from __future__ import print_function import copy #import logging import sys import math import numpy as np from sas.sascalc.dataloader.data_info import Data1D from sas.sascalc.dataloader.data_info import Data2D _SMALLVALUE = 1.0e-10 class FitHandler(object): """ Abstract interface for fit thread handler. The methods in this class are called by the optimizer as the fit progresses. Note that it is up to the optimizer to call the fit handler correctly, reporting all status changes and maintaining the 'done' flag. """ done = False """True when the fit job is complete""" result = None """The current best result of the fit""" def improvement(self): """ Called when a result is observed which is better than previous results from the fit. result is a FitResult object, with parameters, #calls and fitness. """ def error(self, msg): """ Model had an error; print traceback """ def progress(self, current, expected): """ Called each cycle of the fit, reporting the current and the expected amount of work. The meaning of these values is optimizer dependent, but they can be converted into a percent complete using (100*current)//expected. Progress is updated each iteration of the fit, whatever that means for the particular optimization algorithm. It is called after any calls to improvement for the iteration so that the update handler can control I/O bandwidth by suppressing intermediate improvements until the fit is complete. """ def finalize(self): """ Fit is complete; best results are reported """ def abort(self): """ Fit was aborted. """ # TODO: not sure how these are used, but they are needed for running the fit def update_fit(self, last=False): pass def set_result(self, result=None): self.result = result class Model: """ Fit wrapper for SAS models. """ def __init__(self, sas_model, sas_data=None, **kw): """ :param sas_model: the sas model to wrap for fitting """ self.model = sas_model self.name = sas_model.name self.data = sas_data def get_params(self, fitparams): """ return a list of value of paramter to fit :param fitparams: list of paramaters name to fit """ return [self.model.getParam(k) for k in fitparams] def set_params(self, paramlist, params): """ Set value for parameters to fit :param params: list of value for parameters to fit """ for k,v in zip(paramlist, params): self.model.setParam(k,v) def set(self, **kw): self.set_params(*zip(*kw.items())) def eval(self, x): """ Override eval method of model. :param x: the x value used to compute a function """ try: return self.model.evalDistribution(x) except: raise def eval_derivs(self, x, pars=[]): """ Evaluate the model and derivatives wrt pars at x. pars is a list of the names of the parameters for which derivatives are desired. This method needs to be specialized in the model to evaluate the model function. Alternatively, the model can implement is own version of residuals which calculates the residuals directly instead of calling eval. """ raise NotImplementedError('no derivatives available') def __call__(self, x): return self.eval(x) class FitData1D(Data1D): """ Wrapper class for SAS data FitData1D inherits from DataLoader.data_info.Data1D. Implements a way to get residuals from data. """ def __init__(self, x, y, dx=None, dy=None, smearer=None, data=None): """ :param smearer: is an object of class QSmearer or SlitSmearer that will smear the theory data (slit smearing or resolution smearing) when set. The proper way to set the smearing object would be to do the following: :: from sas.sascalc.data_util.qsmearing import smear_selection smearer = smear_selection(some_data) fitdata1d = FitData1D( x= [1,3,..,], y= [3,4,..,8], dx=None, dy=[1,2...], smearer= smearer) :Note: that some_data _HAS_ to be of class DataLoader.data_info.Data1D Setting it back to None will turn smearing off. """ Data1D.__init__(self, x=x, y=y, dx=dx, dy=dy) self.num_points = len(x) self.sas_data = data self.smearer = smearer self._first_unsmeared_bin = None self._last_unsmeared_bin = None # Check error bar; if no error bar found, set it constant(=1) # TODO: Should provide an option for users to set it like percent, # constant, or dy data if dy is None or dy == [] or dy.all() == 0: self.dy = np.ones(len(y)) else: self.dy = np.asarray(dy).copy() ## Min Q-value #Skip the Q=0 point, especially when y(q=0)=None at x[0]. if min(self.x) == 0.0 and self.x[0] == 0 and\ not np.isfinite(self.y[0]): self.qmin = min(self.x[self.x != 0]) else: self.qmin = min(self.x) ## Max Q-value self.qmax = max(self.x) # Range used for input to smearing self._qmin_unsmeared = self.qmin self._qmax_unsmeared = self.qmax # Identify the bin range for the unsmeared and smeared spaces self.idx = (self.x >= self.qmin) & (self.x <= self.qmax) self.idx_unsmeared = (self.x >= self._qmin_unsmeared) \ & (self.x <= self._qmax_unsmeared) def set_fit_range(self, qmin=None, qmax=None): """ to set the fit range""" # Skip Q=0 point, (especially for y(q=0)=None at x[0]). # ToDo: Find better way to do it. if qmin == 0.0 and not np.isfinite(self.y[qmin]): self.qmin = min(self.x[self.x != 0]) elif qmin is not None: self.qmin = qmin if qmax is not None: self.qmax = qmax # Determine the range needed in unsmeared-Q to cover # the smeared Q range self._qmin_unsmeared = self.qmin self._qmax_unsmeared = self.qmax self._first_unsmeared_bin = 0 self._last_unsmeared_bin = len(self.x) - 1 if self.smearer is not None: self._first_unsmeared_bin, self._last_unsmeared_bin = \ self.smearer.get_bin_range(self.qmin, self.qmax) self._qmin_unsmeared = self.x[self._first_unsmeared_bin] self._qmax_unsmeared = self.x[self._last_unsmeared_bin] # Identify the bin range for the unsmeared and smeared spaces self.idx = (self.x >= self.qmin) & (self.x <= self.qmax) ## zero error can not participate for fitting self.idx = self.idx & (self.dy != 0) self.idx_unsmeared = (self.x >= self._qmin_unsmeared) \ & (self.x <= self._qmax_unsmeared) def get_fit_range(self): """ Return the range of data.x to fit """ return self.qmin, self.qmax def size(self): """ Number of measurement points in data set after masking, etc. """ return len(self.x) def residuals(self, fn): """ Compute residuals. If self.smearer has been set, use if to smear the data before computing chi squared. :param fn: function that return model value :return: residuals """ # Compute theory data f(x) fx = np.zeros(len(self.x)) fx[self.idx_unsmeared] = fn(self.x[self.idx_unsmeared]) ## Smear theory data if self.smearer is not None: fx = self.smearer(fx, self._first_unsmeared_bin, self._last_unsmeared_bin) ## Sanity check if np.size(self.dy) != np.size(fx): msg = "FitData1D: invalid error array " msg += "%d <> %d" % (np.shape(self.dy), np.size(fx)) raise RuntimeError, msg return (self.y[self.idx] - fx[self.idx]) / self.dy[self.idx], fx[self.idx] def residuals_deriv(self, model, pars=[]): """ :return: residuals derivatives . :note: in this case just return empty array """ return [] class FitData2D(Data2D): """ Wrapper class for SAS data """ def __init__(self, sas_data2d, data=None, err_data=None): Data2D.__init__(self, data=data, err_data=err_data) # Data can be initialized with a sas plottable or with vectors. self.res_err_image = [] self.num_points = 0 # will be set by set_data self.idx = [] self.qmin = None self.qmax = None self.smearer = None self.radius = 0 self.res_err_data = [] self.sas_data = sas_data2d self.set_data(sas_data2d) def set_data(self, sas_data2d, qmin=None, qmax=None): """ Determine the correct qx_data and qy_data within range to fit """ self.data = sas_data2d.data self.err_data = sas_data2d.err_data self.qx_data = sas_data2d.qx_data self.qy_data = sas_data2d.qy_data self.mask = sas_data2d.mask x_max = max(math.fabs(sas_data2d.xmin), math.fabs(sas_data2d.xmax)) y_max = max(math.fabs(sas_data2d.ymin), math.fabs(sas_data2d.ymax)) ## fitting range if qmin is None: self.qmin = 1e-16 if qmax is None: self.qmax = math.sqrt(x_max * x_max + y_max * y_max) ## new error image for fitting purpose if self.err_data is None or self.err_data == []: self.res_err_data = np.ones(len(self.data)) else: self.res_err_data = copy.deepcopy(self.err_data) #self.res_err_data[self.res_err_data==0]=1 self.radius = np.sqrt(self.qx_data**2 + self.qy_data**2) # Note: mask = True: for MASK while mask = False for NOT to mask self.idx = ((self.qmin <= self.radius) &\ (self.radius <= self.qmax)) self.idx = (self.idx) & (self.mask) self.idx = (self.idx) & (np.isfinite(self.data)) self.num_points = np.sum(self.idx) def set_smearer(self, smearer): """ Set smearer """ if smearer is None: return self.smearer = smearer self.smearer.set_index(self.idx) self.smearer.get_data() def set_fit_range(self, qmin=None, qmax=None): """ To set the fit range """ if qmin == 0.0: self.qmin = 1e-16 elif qmin is not None: self.qmin = qmin if qmax is not None: self.qmax = qmax self.radius = np.sqrt(self.qx_data**2 + self.qy_data**2) self.idx = ((self.qmin <= self.radius) &\ (self.radius <= self.qmax)) self.idx = (self.idx) & (self.mask) self.idx = (self.idx) & (np.isfinite(self.data)) self.idx = (self.idx) & (self.res_err_data != 0) def get_fit_range(self): """ return the range of data.x to fit """ return self.qmin, self.qmax def size(self): """ Number of measurement points in data set after masking, etc. """ return np.sum(self.idx) def residuals(self, fn): """ return the residuals """ if self.smearer is not None: fn.set_index(self.idx) # Get necessary data from self.data and set the data for smearing fn.get_data() gn = fn.get_value() else: gn = fn([self.qx_data[self.idx], self.qy_data[self.idx]]) # use only the data point within ROI range res = (self.data[self.idx] - gn) / self.res_err_data[self.idx] return res, gn def residuals_deriv(self, model, pars=[]): """ :return: residuals derivatives . :note: in this case just return empty array """ return [] class FitAbort(Exception): """ Exception raise to stop the fit """ #pass #print"Creating fit abort Exception" class FitEngine: def __init__(self): """ Base class for the fit engine """ #Dictionnary of fitArrange element (fit problems) self.fit_arrange_dict = {} self.fitter_id = None def set_model(self, model, id, pars=[], constraints=[], data=None): """ set a model on a given in the fit engine. :param model: sas.models type :param id: is the key of the fitArrange dictionary where model is saved as a value :param pars: the list of parameters to fit :param constraints: list of tuple (name of parameter, value of parameters) the value of parameter must be a string to constraint 2 different parameters. Example: we want to fit 2 model M1 and M2 both have parameters A and B. constraints can be ``constraints = [(M1.A, M2.B+2), (M1.B= M2.A *5),...,]`` :note: pars must contains only name of existing model's parameters """ if not pars: raise ValueError("no fitting parameters") if model is None: raise ValueError("no model to fit") if not issubclass(model.__class__, Model): model = Model(model, data) sasmodel = model.model available_parameters = sasmodel.getParamList() for p in pars: if p not in available_parameters: raise ValueError("parameter %s not available in model %s; use one of [%s] instead" %(p, sasmodel.name, ", ".join(available_parameters))) if id not in self.fit_arrange_dict: self.fit_arrange_dict[id] = FitArrange() self.fit_arrange_dict[id].set_model(model) self.fit_arrange_dict[id].pars = pars self.fit_arrange_dict[id].vals = [sasmodel.getParam(name) for name in pars] self.fit_arrange_dict[id].constraints = constraints def set_data(self, data, id, smearer=None, qmin=None, qmax=None): """ Receives plottable, creates a list of data to fit,set data in a FitArrange object and adds that object in a dictionary with key id. :param data: data added :param id: unique key corresponding to a fitArrange object with data """ if data.__class__.__name__ == 'Data2D': fitdata = FitData2D(sas_data2d=data, data=data.data, err_data=data.err_data) else: fitdata = FitData1D(x=data.x, y=data.y, dx=data.dx, dy=data.dy, smearer=smearer) fitdata.sas_data = data fitdata.set_fit_range(qmin=qmin, qmax=qmax) #A fitArrange is already created but contains model only at id if id in self.fit_arrange_dict: self.fit_arrange_dict[id].add_data(fitdata) else: #no fitArrange object has been create with this id fitproblem = FitArrange() fitproblem.add_data(fitdata) self.fit_arrange_dict[id] = fitproblem def get_model(self, id): """ :param id: id is key in the dictionary containing the model to return :return: a model at this id or None if no FitArrange element was created with this id """ if id in self.fit_arrange_dict: return self.fit_arrange_dict[id].get_model() else: return None def remove_fit_problem(self, id): """remove fitarrange in id""" if id in self.fit_arrange_dict: del self.fit_arrange_dict[id] def select_problem_for_fit(self, id, value): """ select a couple of model and data at the id position in dictionary and set in self.selected value to value :param value: the value to allow fitting. can only have the value one or zero """ if id in self.fit_arrange_dict: self.fit_arrange_dict[id].set_to_fit(value) def get_problem_to_fit(self, id): """ return the self.selected value of the fit problem of id :param id: the id of the problem """ if id in self.fit_arrange_dict: self.fit_arrange_dict[id].get_to_fit() class FitArrange: def __init__(self): """ Class FitArrange contains a set of data for a given model to perform the Fit.FitArrange must contain exactly one model and at least one data for the fit to be performed. model: the model selected by the user Ldata: a list of data what the user wants to fit """ self.model = None self.data_list = [] self.pars = [] self.vals = [] self.selected = 0 def set_model(self, model): """ set_model save a copy of the model :param model: the model being set """ self.model = model def add_data(self, data): """ add_data fill a self.data_list with data to fit :param data: Data to add in the list """ if not data in self.data_list: self.data_list.append(data) def get_model(self): """ :return: saved model """ return self.model def get_data(self): """ :return: list of data data_list """ return self.data_list[0] def remove_data(self, data): """ Remove one element from the list :param data: Data to remove from data_list """ if data in self.data_list: self.data_list.remove(data) def set_to_fit(self, value=0): """ set self.selected to 0 or 1 for other values raise an exception :param value: integer between 0 or 1 """ self.selected = value def get_to_fit(self): """ return self.selected value """ return self.selected class FResult(object): """ Storing fit result """ def __init__(self, model=None, param_list=None, data=None): self.calls = None self.fitness = None self.chisqr = None self.pvec = [] self.cov = [] self.info = None self.mesg = None self.success = None self.stderr = None self.residuals = [] self.index = [] self.model = model self.data = data self.theory = [] self.param_list = param_list self.iterations = 0 self.inputs = [] self.fitter_id = None if self.model is not None and self.data is not None: self.inputs = [(self.model, self.data)] def set_model(self, model): """ """ self.model = model def set_fitness(self, fitness): """ """ self.fitness = fitness def __str__(self): """ """ if self.pvec is None and self.model is None and self.param_list is None: return "No results" sasmodel = self.model.model pars = enumerate(sasmodel.getParamList()) msg1 = "[Iteration #: %s ]" % self.iterations msg3 = "=== goodness of fit: %s ===" % (str(self.fitness)) msg2 = ["P%-3d %s......|.....%s" % (i, v, sasmodel.getParam(v)) for i,v in pars if v in self.param_list] msg = [msg1, msg3] + msg2 return "\n".join(msg) def print_summary(self): """ """ print(str(self))