grad_est_subspace {activegp}

Estimate the Active Subspace of a Cheap Function using Gradients

Description

Looks between [-1, 1]

Usage

grad_est_subspace(f, r, m, M = NULL, scale = FALSE)

Arguments

f

The function to eval

r

The max dim of the active subspace

m

The dimension of the underlying/embedding space.

M

optional budget of evaluations, default to 2 * r * log(m)

scale

Scale all gradients to have norm 1?

Value

A list with sub, the active subspace, sv, the singular values (all m of them), fs, which gives function values, gs, function grads, and X, which gives sampled locations.