Partial Dependence Pre-processing

This Component is required to sample the data to be visualized in the Partial Dependence/ICE Plot (JavaScript) node.

You can select only numerical features of Double Type feature columns.

The Component will create for you the desired number of samples for each selected feature and for each instance row. The linear sampling technique will range between the lower and upper bound found in the input Table Spec. You can use Edit Numeric Domain node to customize the Table Spec bounds and the resulting sampling performed by the component.


Selected Numerical Features
Select the numerical features of Double Type to be sampled. If a column is constant, with its upper bound equal to its lower bound, the Component will fail.
Number of Samples
Type in here the number of samples. Each selected numerical feature will be sampled in this amount in each input instance row. Each time a feature column is linearly permuted this number of times on a selected row, the other feature values are kept fixed to the original values belonging to the same row. Minimum is 2.

Input Ports

A table with rows from your test/validation set. Make sure to include all the feature columns of your model. Also the categorical columns (String type) should be in this table if you have any.

Output Ports

A table containing in each row a different sample relative to a test/validation set row where only a single feature was changed. For each different row in the input table we sampled the desired number of samples for each selected numerical feature column. In fact the size of this table is: (number of samples) X (number of selected feature columns) X (number of rows in input table)