KDE nodes Build Options

Use the Build Options tab to specify build options for the KDE nodes, including basic options for kernel density parameters and cluster labels, and advanced options such as tolerance, leaf size, and whether to use a breadth-first approach. For additional information about these options, see the following online resources:

Basic

Bandwidth. Specify the bandwidth of the kernel.

Kernel. Select the kernel to use. Available kernels for the KDE Modeling node are Gaussian, Tophat, Epanechnikov, Eponential, Linear, or Cosine. Available kernels for the KDE Simulation node are Gaussian or Tophat. For details about these available kernels, see the Kernel Density Estimation User Guide.2

Algorithm. Select Auto, Ball Tree or KD Tree for the tree algorithm to use. For more information, see Ball Tree3 and KD Tree.4

Metric. Select a distance metric. Available metrics are Euclidean, Braycurtis, Chebyshev, Canberra, Cityblock, Dice, Hamming, Infinity, Jaccard, L1, L2, Matching, Manhattan, P, Rogerstanimoto, Russellrao, Sokalmichener, Sokalsneath, Kulsinski, or Minkowski. If you select Minkowski, set the P Value as desired.

The metrics available in this drop-down will vary depending on which algorithm you choose. Also note that the normalization of the density output is correct only for the Euclidean distance metric.

Advanced

Absolute Tolerance. Specify the desired absolute tolerance of the result. A larger tolerance will generally result in faster run time. Default is 0.0.

Relative Tolerance. Specify the desired relative tolerance of the result. A larger tolerance will generally result in faster run time. Default is 1E-8.

Leaf Size. Specify the leaf size of the underlying tree. Default is 40. Changing the leaf size may significantly impact performance and required memory. For more information about the Ball Tree and KD Tree algorithms, see Ball Tree3 and KD Tree.4

Breadth first. Select True if you want to use a breadth-first approach or False to use a depth-first approach.

The following table shows the relationship between the settings in the SPSS® Modeler KDE node dialogs and the Python KDE library parameters.
Table 1. Node properties mapped to Python library parameters
SPSS Modeler setting Script name (property name) KDE parameter
Inputs inputs
Bandwidth bandwidth bandwidth
Kernel kernel kernel
Algorithm algorithm algorithm
Metric metric metric
P Value pValue pValue
Absolute Tolerance atol atol
Relative Tolerance rtol Rtol
Leaf Size leafSize leafSize
Breadth first breadthFirst breadthFirst

1 "API Reference." sklearn.neighbors.KernelDensity. Web. © 2007-2018, scikit-learn developers.

2 "User Guide." Kernel Density Estimation. Web. © 2007-2018, scikit-learn developers.

3 "Ball Tree." Five balltree construction algorithms. © 1989, Omohundro, S.M., International Computer Science Institute Technical Report.

4 "K-D Tree." Multidimensional binary search trees used for associative searching. © 1975, Bentley, J.L., Communications of the ACM.