Parameter error estimate with optimize.fmin_l_bfgs_b

Question

Is there a way to estimate the error of the estimated optimization parameters as calculated with scipy.optimize.fmin_l_bfgs_b (or any of the other bounded optimization routines in the scipy.optimize package)?

Answer 1

L-BFGS-B does not compute the error estimates for you, as you can see in its documentation.

However, you can estimate the error yourself by computing (numerically or analytically) the hessian of the objective function at the minimum returned by the optimizer: http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/DAVIES1/rd_bhatt_cvonline/node9.html

You can compute Hessians numerically in Python for example with numdifftools.