This node performs a principal component analysis (PCA) on the given input data. The directions of maximal variance (the principal components) are extracted and can be used in the PCA Apply node to project the input into a space of lower dimension while preserving a maximum of information.
- Type: Data Input data for the PCA
- Type: Data Covariance matrix of the input columns
- Type: Data Table containing parameters extracted from the PCA. Each row in the table represents one principal component, whereby the rows are sorted with decreasing eigenvalues, i.e. variance along the corresponding principal axis. The first column in the table contains the component's eigenvalue, a high value indicates a high variance (or in other words, the respective component dominates the orientation of the input data).
Each subsequent column (labeled with the name of the selected input column) contains a coefficient representing the influence of the respective input dimension to the principal component. The higher the absolute value, the higher the influence of the input dimension on the principal component.
The mapping of the input rows to, e.g. the first principal axis, is computed as follows (all done in the PCA Apply node): For each dimension in the original space subtract the dimension's mean value and then multiply the resulting vector with the vector given by this table (the first row in the spectral decomposition table to get the value on the first PC, the second row for the second PC and so on).
- Type: Transformation Model holding the PCA transformation used by the PCA Apply node to apply the transformation to, e.g. another validation set.