Weighting Function

Use this feature to specify the function used to compute posterior probabilities from distances or to weight observations when computing k-nearest neighbor means.

Available options are described in the table below:

Option	Description
Gaussian	Indicates the Gaussian kernel function as described in the SAS/STAT PROC DISCRIM documentation.
Exponential	Similar to Gaussian but uses distances inside the exponential function instead of squared distances.
Reciprocal	Indicates reciprocal distances. Note: In this case,1 is added to the distances before computing the reciprocal.
Cauchy	Uses reciprocal squared distances. Note: In this case,1 is added to the distances before computing the reciprocal.
Uniform	A uniform kernel creates equal weights of 1.

For continuous responses with nonuniform kernels, a scaling factor for the distances is estimated by minimizing root mean square error.

To Specify the Desired Weighting Function:

8

Select the desired function by clicking within the appropriate radio button.