def run_fitting_schemes(model_types, number_of_generators, random_seeds,
schemes, mean_data, cov_data, max_fun_eval=100000):
all_models = def_all_models_dict(schemes, model_types,
number_of_generators, random_seeds)
overall_start_time = time.time()
for (model_type, n_gen, seed, scheme) in [(model_type, n_gen, seed, scheme)
for scheme in schemes
for model_type in model_types
for n_gen in number_of_generators
for seed in random_seeds]:
random.seed(seed)
entry = def_empty_entry_dict()
erp_model = ERP_Variability_Model_Fit(n_sub=23, n_gen=n_gen)
parameter_list = prepare_variability_parameter_list(erp_model,
model_type, n_gen)
# If parameter_list is None that means the combination of model_type
# and number of generators shouldn't be taken into account.
if parameter_list is None:
continue
parameter_list.insert(0,'amplitudes')
parameter_list.insert(0,'locations and orientations')
erp_model.set_random_parameters(parameter_list)
description = 'Fitting scheme: ' + scheme + ', model type: ' +\
model_type + ', with ' + str(n_gen) +\
' generator(s) and seed ' + str(seed)
print(description + '\n' + '=' * len(description))
entry['starting_error_mean'] = error_mean(mean_data, erp_model, [],
[]) / norm(mean_data, 2)
entry['starting_error_covariance'] = error_cov(cov_data, erp_model,
[], []) / norm(cov_data.reshape((erp_model.n_el**2,1)), 2)
entry['starting_error_mean_and_covariance'] =\
error_mean_and_cov(mean_data, cov_data, erp_model, [], [])
print('Starting error on mean: ' + str(entry['starting_error_mean']))
print('Starting error on covariance: ' +\
str(entry['starting_error_covariance']))
print('Starting error on mean and covariance combined: ' +\
str(entry['starting_error_mean_and_covariance']) + '\n')
# Simplifying calling of fitting functions to make fitting scheme code
# clearer and easier to read.
fit_amplitudes_lambda = lambda erp_model, with_locations:\
fit_amplitudes(erp_model, n_gen, seed, mean_data,
with_locations=with_locations,
max_fun_eval=max_fun_eval)
fit_variability_lambda = lambda erp_model, with_locations:\
fit_variability(erp_model, model_type, n_gen, seed, cov_data,
with_locations=with_locations,
max_fun_eval=max_fun_eval)
if scheme == 'covariance and locations':
entry['fits'].append(fit_variability_lambda(erp_model, True))
elif scheme == 'amplitudes, covariance and locations':
entry['fits'].append(fit_amplitudes_and_variability(erp_model,
model_type, n_gen, seed, mean_data, cov_data,
with_locations=True, max_fun_eval=max_fun_eval))
if scheme == 'amplitudes and locations, then covariance':
entry['fits'].append(fit_amplitudes_lambda(erp_model, True))
entry['fits'].append(fit_variability_lambda(erp_model, False))
if scheme == 'covariance and locations, then amplitudes':
entry['fits'].append(fit_variability_lambda(erp_model, True))
entry['fits'].append(fit_amplitudes_lambda(erp_model, False))
if scheme == 'amplitudes and locations, then covariance and ' +\
'locations, multiple times':
n_repeats = 10
for i in range(n_repeats):
entry['fits'].append(fit_amplitudes_lambda(erp_model, True))
entry['fits'].append(fit_variability_lambda(erp_model, True))
#if entry['fits'][-1] != None:
entry['scheme'] = scheme
entry['erp_model'] = erp_model
all_models[scheme][model_type][n_gen][seed] = entry
entry['final_error_mean'] = error_mean(mean_data, erp_model, [], [])\
/ norm(mean_data, 2)
entry['final_error_covariance'] = error_cov(cov_data, erp_model, [],
[]) / norm(cov_data.reshape((erp_model.n_el**2,1)), 2)
entry['final_error_mean_and_covariance'] =\
error_mean_and_cov(mean_data, cov_data, erp_model, [], [])
print('Final error on mean: ' + str(entry['final_error_mean']))
print('Final error on covariance: ' +\
str(entry['final_error_covariance']))
print('Final error on mean and covariance combined: ' +\
str(entry['final_error_mean_and_covariance']) + '\n')
print('Overall time elapsed: ' + str(round(time.time() -\
overall_start_time,2)))
return all_models
