cov.py

import casadi as ca
import pylab as pl

# [...]

# A is of type SX, column vector, shape (788, 1), dense
# B is of type sx, column vector, shape (984, 1), dense

# V is of type struct_symSX, shape (1594, 1), dense
# W is a numpy array, matrix with shape (788, 788), almost diagonal

fA = ca.SXFunction(ca.nlpIn(x = V), ca.nlpOut(f = A))
fA.init()

fB = ca.SXFunction(ca.nlpIn(x = V), ca.nlpOut(f = B))
fB.init()

J1 = ca.mul(W, fA.jac("x", "f"))
J2 = fB.jac("x", "f")

h1 = ca.horzcat((ca.mul([J1.T, W, J1]), J2.T))
h2 = ca.horzcat((J2, pl.zeros((B.size, B.size))))

Jp = ca.mul([ \

    ca.horzcat((pl.eye(V.size), pl.zeros((V.size, B.size)))), \
    
    ca.solve(ca.vertcat((h1, h2)), pl.eye(V.size, + B.size)), \
    
    ca.vertcat((J1.T, pl.zeros((B.size, A.size)))) \
    
    ])
    
fcov = ca.SXFunction(ca.nlpIn(x = V), ca.nlpOut(f = ca.mul(Jp, Jp.T)))
fcov.init()

# [...] followed by an evaluation of fcov