======================== 
Fitting
========================

To perform a fit you first need start parameters and lower and upper boundaries for the parameters. Usually, you will define them in the parameter file and read it while creating the parameter pool. Then you can get the three arrays like this:
    >>> start, lower, upper = simu.getStartLowerUpper()
    
If you want to have a look on how your simulation with the start values of the parameters compares to the data you can plot both:
    >>> simu.plotData(start)
    
For fitting you can use whatever routine you found to be good. E.g. in the package *scipy* there is a quite good least squares fitter. You can use it like this
    >>> import scipy.optimize
    >>> result = scipy.optimize.least_squares(simu.getResiduals, start, bounds=(l,u), method='trf', x_scale=numpy.array(u)-numpy.array(l), jac='3-point',verbose=2)
    >>> best = result.x
    
*best* is then the parameter array with the fitted values. You can plot your result:
    >>> simu.plotData(best)

You can write the result to a file:
    >>> pp.writeToFile("parameters_best.txt",best)

In general it very difficult to find the parameter set with the lowest SSR in a large parameter space. Different start values might end up in different best fits if you use a deterministic fitter like the least squares fit.
To tackle this problem, you can use the method :meth:`Fitters.Explore`. It is generating a certain number of random start parameter vectors ('seeds') inside the given parameter space and does a least squares fit for all of them. The procedure can take 
very long time depending on the number of seeds. Each least squares fit end up in a fixpoint. The method uses a clustering algorithm to group the fixpoints. All this information is returned by :meth:`Fitters.Explore` as a dictionary. 
    
    >>> out = Fitters.Explore(simu.getResiduals,pp.getStartLowerUpper(),50)

With :meth:`Fitters.plot_clusters_allpars` and :meth:`Fitters.plot_fixpoints_allpars` you can plot the result and get an overview over the fixpoints in the parameter space.