def train_regression(y_desired_index=1, verbose=False):
"""Train multiple datasets (regression problem) under a XGBoost implementation.
:param y_desired_index: the index of the datasset to retrieve an instance
to test. Default is 1.
:type y_desired_index: Integer
:param verbose: define the verbosity. Default is False.
:type verbose: Boolean
:return: an array containing the results found. Each row includes the
dataset used, the algorithm used, the Pareto front, the Pareto set
(i.e. the counterfactual variables) and the multiobjective optimization
algorithm responsible of each counterfactual, respectively.
:rtype: np.array
"""
datasets = ['boston', 'diabetes', 'california']
algorithms = ['XGBRegressor']
trained_models = []
for dataset in datasets:
if verbose:
print(f'Retrieving the dataset {dataset}')
# Setting the sample index to be returned from the database
y_desired_index = 1
# Loading the database (in order to train the model) and
# required additional metadata on the sample in order to generate
# the contrafactuals
X, y, X_current, y_desired, immutable_column_indexes, \
upper_bounds, lower_bounds, y_acceptable_range, \
categorical_columns, integer_columns = \
load_sample_from_dataset(y_desired_index, dataset)
# Training a ML model
for algorithm in algorithms:
if verbose:
print(f'Training using {algorithm}')
model = train_ml_model(X, y, algorithm)
if verbose:
print(f'Starting counterfactual generation.')
pareto_front, pareto_set, pareto_algorithms = \
generate_counterfactuals_regression(model, X_current, \
y_desired, immutable_column_indexes, y_acceptable_range, \
upper_bounds, lower_bounds, categorical_columns, \
integer_columns, n_gen=50, pop_size=300, \
max_changed_vars=10, verbose=verbose, seed=0)
trained_models.append([
dataset, algorithm, pareto_front, pareto_set, pareto_algorithms
])
return np.array(trained_models)
def train_classification_proba(y_desired_index=1, verbose=False):
"""Train multiple datasets (classification problem) under multiple
scikit-learn implementations.
:param y_desired_index: the index of the datasset to retrieve an instance
to test. Default is 1.
:type y_desired_index: Integer
:param verbose: define the verbosity. Default is False.
:type verbose: Boolean
:return: an array containing the results found. Each row includes the
dataset used, the algorithm used, the Pareto front, the Pareto set
(i.e. the counterfactual variables) and the multiobjective optimization
algorithm responsible of each counterfactual, respectively.
:rtype: np.array
"""
datasets = ['breast_cancer', 'digits', 'iris', 'wine']
algorithms = ['AdaBoostClassifier', 'ExtraTreesClassifier', \
'GradientBoostingClassifier', 'RandomForestClassifier', \
'DecisionTreeClassifier', 'SGDClassifier', \
'LogisticRegression', 'KNeighborsClassifier']
trained_models = []
for dataset in datasets:
if verbose:
print(f'Retrieving the dataset {dataset}.')
# Loading the database (in order to train the model) and
# required additional metadata on the sample in order to generate
# the contrafactuals
X, y, X_current, y_desired, immutable_column_indexes, \
upper_bounds, lower_bounds, y_acceptable_range, \
categorical_columns, integer_columns = \
load_sample_from_dataset(y_desired_index, dataset)
# Training a ML model
for algorithm in algorithms:
if verbose:
print(f'Training using {algorithm}.')
model = train_ml_model(X, y, algorithm)
if verbose:
print(f'Starting counterfactual generation.')
pareto_front, pareto_set, pareto_algorithms = \
generate_counterfactuals_classification_proba(model, X_current, \
y_desired, immutable_column_indexes, y_acceptable_range, \
upper_bounds, lower_bounds, categorical_columns, \
integer_columns, n_gen=50, pop_size=100, \
max_changed_vars=5, verbose=verbose, seed=0)
trained_models.append([
dataset, algorithm, pareto_front, pareto_set, pareto_algorithms
])
return np.array(trained_models)
def train_classification_simple(y_desired_index=1, verbose=False):
datasets = ['breast_cancer', 'digits', 'iris', 'wine']
algorithms = ['LinearSVC', 'NuSVC', 'RidgeClassifier']
trained_models = []
for dataset in datasets:
if verbose:
print(f'Retrieving the dataset {dataset}.')
# Loading the database (in order to train the model) and
# required additional metadata on the sample in order to generate
# the contrafactuals
X, y, X_current, y_desired, immutable_column_indexes, \
upper_bounds, lower_bounds, y_acceptable_range, \
categorical_columns, integer_columns = \
load_sample_from_dataset(y_desired_index, dataset)
# Training a ML model
for algorithm in algorithms:
if verbose:
print(f'Training using {algorithm}.')
model = train_ml_model(X, y, algorithm)
if verbose:
print(f'Starting counterfactual generation.')
pareto_front, pareto_set, pareto_algorithms = \
generate_counterfactuals_classification_simple(model, X_current, \
y_desired, immutable_column_indexes, upper_bounds, \
lower_bounds, categorical_columns, integer_columns, \
n_gen=100, pop_size=100, max_changed_vars=10, \
verbose=verbose, seed=0)
trained_models.append([
dataset, algorithm, pareto_front, pareto_set, pareto_algorithms
])
return trained_models