Check that the new point is not too close to already known observations to avoid numerical issues. Closeness can be estimated with several distances.
checkPredict(x, model, threshold = 1e-04, distance = "euclidean", type = "UK")TRUE if the point should not be tested.
a vector representing the input to check, alternatively a matrix with one point per row,
list of objects of class km, one for each objective functions,
optional value for the minimal distance to an existing observation, default to 1e-4,
selection of the distance between new observations, between "euclidean" (default), "none",
"covdist" and "covratio", see details,
"SK" or "UK" (default), depending whether uncertainty related to trend estimation has to be taken into account.
If the distance between x and the closest observations in model is below
threshold, x should not be evaluated to avoid numerical instabilities.
The distance can simply be the Euclidean distance or the canonical distance associated with the kriging predictive covariance k:
$$d(x,y) = \sqrt{k(x,x) - 2k(x,y) + k(y,y)}.$$
The last solution is the ratio between the prediction variance at x and the variance of the process.
none can be used, e.g., if points have been selected already.