PredictCaseLikelihood (DMX)

Syntax

PredictCaseLikelihood([NORMALIZED|NONNORMALIZED])

Arguments

• NORMALIZED
  Return value contains the probability of the case within the model divided by the probability of the case without the model.

• NONNORMALIZED
  Return value contains the raw probability of the case, which is the product of the probabilities of the case attributes.

Applies To

Models that are built by using the Microsoft Clustering and Microsoft Sequence Clustering algorithms.

Return Type

Double-precision floating point number between 0 and 1. A number closer to 1 indicates that the case has a higher probability of occurring in this model. A number closer to 0 indicates that the case is less likely to occur in this model.

Remarks

By default, the result of the PredictCaseLikelihood function is normalized. Normalized values are typically more useful as the number of attributes in a case increase and the differences between the raw probabilities of any two cases become much smaller.

The following equation is used to calculate the normalized values, given x and y:

• x = likelihood of the case based on the clustering model

• y = Marginal case likelihood, calculated as the log likelihood of the case based on counting the training cases

• Z = Exp( log(x) – Log(Y))

Normalized = (z/ (1+z))

Examples

The following example returns the likelihood that the specified case will occur within the clustering model that was created in the Basic Data Mining Tutorial.

SELECT
  PredictCaseLikelihood() AS Default_Likelihood,
  PredictCaseLikelihood(NORMALIZED) AS Normalized_Likelihood,
  PredictCaseLikelihood(NONNORMALIZED) AS Raw_Likelihood,
FROM
  [TM Clustering]
NATURAL PREDICTION JOIN
(SELECT 28 AS [Age],
  '2-5 Miles' AS [Commute Distance],
  'Graduate Degree' AS [Education],
  0 AS [Number Cars Owned],
  0 AS [Number Children At Home]) AS t

Expected results:

The difference between these results demonstrates the effect of normalization.

See Also

Reference

Concepts

Change History