PredictorModels Table

  • Article

The actual models and the parameters used to build them are stored in this table by the Predictor resource when the models are built. The models are displayed in Commerce Server Manager when the Models node, under the Predictor resource, is selected. In/Out specifies whether the parameter is an input or output parameter. The Model Type specifies which model type uses that parameter. The last column, Data, is a Blob containing the binary representation for the actual model data.

Column Name Data Type In/
Out		 Model
Type		 Description Required?
ModelName DBTYPE_WSTR In Both Model name. Yes
ModelCfgName DBTYPE_WSTR In Both Model configuration name (foreign key to the PredictorModelCfgs table). Yes
ModelType DBTYPE_UI2 In Both Algorithm used to build the model. Default is Prediction.

Valid values are:

  • 0 = Prediction (Dependency Net)

  • 1 = Segment (Clustering)
 No
BuildTime DBTYPE_UI4 Out Both Time spent building model (milliseconds). No
ComplexityPenality DBTYPE_UI4 In Prediction Larger values prevent trees from growing too large (Prediction models only).

Valid range is 1 to MaxInt.

Default = 100

 No
DataFitScore DBTYPE_R8 Out Prediction Measures the quality of predictions made to supply missing profile data. The higher the score, the better the model.

Valid range is minus infinity to plus infinity.

 No
DateCreated DBTYPE_DB
TIMESTAMP		 Out Both Date model was built.

Default is getdate()

 Yes
InputAttributeFraction DBTYPE_R8 In Both Fraction of attributes used for feature selection as input to predictions.

The default number of attributes is 1000 or all attributes, whichever is less.

Valid range is 0 to 1.

  • 1 - use all attributes.

  • < 1 - use feature selection to identify the most significant attributes. Results in a smaller model.
 No
K DBTYPE_UI4 In Segment Number of segments to build.

The default is 10.

Valid range is 1 to 32756.

 No
MaxBufferSize DBTYPE_UI4 In Segment Maximum buffer size in bytes.

Default is 1,056,784 bytes (1 MB)

 No
MeasuredAccuracy
MaxPredictions		 DBTYPE_UI4 In Both Maximum number of recommendation made.

Valid range is 1 to MaxInt.

Default is 10.

 No
MeasuredAccuracy
SampleFraction		 DBTYPE_R8 In Both Fraction of sample data to hold out for automatic accuracy measurement.

Valid range is 0 to 1.

Default is 40% (0.4) of all cases or 1000, whichever is less.

 No
MinimumCasesToSplit DBTYPE_UI4 In Prediction Minimum number of cases for a split to occur in a tree.

Default = 5

 No
OutputAttributeFraction DBTYPE_R8 In Both Fraction of attributes used for feature selection as output of predictions.

The default number of attributes is 1000 or all attributes, whichever is less.

Valid range is 0 to 1.

  • 1 - build a model that can predict all attributes.

  • < 1 - use feature selection to identify attributes that are significant. Results in a smaller model.
 No
RecommendScore DBTYPE_R8 Out Prediction Measures the quality of a ranked list of recommendations. The higher the score, the better the model.

Valid range is 0 to 1.

 No
SampleSize DBTYPE_UI4 In Both Number of cases to randomly sample from the input data.

Default is -1 (use all data) or 20,000, whichever is less.

 No
Data DBTYPE_BYTES Blob containing the model data. No


All rights reserved.