def get_interped_y(label, data, x, x_to): run_y = lfu.grab_mobj_by_name(label, data) run_interped = lgeo.scalars( label = 'interpolated best result - ' + label, scalars = lm.linear_interpolation( x.scalars, run_y.scalars, x_to.scalars, 'linear')) return run_interped
def measure(self, *args, **kwargs): to_fit_to = args[0] run_data = args[1] dom_weight_max = 5.0 domain_weights = np.exp(np.linspace(dom_weight_max, 0, len(to_fit_to[0].scalars))) #domain_weights = [1 for val in # np.linspace(0, 1, len(to_fit_to[0].scalars))] #domain_weights = np.linspace(dom_weight_max, # 1, len(to_fit_to[0].scalars)) if self.best_flag: self.best_flag = False self.acceptance_weight /= 2.0 try: report_only = kwargs['report_only'] except KeyError: report_only = False labels = [[do.label for do in de] for de in args] run_data_domain = run_data[labels[1].index(labels[0][0])] try: run_data_codomains = [run_data[labels[1].index( labels[0][lab_dex+1])] for lab_dex in range(len(labels[0][1:]))] except ValueError: pdb.set_trace() run_data_interped = [scalars( label = 'interpolated result - ' + codomain.label, scalars = lm.linear_interpolation(run_data_domain.scalars, codomain.scalars, to_fit_to[0].scalars, 'linear')) for codomain in run_data_codomains] x_meas_bnds = (0, len(to_fit_to[0].scalars)) meas = [[diff for diff in kwargs['measurement']( fit_to.scalars, interped.scalars, x_meas_bnds, to_fit_to[0].scalars, domain_weights) if not math.isnan(diff)] for fit_to, interped in zip(to_fit_to[1:], run_data_interped)] meas = np.mean([np.mean(mea) for mea in meas]) if not report_only: self.data[0].scalars.append(meas) if meas == min(self.data[0].scalars) and\ meas < self.data[0].scalars[self.best_measure]: self.best_measure = len(self.data[0].scalars) - 1 #self.best_flag = self.best_measure > self.display_threshold self.best_flag = True if self.best_flag: if 'weight' in kwargs.keys(): self.acceptance_weight = kwargs['weight'] else: self.acceptance_weight *= 2.0 if (self.best_flag or kwargs['display']) and self.is_heaviest: print ' '.join(['metric', self.label, 'measured', str(meas), str(len(self.data[0].scalars))]) print self.parent.parameter_space.get_current_position() lgd.quick_plot_display(to_fit_to[0], to_fit_to[1:] + run_data_interped, delay = self.display_time) else: return meas
def raw_measure(*args): to_fit_to = args[0] run_data = args[1] labels = args[2] domain_weights = args[3] methods = args[4] run_data_domain = run_data[labels[1].index(labels[0][0])] run_data_codomains = [run_data[labels[1].index( labels[0][lab_dex+1])] for lab_dex in range(len(labels[0][1:]))] run_data_interped = [lm.linear_interpolation( run_data_domain, codomain, to_fit_to[0], 'linear') for codomain in run_data_codomains] x_meas_bnds = (0, len(to_fit_to[0])) meas = [[[diff for diff in method(domain_weights, fit_to, interped, x_meas_bnds, to_fit_to[0]) if not math.isnan(diff)] for fit_to, interped in zip(to_fit_to[1:], run_data_interped)] for method in methods] measurement = [np.mean([np.mean(mea) for mea in measur]) for measur, method in zip(meas, methods)] return measurement
