grad_est_subspace

The active subspace method is a sensitivity analysis technique that finds important linear combinations of input variables for a simulator. This package provides functions allowing estimation of the active subspace without gradient information using Gaussian processes as well as sequential experimental design tools to minimize the amount of data required to do so. Implements Wycoff et al. (JCGS, 2021) <doi:10.48550/arXiv.1907.11572>.

Nathan Wycoff

activegp

Gaussian Process Based Design and Analysis for the Active
Subspace Method

grad_est_subspace function

<dl><dt>f</dt>
<dd>The function to eval</dd>
<dt>r</dt>
<dd>The max dim of the active subspace</dd>
<dt>m</dt>
<dd>The dimension of the underlying/embedding space.</dd>
<dt>M</dt>
<dd>optional budget of evaluations, default to <code>2 * r * log(m)</code></dd>
<dt>scale</dt>
<dd>Scale all gradients to have norm 1?</dd></dl>

Arguments

Estimate the Active Subspace of a Cheap Function using Gradients — grad_est_subspace

<dl>

<dt>f</dt>
<dd>The function to eval</dd>


<dt>r</dt>
<dd>The max dim of the active subspace</dd>


<dt>m</dt>
<dd>The dimension of the underlying/embedding space.</dd>


<dt>M</dt>
<dd>optional budget of evaluations, default to <code>2 * r * log(m)</code></dd>


<dt>scale</dt>
<dd>Scale all gradients to have norm 1?</dd>

</dl>

grad_est_subspace: Estimate the Active Subspace of a Cheap Function using Gradients

Description

Usage

Value

Arguments