多维ODR配件

Question

I want to fit M=2 sets of N=3 observations (X,Y) using scipy.odr in a single fitting step, from which I expect to get 2*M best-fit values (slope and intercept estimates within each of the M sets of observations). From reading the scipy.odr documentation and a few related stackoverflow questions, it seems that this should be possible, but when I try using the following minimal example, the fitting fails to converge ( Reason(s) for Halting: NP < 1 or NP > N ).

I'm starting with a reasonably good approximation of the best-fit beta values. Any ideas why this fails so miserably?

from pylab import *
from scipy import odr

x = array([[1.0,2.0,3.0],[1.1,2.1,3.1]])
y = array([[1.1,2.3,3.1],[5.9,7.0,8.2]])
sx = x*0 + .1
sy = y*0 + .1

def f(B, x):
    out = x * 0
    for k in range(x.shape[0]) :
        out[k,:] = B[2*k] * x[k,:] + B[2*k+1]
    return out

result = odr.ODR(
    odr.RealData( x, y, sx = sx, sy = sy ),
    odr.Model(f), beta0 = array([1.,0.,1.,5.])
    ).run()

result.pprint()

Answer 1

The error message has nothing to do with your start values. I am not sure if the ODR can handle this data, as it is virtually x,y,z . My interpretation is that it counts the members of x and y , which is N=2 each (arrays, but nevertheless) and compares this to your free parameters, which is NP=4 , so NP>N .