def get_number_of_individual_by_sample(
sample_name,
remainder_division_number_individuals_all_samples_except_testing_and_number_of_samples,
indexes_individuals_sample_training, indexes_individuals_sample_test,
number_of_samples_except_testing, number_of_individuals):
if remainder_division_number_individuals_all_samples_except_testing_and_number_of_samples == 0:
if sample_name == 'training_1' or sample_name == 'training_2' or sample_name == 'validation':
number_of_individuals[sample_name] = int(
len(indexes_individuals_sample_training) /
number_of_samples_except_testing)
elif sample_name == 'test':
number_of_individuals[sample_name] = len(
indexes_individuals_sample_test)
else:
raise error.my_custom_error(
"The name of the sample introduced does not correspond to the given configuration. Please, have a look at the sample_name variable"
)
elif remainder_division_number_individuals_all_samples_except_testing_and_number_of_samples == 1:
if sample_name == 'training_1':
number_of_individuals[sample_name] = int(
len(indexes_individuals_sample_training) /
number_of_samples_except_testing) + 1
elif sample_name == 'training_2' or sample_name == 'validation':
number_of_individuals[sample_name] = int(
len(indexes_individuals_sample_training) /
number_of_samples_except_testing)
elif sample_name == 'test':
number_of_individuals[sample_name] = len(
indexes_individuals_sample_test)
else:
raise error.my_custom_error(
"The name of the sample introduced does not correspond to the given configuration. Please, have a look at the sample_name variable"
)
elif remainder_division_number_individuals_all_samples_except_testing_and_number_of_samples == 2:
if sample_name == 'training_1' or sample_name == 'training_2':
number_of_individuals[sample_name] = int(
len(indexes_individuals_sample_training) /
number_of_samples_except_testing) + 1
elif sample_name == 'validation':
number_of_individuals[sample_name] = int(
len(indexes_individuals_sample_training) /
number_of_samples_except_testing)
elif sample_name == 'test':
number_of_individuals[sample_name] = len(
indexes_individuals_sample_test)
else:
raise error.my_custom_error(
"The name of the sample introduced does not correspond to the given configuration. Please, have a look at the sample_name variable"
)
return number_of_individuals
def kernel_function_second_step(model, time_period_1, time_period_2):
individual_1 = model.data_to_train.iloc[time_period_1 - 1, :]
individual_2 = model.data_alpha_variables.iloc[time_period_2 - 1, :]
correspondence_time_period_line_all_samples = model.correspondence_time_period_line_all_samples
original_index_time_period_1 = model.sample_by_line[model.sample_to_train][
time_period_1 - 1]
original_index_time_period_2 = model.sample_by_line[
model.sample_alpha_variables][time_period_2 - 1]
correspondence_time_period_line_sample_to_train = correspondence_time_period_line_all_samples[
model.sample_to_train]
correspondence_time_period_line_sample_alpha_variables = correspondence_time_period_line_all_samples[
model.sample_alpha_variables]
squared_difference_between_individuals = pd.DataFrame(
data=(individual_1 - individual_2)**2)
line_associated_to_time_period_1 = correspondence_time_period_line_sample_to_train.loc[
correspondence_time_period_line_sample_to_train['time_period'] ==
original_index_time_period_1, 'line'].item()
line_associated_to_time_period_2 = correspondence_time_period_line_sample_alpha_variables.loc[
correspondence_time_period_line_sample_alpha_variables['time_period']
== original_index_time_period_2, 'line'].item()
if (model.number_of_renewable_energy > 0):
raise error.my_custom_error(
"The optimization of the second step is not designed for a number of renewable energy greater than 0. Please, check the value variable in the kernel function."
)
value = (
1 + msvm.dirac_delta(value_1=line_associated_to_time_period_1,
value_2=line_associated_to_time_period_2)
) * pe.exp(-sum(squared_difference_between_individuals.iloc[node - 1, 0] *
model.weights_variables[node]
for node in model.indexes_nodes))
return value
def get_indexes_individuals_by_sample(sample_name, seed_sampling,
indexes_individuals_sample_training,
indexes_individuals_sample_test,
number_of_individuals,
indexes_individuals):
np.random.seed(seed=seed_sampling)
if sample_name == 'training_1':
indexes_individuals[sample_name] = pd.DataFrame(
data=sorted(
np.random.choice(
a=indexes_individuals_sample_training['time_period'],
size=number_of_individuals[sample_name],
replace=False)),
columns=['time_period'],
index=np.array(range(1, number_of_individuals[sample_name] + 1)))
elif sample_name == 'training_2':
difference_all_individuals_and_individuals_training_1 = np.setdiff1d(
ar1=np.array(indexes_individuals_sample_training),
ar2=indexes_individuals['training_1'])
indexes_individuals[sample_name] = pd.DataFrame(
data=sorted(
np.random.choice(a=np.array(
difference_all_individuals_and_individuals_training_1),
size=number_of_individuals[sample_name],
replace=False)),
columns=['time_period'],
index=np.array(range(1, number_of_individuals[sample_name] + 1)))
elif sample_name == 'validation':
difference_all_individuals_and_individuals_training_1 = np.setdiff1d(
ar1=np.array(indexes_individuals_sample_training),
ar2=indexes_individuals['training_1'])
indexes_individuals[sample_name] = pd.DataFrame(
data=np.setdiff1d(
ar1=difference_all_individuals_and_individuals_training_1,
ar2=indexes_individuals['training_2']),
columns=['time_period'],
index=np.array(range(1, number_of_individuals[sample_name] + 1)))
elif sample_name == 'test':
indexes_individuals[sample_name] = indexes_individuals_sample_test
indexes_individuals[sample_name].index = range(
1, number_of_individuals[sample_name] + 1)
else:
raise error.my_custom_error(
"The name of the sample introduced does not correspond to the given configuration. Please, have a look at the sample_name variable"
)
return indexes_individuals
def get_indexes_individuals_all_lines(sample_name,
indexes_individuals_all_lines,
indexes_individuals, number_of_lines,
seed_shuffle):
if sample_name == 'training_1' or sample_name == 'training_2':
indexes_individuals_all_lines[sample_name] = list_split(
list_to_split=indexes_individuals[sample_name]
['time_period'].values.tolist(),
number_of_parts=number_of_lines,
seed_shuffle=seed_shuffle)
elif sample_name == 'validation' or sample_name == 'test':
indexes_individuals_all_lines[sample_name] = [
indexes_individuals[sample_name]['time_period'].values.tolist()
] * number_of_lines
else:
raise error.my_custom_error(
"The name of the sample introduced does not correspond to the given configuration. Please, have a look at the sample_name variable"
)
return indexes_individuals_all_lines
def run_optimization_problem_given_solver(
solver, problem, neos_flag, number_of_variables, number_of_constraints,
sense_opt_problem, maximum_number_iterations_multistart,
folder_results, csv_file_name_multistart, ampl_flag):
if problem == "m2svm_optimal_weights":
run_m2svm_optimal_weights(
solver=solver,
problem=problem,
neos_flag=neos_flag,
number_of_variables=number_of_variables,
number_of_constraints=number_of_constraints,
sense_opt_problem=sense_opt_problem,
maximum_number_iterations_multistart=
maximum_number_iterations_multistart,
folder_results=folder_results,
csv_file_name_multistart=csv_file_name_multistart,
ampl_flag=ampl_flag)
elif problem == "MINLP_trigonometric_functions":
run_minlp_trigonometric_functions(
solver=solver,
problem=problem,
neos_flag=neos_flag,
number_of_variables=number_of_variables,
number_of_constraints=number_of_constraints,
sense_opt_problem=sense_opt_problem,
maximum_number_iterations_multistart=
maximum_number_iterations_multistart,
folder_results=folder_results,
csv_file_name_multistart=csv_file_name_multistart,
ampl_flag=ampl_flag)
else:
raise error.my_custom_error(
"The given optimization problem does not exist❌🚫 Please, check 👁🗨 that the name is well-written ✍"
)
return "The optimization problem has been solved"
def alternating_approach(
maximum_number_iterations_alternating_approach,
threshold_difference_objective_values_second_step,
default_difference_objective_values_second_step,
seed_random_prediction_values, index_SVM_regularization_parameter,
sample_names, lowest_label_value, highest_label_value,
transformed_label_all_samples, SVM_regularization_parameter,
seed_initialize_parameters, number_of_nodes, bounds_weights,
new_label_values, data_all_samples, number_of_renewable_energy,
correspondence_time_period_line_all_samples, sample_by_line,
maximum_number_iterations_multistart,
perturbation_multistart_variables, seed_multistart,
default_new_objective_value_second_step, line, initial_weights,
label_values, solver, neos_flag):
iteration_alternating_approach = 0
difference_objective_values_second_step = default_difference_objective_values_second_step
variables_previous_iteration = np.array([math.nan] * number_of_nodes)
new_objective_value_second_step = default_new_objective_value_second_step
output_alternating_approach_by_iteration = {}
while iteration_alternating_approach <= maximum_number_iterations_alternating_approach and difference_objective_values_second_step >= threshold_difference_objective_values_second_step:
initial_variables = get_initial_variables(
iteration_alternating_approach=iteration_alternating_approach,
seed_initialize_parameters=seed_initialize_parameters,
number_of_nodes=number_of_nodes,
index_SVM_regularization_parameter=
index_SVM_regularization_parameter,
bounds_weights=bounds_weights,
variables_previous_iteration=variables_previous_iteration,
initial_weights=initial_weights,
line=line)
###################################################################################################################################################################################################
# ASUN: The alpha variables are the optimal solution of the Multiclass Support Vector Machine optimization problem. Such a problem is a convex quadratic problem which is solved using Cplex.
# Since the objective of this repo is to compare the performance of the different non-linear solvers, it is desirable to avoid the computation of any extra optimization problem. Hence,I copy
# and fix the results of the alpha variables obtained for a toy example with 12 individuals and SVM_regularization_parameter = 1e-3, i.e., C = 1/(1e-3) = 1e3. However, the code which computes
# the optimal variables in a general case will be just commented and not deleted.
# If further information about this point is necessary, please do not hesitate to contact Asun =)
# output_first_step = fsap.run_first_step_alternating_approach(SVM_regularization_parameter = SVM_regularization_parameter,
# number_of_nodes = number_of_nodes,
# index_SVM_regularization_parameter = index_SVM_regularization_parameter,
# sample_to_train = sample_names[0],
# transformed_label_all_samples = transformed_label_all_samples,
# new_label_values = new_label_values,
# data_all_samples = data_all_samples,
# number_of_renewable_energy = number_of_renewable_energy,
# correspondence_time_period_line_all_samples = correspondence_time_period_line_all_samples,
# sample_by_line = sample_by_line,
# initial_variables = initial_variables,
# sample_names = sample_names,
# line = line,
# label_values = label_values)
# alpha_variables = output_first_step['alpha_variables']
###################################################################################################################################################################################################
output_first_step = {}
output_first_step['accuracy'] = {}
output_first_step['accuracy'][sample_names[1]] = 1e2
alpha_variables = pd.DataFrame(data={
1: [1e-6, 1e-7, 0.999],
2: [1e-6, 1e-7, 0.999],
3: [1e-5, 1e-7, 0.999],
4: [0.000112, 0.203, 0.796],
5: [1e-5, 0.999, 1e-7],
6: [1e-5, 1e-6, 0.999],
7: [0.00108, 0.796, 0.203],
8: [1e-5, 0.999, 1e-7],
9: [1e-5, 0.999, 1e-7],
10: [1e-6, 0.999, 1e-7],
11: [1e-6, 0.999, 1e-7],
12: [1e-6, 0.999, 1e-7]
},
index=new_label_values)
if (len(data_all_samples['training_1']) != alpha_variables.shape[1]
or SVM_regularization_parameter != 1e-3):
raise error.my_custom_error(
"The data set or the regularization parameter has changed. Please, ask Asun to check the status of the alpha variables"
)
output_second_step = ssap.run_second_step_alternating_approach(
alpha_variables=alpha_variables,
initial_variables=initial_variables,
maximum_number_iterations_multistart=
maximum_number_iterations_multistart,
new_label_values=new_label_values,
transformed_label_all_samples=transformed_label_all_samples,
sample_alpha_variables=sample_names[0],
sample_to_train=sample_names[1],
number_of_nodes=number_of_nodes,
bounds_weights=bounds_weights,
perturbation_multistart_variables=perturbation_multistart_variables,
seed_multistart=seed_multistart,
index_SVM_regularization_parameter=
index_SVM_regularization_parameter,
iteration_alternating_approach=iteration_alternating_approach,
sample_names=sample_names,
SVM_regularization_parameter=SVM_regularization_parameter,
data_all_samples=data_all_samples,
correspondence_time_period_line_all_samples=
correspondence_time_period_line_all_samples,
sample_by_line=sample_by_line,
number_of_renewable_energy=number_of_renewable_energy,
line=line,
solver=solver,
neos_flag=neos_flag)
if (output_first_step['accuracy'][sample_names[1]] >
output_second_step['best_accuracy'][sample_names[1]]):
variables_previous_iteration = initial_variables
else:
variables_previous_iteration = output_second_step[
'optimal_weights']
old_objective_value_second_step = new_objective_value_second_step
new_objective_value_second_step = output_second_step[
'optimal_objective_value']
difference_objective_values_second_step = abs(
new_objective_value_second_step - old_objective_value_second_step)
last_iteration_alternating_approach = iteration_alternating_approach
output_alternating_approach_by_iteration[
iteration_alternating_approach] = {
'first_step': output_first_step,
'second_step': output_second_step
}
iteration_alternating_approach = iteration_alternating_approach + 1
output_alternating_approach = output_alternating_approach_by_iteration[
last_iteration_alternating_approach]
output_alternating_approach[
'last_iteration_alternating_approach'] = last_iteration_alternating_approach
return output_alternating_approach
def get_correspondence_time_period_line_all_samples(
correspondence_time_period_line_all_samples, number_of_lines,
indexes_individuals_all_lines,
indexes_individuals_all_samples_and_lines, sample_names):
for sample_name in sample_names:
correspondence_time_period_line_all_samples[
sample_name] = pd.DataFrame(index=[],
columns=['time_period', 'line'])
if sample_name in ['training_1', 'training_2']:
for line in range(0, number_of_lines):
correspondence_time_period_line_by_sample = pd.DataFrame(
index=range(
1,
len(indexes_individuals_all_lines[sample_name][line]) +
1),
columns=['time_period', 'line'])
for time_period in range(
1,
len(indexes_individuals_all_samples_and_lines[line]
[sample_name]) + 1):
correspondence_time_period_line_by_sample.at[
time_period,
'time_period'] = indexes_individuals_all_samples_and_lines[
line][sample_name][time_period - 1]
correspondence_time_period_line_by_sample.at[
time_period, 'line'] = line + 1
correspondence_time_period_line_all_samples[
sample_name] = correspondence_time_period_line_all_samples[
sample_name].append(
correspondence_time_period_line_by_sample
).sort_values(by='time_period')
correspondence_time_period_line_all_samples[
sample_name].index = range(
1,
len(correspondence_time_period_line_all_samples[
sample_name]) + 1)
elif sample_name in ['validation', 'test']:
default_line = 0 #Since in the validation and testing set, all the individuals belong to all the lines
default_value_for_line = -1 #In order to indicate that all the individuals share the same lines.
correspondence_time_period_line_by_sample = pd.DataFrame(
index=range(
1,
len(indexes_individuals_all_lines[sample_name]
[default_line]) + 1),
columns=['time_period', 'line'])
for time_period in range(
1,
len(indexes_individuals_all_samples_and_lines[default_line]
[sample_name]) + 1):
correspondence_time_period_line_by_sample.at[
time_period,
'time_period'] = indexes_individuals_all_samples_and_lines[
default_line][sample_name][time_period - 1]
correspondence_time_period_line_by_sample.at[
time_period, 'line'] = default_value_for_line
correspondence_time_period_line_all_samples[
sample_name] = correspondence_time_period_line_all_samples[
sample_name].append(
correspondence_time_period_line_by_sample).sort_values(
by='time_period')
correspondence_time_period_line_all_samples[
sample_name].index = range(
1,
len(correspondence_time_period_line_all_samples[
sample_name]) + 1)
else:
raise error.my_custom_error(
"The name of the sample introduced does not correspond to the given configuration. Please, have a look at the sample_name variable"
)
return correspondence_time_period_line_all_samples