransformationDictionary
  • allow mixed sequence of Itemsets and AssociationRules, but require precedence of Itemsets before the respective AssociationRules
  • attribute numberOfItems is now a non-negative number
  • require unique ids for Items as well as the attribute value in Item
  • add attributes lift and id to element AssociationRule

BuiltinFunctions

Introduced with PMML 3.0.

ClusteringModel

  • added new element MissingValueWeights
  • require attribute fieldWeight in element ClusteringField to be >0
  • moved element Matrix over to GeneralStructure

Conformance

No changes.

DataDictionary

  • added the following values to attribute dataType in element DataDictionary:
    • date
    • time
    • dateTime
    • dateDaysSince[0]
    • dateDaysSince[1960]
    • dateDaysSince[1970]
    • dateDaysSince[1980]
    • timeSeconds
    • dateTimeSecondsSince[0]
    • dateTimeSecondsSince[1960]
    • dateTimeSecondsSince[1970]
    • dateTimeSecondsSince[1980]

Functions

Introduced with PMML 3.0.

GeneralRegression

  • made attribute targetVariableName in GeneralRegressionModel optional
  • removed value logLinear from attribute modelType in element GeneralRegressionModel
  • added example and scoring algorithm for General Linear Regression

GeneralStructure

  • clarified that field names, as well as all other elements of PMML, are case sensitive
  • include element Extension in all elements and groups as first item
  • added elements Output, Targets and ModelVerification to all predictive models
  • added element LocalTransformations to all models
  • attribute extensions via x- are deprecated but still tolerated in PMML 3.0 - use element Extension instead
  • clarified that data types NUMBER and REAL-NUMBER do not allow the symbols INF, -INF and NaN
  • required attribute type in element Array
  • added new elements INT-SparseArray and REAL-SparseArray
  • added element Matrix

Header

No changes.

MiningSchema

  • clarified that fields listed in MiningSchema must be unique
  • added attributes optype and importance to element MiningField
  • introduced new value order for attribute usageType in element MiningField

ModelComposition

Introduced with PMML 3.0.

ModelVerification

Introduced with PMML 3.0.

NaiveBayes

No changes.

NeuralNetwork

  • added activation functions cosine and radialBasis
  • added default value 0 for attribute threshold in element NeuralNetwork
  • added attribute normalizationMethod to element NeuralNetwork
  • added attributes threshold, width and altitude to elements NeuralNetwork and NeuralLayer consistently
  • removed attributes threshold and altitude from element Neuron

Output

Introduced with PMML 3.0.

Regression

  • attributes modelType and targetFieldName in element RegressionModel is now optional
  • added new normalization methods logit, probit, cloglog and exp
  • added default value 1 for attribute exponent in element NumericPredictor
  • introduced interaction terms by adding new element PredictorTerm to element RegressionTable
  • removed attribute mean from element NumericPredictor

RuleSet

Introduced with PMML 3.0.

Sequence

No changes.

Statistics

No changes.

SupportVectorMachine

Introduced with PMML 3.0.

Targets

Introduced with PMML 3.0.

Taxonomy

No changes.

Text

Introduced with PMML 3.0.

Transformations

  • incorporated new elements DefineFunction and Apply to support built-in and self-defined functions
  • introduced new element LocalTransformations to hold all transformations that are local to a model
  • added elements Interval and Value to element DerivedField, to define valid result value ranges
  • added attributes optype and dataType to element DerivedField
  • required attribute name in element DerivedField in case it is featured in TransformationDictionary or LocalTransformations
  • redefined element Constant to have the value in the content
  • added attribute mapMissingTo to elements NormDiscrete and MapValues to handle missing value
  • added attributes mapMissingTo and defaultValue to element Discretize

TreeModel

  • required that element Node can have either a sequence of elements ScoreDistribution and Node, or the element EmbeddedModel, not both at the same time
  • added element Embedded to element Node to support model composition

e-mail info at dmg.org