deriva_grad

The function computes the information score matrix in the case where the first
derivatives of the function to optimize are analytically known. Therefore,
minus the derivatives of the gradient are computed by central finite differences.

This algorithm provides a numerical solution to the
problem of unconstrained local minimization (or maximization). It is particularly suited for complex problems and more efficient than
the Gauss-Newton-like algorithm when starting from points very
far from the final minimum (or maximum). Each iteration is parallelized and convergence relies on a stringent stopping criterion based on the first and second derivatives. See Philipps et al, 2021 <doi:10.32614/RJ-2021-089>.

Viviane Philipps

marqLevAlg

A Parallelized General-Purpose Optimization Based on
Marquardt-Levenberg Algorithm

deriva_grad function

<dl><dt>nproc</dt>
<dd>number of processors for parallel computing</dd>
<dt>b</dt>
<dd>value of parameters to be optimized over</dd>
<dt>grad</dt>
<dd>the gradient of the function to be minimized (or maximized)</dd>
<dt>.packages</dt>
<dd>character vector of packages that grad depends on</dd>
<dt>...</dt>
<dd>other arguments of the grad function</dd></dl>

Arguments

Viviane Philipps, Boris Hejblum, Cecile Proust-Lima, Daniel Commenges

Author

Numerical derivatives of the gradient function — deriva_grad

<dl>

<dt>nproc</dt>
<dd>number of processors for parallel computing</dd>


<dt>b</dt>
<dd>value of parameters to be optimized over</dd>


<dt>grad</dt>
<dd>the gradient of the function to be minimized (or maximized)</dd>


<dt>.packages</dt>
<dd>character vector of packages that grad depends on</dd>


<dt>...</dt>
<dd>other arguments of the grad function</dd>

</dl>

deriva_grad: Numerical derivatives of the gradient function

Description

Usage

Value

Arguments

Author