def evaluation_function(chromosome):
"""
Function the be used to evaluate the NSGA-II chromosomes and return fitness scores.
Contains the svm with disparate impact remover algorithm,
that returns the fitness scores for the specified metrics.
:param chromosome: Chromosome to specify parameters for SVM
:return: The fitness scores in a list: [accuracy_score, fairness_score]
"""
# Check if scores already calculated for identical chromosome, in which case return those scores
if str(chromosome) in FITNESS_SCORES:
return FITNESS_SCORES[str(chromosome)]
else:
C = get_C(chromosome)
gamma = get_gamma(chromosome)
selected_features = get_selected_features(chromosome, 30)
accuracy_score, fairness_score = svm_dir(training_data=training_data, test_data=test_data,
fairness_metric=fairness_metric,
accuracy_metric=accuracy_metric,
C=C, gamma=gamma, keep_features=selected_features,
privileged_groups=privileged_groups,
unprivileged_groups=unprivileged_groups, max_iter=max_iter,
svm_seed=svm_seed)
FITNESS_SCORES[str(chromosome)] = [accuracy_score, fairness_score]
return [accuracy_score, fairness_score]
def svm_dir_experiment(classifier_chromosome, num_generations, population_size,
mutation_rate, crossover_rate, chromosome_length,
fairness_metric, accuracy_metric, training_data,
test_data, privileged_groups, unprivileged_groups,
max_iter, svm_seed, name_postfix):
"""
SVM with Disparate Impact Remover experiment.
:param classifier_chromosome: Chromosome representing the SVM classifier,
containing the C & gamma values and the feature subset
:param num_generations: Number of generations for NSGA-II
:param population_size: Population size for NSGA-II
:param mutation_rate: Mutation rate for NSGA-II
:param crossover_rate: Crossover probability for NSGA-II
:param chromosome_length: Length of chromosome used in NSGA-II
:param fairness_metric: Fairness metric used to calculate the fairness score
:param accuracy_metric: Accuracy metric used to calculate the accuracy score
:param training_data: The training data set to run experiment on
:param test_data: The test data set to test the experiment on
:param privileged_groups: The privileged groups in the data set
:param unprivileged_groups: The unprivileged groups in the data set
:param max_iter: Max iterations for SVM
:param svm_seed: Seed used for RNG in SVM
:param name_postfix: Postfix to add to name of saved result file
:return: Resulting Pareto front
"""
# Train classifier
C = get_C(classifier_chromosome)
gamma = get_gamma(classifier_chromosome)
selected_features = get_selected_features(classifier_chromosome, 30)
sensitive_attribute = list(privileged_groups[0].keys())[0]
svm_dir_classifier, svm_dir_scale = train_svm_dir(
training_data=training_data,
C=C,
gamma=gamma,
keep_features=selected_features,
max_iter=max_iter,
svm_seed=svm_seed,
sensitive_attribute=sensitive_attribute)
# Defines the evaluation function
def evaluation_function(chromosome):
"""
Function the be used to evaluate the NSGA-II chromosomes and return fitness scores.
Contains the svm with disparate impact remover algorithm,
that returns the fitness scores for the specified metrics.
:param chromosome: Chromosome to specify parameters for SVM
:return: The fitness scores in a list: [accuracy_score, fairness_score]
"""
# Check if scores already calculated for identical chromosome, in which case return those scores
if str(chromosome) in FITNESS_SCORES:
return FITNESS_SCORES[str(chromosome)]
else:
privileged_threshold = get_classification_threshold(
chromosome, 0, 10)
unprivileged_threshold = get_classification_threshold(
chromosome, 10, 20)
accuracy_score, fairness_score = test_classifier(
classifier=svm_dir_classifier,
scale=svm_dir_scale,
test_data=test_data,
fairness_metric=fairness_metric,
accuracy_metric=accuracy_metric,
keep_features=selected_features,
privileged_threshold=privileged_threshold,
unprivileged_threshold=unprivileged_threshold,
privileged_groups=privileged_groups,
unprivileged_groups=unprivileged_groups)
FITNESS_SCORES[str(chromosome)] = [accuracy_score, fairness_score]
return [accuracy_score, fairness_score]
# Runs NSGA-II
result = nsga2(pop_size=population_size,
num_generations=num_generations,
chromosome_length=chromosome_length,
crossover_rate=crossover_rate,
mutation_rate=mutation_rate,
evaluation_algorithm=evaluation_function)
# Writes summary to file
result_summary = {
'name': 'SVM_DIR',
'result': result,
'fairness_metric': function_name_to_string(fairness_metric),
'accuracy_metric': function_name_to_string(accuracy_metric),
'nsga2_parameters': {
'num_generations': num_generations,
'population_size': population_size,
'crossover_rate': crossover_rate,
'mutation_rate': mutation_rate,
'chromosome_length': chromosome_length
}
}
write_result_to_file(result_summary, "svm_dir_" + name_postfix)
# Return only the result, not the summary
return result