DerivativeTests.jl

Since AD is not yet sufficiciently performant for many applications, one often has to resort to "manual" derivative (gradient, jacobian, hessian, etc) implementations. This package facilitates testing these implementations using either AD or FD.

Start using it by

add DerivativeTests
using DerivativeTests
?fdtest

Example

A test that will pass:

using DerivativeTests
F(x) = exp(x[1]*x[2]),
dF(x) = [ x[2], x[1] ] * F(x)
fdtest(F, dF, rand(2))

A test that will fail:

using DerivativeTests
F(x) = exp(x[1]*x[2]),
dF(x) = [ x[1], x[2] ] * F(x)
fdtest(F, dF, rand(2))