def calculate_score_vs_parameters(best_parameters, err_settings, fit_settings, simulator, exp_data, spinA, spinB, calc_settings):
sys.stdout.write("Calculating chi2 in dependence of fitting parameters ...\n")
score_vs_parameters = []
Ne = len(err_settings['variables'])
Ns = err_settings['Ns']
for i in range(Ne):
sys.stdout.write('\r')
sys.stdout.write("Calculation step %d / %d" % (i+1, Ne))
sys.stdout.flush()
# Create a generation
generation = Generation(Ns)
generation.first_generation(fit_settings['parameters']['bounds'])
# Set all genes to the optimized values except the ones which will to be varied
for name in const['variable_names']:
index = fit_settings['parameters']['indices'][name]
if not (index == -1) and not (name in err_settings['variables'][i]):
for j in range(Ns):
generation.chromosomes[j].genes[index] = best_parameters[name]['value']
# Score the generation
generation.score_chromosomes(fit_settings, simulator, exp_data, spinA, spinB, calc_settings)
# Sort chromosomes according to their score
generation.sort_chromosomes()
# Store the variables and the corresponding score values
score_vs_variables = {}
for name in err_settings['variables'][i]:
score_vs_variables[name] = []
index = fit_settings['parameters']['indices'][name]
for j in range(Ns):
score_vs_variables[name].append(generation.chromosomes[j].genes[index])
score_vs_variables['score'] = []
for j in range(Ns):
score_vs_variables['score'].append(generation.chromosomes[j].score)
score_vs_parameters.append(score_vs_variables)
sys.stdout.write('\n')
return score_vs_parameters
def calculate_fit_and_noise_std(best_parameters, fit_settings, simulator,
exp_data, spinA, spinB, calc_settings):
# Create a generation with one chromosome and set all genes to the optimized values
generation = Generation(1)
generation.first_generation(fit_settings['parameters']['bounds'])
for name in const['variable_names']:
index = fit_settings['parameters']['indices'][name]
if not (index == -1):
generation.chromosomes[0].genes[index] = best_parameters[name][
'value']
# Calculate the corresponding fit
fit = get_fit(generation.chromosomes[0].genes, fit_settings, simulator,
exp_data, spinA, spinB, calc_settings)
# Calculate the standard deviation of noise
noise_std = calculate_noise_std(fit, exp_data, fit_settings, calc_settings)
return noise_std
def calculate_numerical_error(best_parameters, err_settings, fit_settings, simulator, exp_data, spinA, spinB, calc_settings):
# Create a generation
Ns = err_settings['Ns']
generation = Generation(Ns)
generation.first_generation(fit_settings['parameters']['bounds'])
# Set all genes to the optimized values
for name in const['variable_names']:
index = fit_settings['parameters']['indices'][name]
if not (index == -1):
for j in range(Ns):
generation.chromosomes[j].genes[index] = best_parameters[name]['value']
# Score the generation
generation.score_chromosomes(fit_settings, simulator, exp_data, spinA, spinB, calc_settings)
# Sort chromosomes according to their score
generation.sort_chromosomes()
# Determine the variation of the score
score_min = generation.chromosomes[0].score
score_max = generation.chromosomes[Ns-1].score
numerical_error = score_max - score_min
return numerical_error
def run_optimization(self, fit_settings, simulator, exp_data, spinA, spinB,
calc_settings):
sys.stdout.write('Starting the fitting...\n')
time_start = time.time()
for i in range(self.num_generations):
if (i == 0):
# Create the first generation
generation = Generation(self.size_generation)
generation.first_generation(
fit_settings['parameters']['bounds'])
else:
# Create the next generation
generation.produce_offspring(
fit_settings['parameters']['bounds'], self.prob_crossover,
self.prob_mutation)
# Score the generation
generation.score_chromosomes(fit_settings, simulator, exp_data,
spinA, spinB, calc_settings)
# Sort chromosomes according to their score
generation.sort_chromosomes()
# Save the best score in each optimization step
self.best_score[i] = generation.chromosomes[0].score
# Display some graphics
if self.display_graphics:
# Calculate the best fit so far
best_fit = get_fit(generation.chromosomes[0].genes,
fit_settings, simulator, exp_data, spinA,
spinB, calc_settings)
if (i == 0):
fig_fit, graph_fit = plot_fit(best_fit, exp_data,
self.fitted_data,
calc_settings)
fig_score, axes_score = plot_score(
self.best_score, bool(calc_settings['noise_std']))
elif ((i > 0) and (i < self.num_generations - 1)):
update_fit_plot(fig_fit, graph_fit, best_fit)
update_score_plot(axes_score, self.best_score,
bool(calc_settings['noise_std']))
elif (i == self.num_generations - 1):
close_fit_plot(fig_fit)
close_score_plot(fig_score)
sys.stdout.write('\r')
sys.stdout.write("Optimization step %d / %d: chi2 = %f" %
(i + 1, self.num_generations, self.best_score[i]))
sys.stdout.flush()
# Calculate the best fit
if self.display_graphics:
self.best_fit = best_fit
else:
self.best_fit = get_fit(generation.chromosomes[0].genes,
fit_settings, simulator, exp_data, spinA,
spinB, calc_settings)
# Check that Chi2 is normalized by the variance of noise
if calc_settings['noise_std'] == 0:
noise_std = calculate_noise_std(self.best_fit, exp_data,
fit_settings, calc_settings)
calc_settings['noise_std'] = noise_std
self.best_score = self.best_score / noise_std**2
# Store the best genes
self.best_parameters = get_parameters(
generation.chromosomes[0].genes,
fit_settings['parameters']['indices'],
fit_settings['parameters']['fixed'])
time_finish = time.time()
time_elapsed = str(datetime.timedelta(seconds=time_finish -
time_start))
sys.stdout.write('\n')
sys.stdout.write('The fitting is finished. Total duration: %s\n\n' %
(time_elapsed))