ScipyFitter

ScipyFitter = class ScipyFitter(MaxLikelihoodFitter)

ScipyFitter(xdata, model, method=None, gradient=True, hessp=None, **kwargs) Unified interface to the Scipy minimization module minimize, to fit data to a model. For documentation see scipy.org->Docs->Reference Guide->optimization and root finding. scipy.optimize.minimize Examples -------- # assume x and y are Double1d data arrays. >>> x = numpy.arange( 100, dtype=float ) / 10 >>> y = numpy.arange( 100, dtype=float ) / 122 # make slope >>> y += 0.3 * numpy.random.randn( 100 ) # add noise >>> y[9:12] += numpy.asarray( [5,10,7], dtype=float ) # make some peak >>> gauss = GaussModel( ) # Gaussian >>> gauss += PolynomialModel( 1 ) # add linear background >>> print( gauss.npchain ) >>> cgfit = ConjugateGradientFitter( x, gauss ) >>> param = cgfit.fit( y ) >>> print( len( param ) ) 5 >>> stdev = cgfit.stdevs >>> chisq = cgfit.chisq >>> scale = cgfit.scale # noise scale >>> yfit = cgfit.getResult( ) # fitted values >>> yband = cgfit.monteCarloEoor( ) # 1 sigma confidence region Notes ----- 1. CGF is <b>not</b> guaranteed to find the global minimum. 2. CGF does <b>not</b> work with fixed parameters or limits Attributes ---------- gradient : callable gradient( par ) User provided method to calculate the gradient of chisq. It can be used to speed up the dotproduct calculation, maybe because of the sparseness of the partial. default dotproduct of the partial with the residuals tol : float (1.0e-5) Stop when the norm of the gradient is less than tol. norm : float (inf) Order to use for the norm of the gradient (-np.Inf is min, np.Inf is max). maxIter : int (200*len(par)) Maximum number of iterations verbose : bool (False) if True print status at convergence debug : bool (False) return the result of each iteration in `vectors` yfit : ndarray (read only) the result of the model at the optimal parameters ntrans : int (read only) number of function calls ngrad : int (read only) number of gradient calls vectors : list of ndarray (read only when debug=True) list of intermediate vectors Hidden Attributes ----------------- _Chisq : class to calculate chisq in the method Chisq.func() and Chisq.dfunc() Returns ------- pars : array_like the parameters at the minimum of the function (chisq).

Method resolution order:: ScipyFitter; MaxLikelihoodFitter; IterativeFitter; BaseFitter; builtins.object

Constructor:

ScipyFitter( xdata, model, method=None, gradient=True, hessp=None, **kwargs ): Constructor. Create a class, providing inputs and model. Parameters ---------- xdata : array_like array of independent input values model : Model a model function to be fitted (linear or nonlinear) method : None | 'CG' | 'NELDER-MEAD' | 'POWELL' | 'BFGS' | 'NEWTON-CG' | 'L-BFGS-B' | 'TNC' | 'COBYLA' | 'SLSQP' | 'DOGLEG' | 'TRUST-NCG' the method name is case invariant. None Automatic selection of the method. 'SLSQP' when the problem has constraints 'L-BFGS-B' when the problem has limits 'BFGS' otherwise 'CG' Conjugate Gradient Method of Polak and Ribiere :ref:`(see here) <optimize.minimize-cg>` 'NELDER-MEAD' Nelder Mead downhill simplex :ref:`(see here) <optimize.minimize-neldermead>` 'POWELL' Powell's conjugate direction method :ref:`(see here) <optimize.minimize-powell>` 'BFGS' Quasi-Newton method of Broyden, Fletcher, Goldfarb, and Shannon :ref:`(see here) <optimize.minimize-bfgs>` 'NEWTON-CG' Truncated Newton method :ref:`(see here) <optimize.minimize-newtoncg>` 'L-BFGS-B' Limited Memory Algorithm for Bound Constrained Optimization :ref:`(see here) <optimize.minimize-lbfgsb>` 'TNC' Truncated Newton method with limits :ref:`(see here) <optimize.minimize-tnc>` 'COBYLA' Constrained Optimization BY Linear Approximation :ref:`(see here) <optimize.minimize-cobyla>` 'SLSQP' Sequential Least Squares :ref:`(see here) <optimize.minimize-slsqp>` 'DOGLEG' Dog-leg trust-region algorithm :ref:`(see here) <optimize.minimize-dogleg>` 'TRUST-NCG' Newton conjugate gradient trust-region algorithm :ref:`(see here) <optimize.minimize-trustncg>` gradient : bool or None or callable gradient( par ) if True use gradient calculated from model. It is the default. if False/None dont use gradient (use numeric approximation in stead) if callable use the method as gradient hessp : callable hessp(x, p, *args) or None Function which computes the Hessian times an arbitrary vector, p. The hessian itself is always provided. kwargs : dict Possibly includes keywords from MaxLikelihoodFitter : errdis, scale, power IterativeFitter : maxIter, tolerance, verbose BaseFitter : map, keep, fixedScale

Methods defined here:

collectVectors( par )

fit( data, weights=None, par0=None, keep=None, limits=None,
maxiter=None, tolerance=None, constraints=(), verbose=0, accuracy=None,
plot=False, callback=None, **options): Return parameters for the model fitted to the data array. Parameters ---------- ydata : array_like the data vector to be fitted weights : array_like weights pertaining to the data accuracy : float or array_like accuracy of (individual) data par0 : array_like initial values of the function. Default from Model. keep : dict of {int:float} dictionary of indices (int) to be kept at a fixed value (float) The values of keep are only valid for *this* fit See also `ScipyFitter( ..., keep=dict )` limits : None or list of 2 floats or list of 2 array_like None : no limits applied [lo,hi] : low and high limits for all values [la,ha] : low array and high array limits for the values constraints : list of callables constraint functions cf. All are subject to cf(par) > 0. maxiter : int max number of iterations tolerance : float stops when ( |hi-lo| / (|hi|+|lo|) ) < tolerance verbose : int 0 : silent >0 : print output if iter % verbose == 0 plot : bool Plot the results callback : callable is called each iteration as `val = callback( val )` where `val` is the minimizable array options : dict options to be passed to the method Raises ------ ConvergenceError if it stops when the tolerance has not yet been reached.

Methods inherited from MaxLikelihoodFitter:

Methods inherited from IterativeFitter:

Methods inherited from BaseFitter: