CODEWITHSUNDEEP
pythonnumpyoptimizationscipy
j_e
j_e

Reputation: 57

Estimate of Inverse Hessian Using SciPy Minimization

I am using SciPy's 'minimize' function to minimize a function. The function returns the optimal value, along with an estimated Jacobian and Hessian. As below:

fun: -675.09792378630596
hess_inv: <8x8 LbfgsInvHessProduct with dtype=float64>
jac: array([  6.34713615e-02,   1.15960574e-03,   1.63709046e-03, 2.16914486e-02,  -8.02970135e-02,  -4.39513315e-02,
6.69160727e-02,  -5.68434189e-05])
message: b'CONVERGENCE: REL_REDUCTION_OF_F_<=_FACTR*EPSMCH'
nfev: 684
nit: 60
status: 0
success: True
x: array([  9.93756778e-01,   3.51823214e+00,  -2.06368373e-01, 7.37395700e-04,   2.11222756e-02,   3.29367564e-02, 1.22886906e-01,  -2.75434386e-01])

I want the estimated Hessian, but when I have it return hess_inv, all I get returned is 

<8x8 LbfgsInvHessProduct with dtype=float64>

rather than the maxtrix itself. How do I have it return the matrix?

Upvotes: 3

Views: 2663

Answers (1)

skoocda
skoocda

Reputation: 74

From the SciPy documentation for the LbfgsInvHessProduct; you can use the method todense() to obtain the LbfgsInvHessProduct's values as a dense array.

However, keep in mind the LbfgsInvHessProduct is still a matrix! It's a special memory-optimized format, but you can still call other matrix functions such as matmat(), transpose(), dot() etc.

Upvotes: 4

Related Questions