Hints

Fitting

The fitting algorithms in Fitters are not well develeped yet. It is better to use existing optimizers.

One option which delivers good performance is the least squares optimizer of scipy (scipy.optimize.least_squares) used in the following way (where simu is an instance of Experiment.ReflDataSimulator and pp is an instance of Parameters.Parameterpool):

>>> (start, l, u)=pp.getStartLowerUpper()
>>> res= scipy.optimize.least_squares(simu.getResiduals, start, bounds=(l,u), method='trf', x_scale=u-l, jac='3-point',verbose=2)
>>> best=res.x

With the given parameters a “trusted region reflective algorithm* (method=’trf’) will be used. Each parameter is scaled by the difference between upper and lower boundary (x_scale=u-l). And for the approximation of the Jacobian 3 points are used (jac=‘3-point’).