Approximate jacobian matrix for loss function

A numerical estimation for the jacobian matrix giving the change of loss function dimensions \(\mathcal{l}(\theta)\) for marginal change in each dimension of \(\theta\) vector

approx_jacobian_epsilon(theta, model_function, step = 1e-06, ...)

Arguments

theta	Vector of structural parameters. Assuming a named vector.
model_function	Function that should be used to transform \(\theta\) parameter into moment conditions
step	\(h\) step to numerically compute derivative
...	Additional arguments

Details

Jacobian matrix is not derived from gradient methods but is numerically approximated using a small \(h\) step (step argument).

Parallel implementation is proposed but is not efficient for the moment: it is usually slower than the sequential approach