Select

Select

Provides efficient algorithms for information-theoretic sparsity selection.

select(kernel::Kernel, X_train, X_test, s)

Select the s points from X_train most informative to X_test greedily.

This function dispatches to select_single or select_mult depending on whether length(X_test) == 1. This function assumes each point of X_test can be conditioned by X_train.

See also select(kernel::Kernel, X, train, test, s).

select(kernel::Kernel, X, train, test, s; budget=true)

Select the s points from X[train] most informative to X[test] greedily.

This function dispatches to select_single if length(X_test) == 1, otherwise select_budget or select_nonadj depending on whether budget is true.

See also select(kernel::Kernel, X_train, X_test, s).

select_single(kernel::Kernel, X_train, X_test, s)

Select the s points from X_train most informative to X_test greedily.

This function assumes length(X_test) == 1.

See also select, select_mult.

select_mult(kernel::Kernel, X_train, X_test, s)

Select the s points from X_train most informative to X_test greedily.

This function assumes each point of X_test can be conditioned by X_train.

See also select, select_single.

select_nonadj(kernel::Kernel, X, train, test, s)

Select the s points from X_train most informative to X_test greedily.

See also select(kernel::Kernel, X, train, test, s), select_budget.

select_budget(kernel::Kernel, X, train, test, s)

Select the s points from X_train most informative to X_test greedily.

See also select(kernel::Kernel, X, train, test, s), select_nonadj.

covariance!(kernel::Kernel, X_train, X_test, L, i, k)

Fill the factor L's i-th column by the covariance with the k-th point.

cholupdate!(L, i, k, cond_var; rows=size(L, 1))

Conditions the i-th column of the Cholesky factor L by the k-th point.

This function modifies both L and cond_var in-place.

See also cholupdate!(L, i, k, cond_var, cond_cov).

cholupdate!(L, i, k, cond_var, cond_cov)

Additionally updates cond_cov in-place if it is provided.

See also cholupdate!(L, i, k, cond_var; rows).

log2prod(x)

Compute log2(prod(x)) for a vector x efficiently without under/overflow.

choldowndate!(L, u, order; start=1, finish=size(L, 2))

Computes the rank-one downdate of L by u, modifying L in-place.

Given a Cholesky factor $L$, a rank-one downdate by a vector $u$ computes

\[L \to \mathsf{chol}(L L^{\top} - u u^{\top}).\]

This function only computes the columns from start to finish.

See also: cholinsert!.

cholinsert!(L, index, k, order, i)

Insert the k-th point at the index column in the Cholesky factor L.

See also: choldowndate!.

insertindex(k, order, locations, i; start=1)

Finds the index to insert index k into order.

selectpoint!(kernel::Kernel, points, L, k, var, order, locations, i)

Add the k-th point to the Cholesky factor L.

This function modifies L and var in-place.

scoresupdate!(scores, L, var, pos, vars, order, locations, i)

Update scores in-place for all candidates.