Geometry-Based Active Learning Distance

Abstract base class for active-learning distances based on Euclidean distances in a geometry space.

Subclasses implement how the pool or search space determines geometry-space scaling and how reference and query points are transformed into that same space.

Subclasses must implement the following methods:

private$.fit_geometry(xdt, search_space): Fit the geometry from the pool and search space, or from the search space only when xdt is NULL. The result must contain an integer dimension giving the number of geometry columns.
private$.transform(xdt, state): Transform query points into the geometry space. The result must be a matrix with one row per query point and state$dimension columns.

Fields

dimension: (integer(1) | NULL) Dimension of the fitted geometry space.
reference_embedding: (matrix() | NULL) Current reference points transformed into the fitted geometry space. NULL if the distance has not been fitted or reference points have not been set.

Fields

See also