PredictorClient.fpDataFitScore
.gif "Ee784385.c++_on(en-US,CS.10).gif")
.gif "Ee784385.vb_off(en-US,CS.10).gif")
The read-only fpDataFitScore property returns the Data Fit Score, measuring the effectiveness of a prediction model to fill in missing profile data.
Definition
Property fpDataFitScore As Single
Parameters
None.
Error Values
This method sets the Number property of the global Err object to S_OK (&H00000000) to indicate success and to standard COM error values to indicate failure. For more information about standard COM errors, see Standard COM Errors. Additional information may be available using the global Err object. In particular, the Description property may contain a text description of the error.
Remarks
The Data Fit Score is automatically computed when a model is built, using a fraction of the data as test data. The amount of data to set aside for testing purposes is specified as the Measured Accuracy Sample Fraction. For more information, see Building a New Analysis Model and PredictorModels Table.
Use the Data Fit Score to test and fine-tune the input parameters for a prediction model. Known cases, taken from the input cases used to build the model, are tested against both the prediction model and a "stalking horse" model. The " stalking horse" model is a marginal model that assumes all properties are independent. An intermediate value is calculated by dividing the probability the prediction model assigns to a case in the test data by the probability the "stalking horse" model assigns to the same case. The score is the logarithm of the geometric average of these values for each output property. Thus, any score greater than zero means that the built model makes predictions that are more accurate than the "stalking horse" model.
Load the prediction model using the LoadModelFromDB or LoadModelFromFile methods prior to accessing this property.
For more information about memory ownership issues related to COM property and method parameters, see Managing COM Parameter Memory.