TaylorTest

Simple package to check derivatives

Usage

  check(f, Jf, x[, constant_components]; taylortestplot=false, taylortestdirection=nothing, f_kwargs...)

Returns true if Jf approximates the derivative/gradient/Jacobian of f at point x (along a random direction unless specified using taylortestdirection). f_kwargs are keywords arguments to be passed to f and Jf. constant_components is an optional Vector{Int} corresponding to components of the direction which should be set to zero, effectively ignoring the dependency of f on these components. If taylortestplot is true, a log-log plot of the error against the perturbation size will be shown.

  check!(f!, Jf!, x, size_f_x, size_Jf_x, [, constant_components]; taylortestplot=false, taylortestdirection=nothing, f_kwargs...)

Like check but handling non-allocating functions. Output size for both f! and the Jacobian Jf! must be provided (as Tuples) via size_f_x and size_Jf_x.

Examples (see test directory for more)

import TaylorTest

f = x -> cos(x)
Jf = x -> -sin(x)

TaylorTest.check(f, Jf, rand())

# [ Info: Approximation order ~ 1.0
# true