Skip to main content

%iKnow.Classification.Definition.ClassificationMethod

Class %iKnow.Classification.Definition.ClassificationMethod Extends node [ Deprecated, System = 3 ]

Properties​

method​

Property method As %String(VALUELIST = ",cosineSimilarity,euclideanDistance,naiveBayes,linearRegression,pmml,rules", XMLPROJECTION = "ATTRIBUTE") [ Required ];

This is the method the classifier will use to match the document vector against the different Categories. Depending on this method, the weights in the CategoryTermWeights vectors will represent different metrics.

localTermWeights​

Property localTermWeights As %String(VALUELIST = ",linear,binary,logarithmic", XMLPROJECTION = "ATTRIBUTE") [ InitialExpression = "binary", Required ];

The local term weight metric to use to build the Document Vector. The default setting (binary) means simple 0s and 1s will be used to identify if the term occurred in the document or not. Some values for method impose restrictions on this property.

normalization​

Property normalization As %String(VALUELIST = ",none,cosine", XMLPROJECTION = "ATTRIBUTE") [ InitialExpression = "none", Required ];

The normalization function to apply to the Document Vector. Some values for method impose restrictions on this property.

GlobalTermWeights​

Property GlobalTermWeights As %DeepSee.PMML.Definition.Util.Array;

Any global term weights for the terms in the TermDictionary, based on their corpus-wide relevance. Used to weigh the Document Vector. This property cannot be used when method = "naiveBayes".

CategoryTermWeights​

Property CategoryTermWeights As %DeepSee.PMML.Definition.Util.Matrix;

This matrix contains the term weights within the specific categories of this classifier when method is "cosineSimilarity", "euclideanDistance", "naiveBayes" or "linearRegression". Rows correspond positionally to the categories in the Categories element and columns to the terms in the TermDictionary, appended with any MetadataProperty elements. The meaning of the "weight" value is dependent on the classification method and might correspond to an actual weight (vector space), a probability (naiveBayes) or coefficient (regression)

CategoryWeights​

Property CategoryWeights As %DeepSee.PMML.Definition.Util.Array;

When method="naiveBayes", this array contains any a-priori probabilities of the categories, used to weigh the per-category term-based probabilities.

PMML​

Property PMML As %DeepSee.PMML.Definition.PMML;

If method="pmml", this element contains the PMML model to be used for categorization

ClassificationRules​

Property ClassificationRules As ClassificationRules;

If method="rules", this element contains the actual decision rules for deriving the category from the text input

Methods​

%Validate​

Method %Validate(ByRef pInfo) As %Status