Cross-validation

apbase.cross_validate.cross_validate compares IDW and ordinary kriging on the same source data. It uses leave-one-out cross-validation with one shared setup: a single fitted variogram sets the common search radius and the kriging model, and both methods use the same spatial index.

Its result feeds the decision step behind apbase.Map: Map passes it to select_best_model, which picks the method with the lower leave-one-out RMSE on the points both methods predicted (a tie goes to IDW). Call cross_validate directly when you want to inspect the comparison or run it with your own limits.

from apbase.cross_validate import cross_validate

result = cross_validate(x, y, z, max_points=500)
result.idw.rmse, result.kriging.rmse

use apbase::cross_validate::{CrossValidateOptions, cross_validate};

let result = cross_validate(&x, &y, &z, CrossValidateOptions { max_points: 500, ..Default::default() })?;
(result.idw.rmse, result.kriging.rmse);

Parameters

x, y : array_like : Source coordinates, in a projected coordinate system (no longitude/latitude conversion).

z : array_like : Source values. Must have the same length as x and y.

max_points : int, default 500 : Maximum source points evaluated. Larger datasets are randomly sampled to keep runtime bounded.

seed : int, optional : Seed for the validation subsample.

radius : float, optional : Shared local search radius for IDW and kriging. If omitted, APbase derives it from the fitted variogram (see Variogram fitting).

power : float, default 2.0 : IDW distance exponent.

max_neighbors, min_neighbors : int : Shared neighbor limits for both methods.

model_values : array_like, optional : Precomputed variogram model reused by the kriging evaluation.

Reading the result

result.idw and result.kriging expose mae, rmse, r2, and the per-sample predictions and actuals used to compute them. These values help you see whether one method's error is being driven by a few outlier points.

apbase.Map runs this function, then select_best_model, to pick between IDW and kriging automatically. See Cross-validation for the leave-one-out procedure and the selection rule.

For cokriging (collocated/ICM/LMC), see cross_validate_cokriging in Cokriging, which cross-validates all three methods together.