def generate_oversamplers(oversamplers_names):
"Generate oversamplers."
oversamplers = [
('NO OVERSAMPLING', None, {}),
('RANDOM OVERSAMPLING', RandomOverSampler(), {}),
('SMOTE', SMOTE(), {
'k_neighbors': [3, 5]
}), ('BORDERLINE SMOTE', BorderlineSMOTE(), {
'k_neighbors': [3, 5]
}), ('ADASYN', ADASYN(), {
'n_neighbors': [2, 3]
}),
('G-SMOTE', GeometricSMOTE(), {
'k_neighbors': [3, 5],
'selection_strategy': ['combined', 'minority', 'majority'],
'truncation_factor': [-1.0, -0.5, .0, 0.25, 0.5, 0.75, 1.0],
'deformation_factor': [.0, 0.2, 0.4, 0.5, 0.6, 0.8, 1.0]
}),
('K-MEANS RANDOM OVERSAMPLING',
RandomOverSampler(clusterer=KMeans(),
distributor=DensityDistributor()),
{
'k_neighbors': [3,
5],
'clusterer__n_clusters':
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
'distributor__distances_exponent': [0, 1, 2, 5],
'distributor__filtering_threshold': [0.0, 0.5, 1.0, 2.0]
}),
('K-MEANS SMOTE',
SMOTE(clusterer=KMeans(), distributor=DensityDistributor()), {
'k_neighbors': [3, 5],
'clusterer__n_clusters':
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
'distributor__distances_exponent': [0, 1, 2, 5],
'distributor__filtering_threshold': [0.0, 0.5, 1.0, 2.0]
}),
('K-MEANS BORDERLINE SMOTE',
BorderlineSMOTE(clusterer=KMeans(), distributor=DensityDistributor()),
{
'k_neighbors': [3, 5],
'clusterer__n_clusters':
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
'distributor__distances_exponent': [0, 1, 2, 5],
'distributor__filtering_threshold': [0.0, 0.5, 1.0, 2.0]
}),
('K-MEANS G-SMOTE',
GeometricSMOTE(clusterer=KMeans(), distributor=DensityDistributor()),
{
'k_neighbors':
[3, 5],
'selection_strategy': ['combined', 'minority', 'majority'],
'truncation_factor': [-1.0, -0.5, .0, 0.25, 0.5, 0.75, 1.0],
'deformation_factor': [.0, 0.2, 0.4, 0.5, 0.6, 0.8, 1.0],
'clusterer__n_clusters':
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
'distributor__distances_exponent': [0, 1, 2, 5],
'distributor__filtering_threshold': [0.0, 0.5, 1.0, 2.0]
}),
('SOMO', SMOTE(clusterer=SOM(), distributor=DensityDistributor()), {
'k_neighbors': [3, 5],
'clusterer__n_clusters':
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
'distributor__distances_exponent': [0, 1, 2, 5],
'distributor__filtering_threshold': [0.0, 0.5, 1.0, 2.0],
'distributor__distribution_ratio': [0.0, 0.25, 0.5, 0.75, 1.0]
}),
('G-SOMO',
GeometricSMOTE(clusterer=SOM(), distributor=DensityDistributor()), {
'k_neighbors': [3, 5],
'selection_strategy': ['combined', 'minority', 'majority'],
'truncation_factor': [-1.0, -0.5, .0, 0.25, 0.5, 0.75, 1.0],
'deformation_factor': [.0, 0.2, 0.4, 0.5, 0.6, 0.8, 1.0],
'clusterer__n_clusters':
[0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
'distributor__distances_exponent': [0, 1, 2, 5],
'distributor__filtering_threshold': [0.0, 0.5, 1.0, 2.0],
'distributor__distribution_ratio': [0.0, 0.25, 0.5, 0.75, 1.0]
})
]
if oversamplers_names in ('basic', 'scaled', 'undersampled'):
oversamplers = select_pipelines(
oversamplers, ('NO OVERSAMPLING', 'RANDOM OVERSAMPLING', 'SMOTE',
'BORDERLINE SMOTE', 'ADASYN', 'G-SMOTE'))
if oversamplers_names == 'scaled':
oversamplers = append_transformer(MinMaxScaler(), oversamplers)
elif oversamplers_names == 'undersampled':
oversamplers = set_sampling_strategy(
lambda y: generate_sampling_strategy(y, 1 / 3), oversamplers)
oversamplers = append_transformer(
RandomUnderSampler(
sampling_strategy=lambda y: generate_sampling_strategy(y, 3)),
oversamplers)
return oversamplers
from sklearnext.over_sampling.base import DensityDistributor
from sklearnext.tools import report_model_search_results
# Load data
X, y = make_classification(n_informative=15,
n_clusters_per_class=3,
weights=[0.9, 0.1])
# Define estimators
estimators = [('GBC', GradientBoostingClassifier()),
('SMOTE+GBC',
Pipeline([('smote', SMOTE()),
('gbc', GradientBoostingClassifier())])),
('KMeanSMOTE+GBC',
Pipeline([('smote',
SMOTE(clusterer=KMeans(n_init=1),
distributor=DensityDistributor())),
('gbc', GradientBoostingClassifier())]))]
# Define parameters grid
param_grids = [{
'SMOTE+GBC__smote__k_neighbors': [2, 3, 4, 5],
'SMOTE+GBC__gbc__max_depth': [2, 4]
}, {
'KMeanSMOTE+GBC__smote__k_neighbors': [2, 3, 4, 5],
'KMeanSMOTE+GBC__smote__clusterer__n_clusters':
[0.1, 0.3, 0.5, 0.7, 0.9, 1.0],
'KMeanSMOTE+GBC__smote__distributor__filtering_threshold':
[0.8, 1.0, 1.2, 1.5],
'KMeanSMOTE+GBC__gbc__max_depth': [2, 4]
}]
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors.classification import KNeighborsClassifier
from sklearn.ensemble import GradientBoostingClassifier
from sklearnext.cluster import KMeans
from sklearnext.tools import evaluate_binary_imbalanced_experiments, read_csv_dir, summarize_binary_datasets
from sklearnext.over_sampling import RandomOverSampler, DensityDistributor
# Paths
datasets_path = join(dirname(__file__), '..', '..', 'data', 'binary-numerical-imbalanced')
results_path = join(dirname(__file__), '..', '..', 'data', 'results', 'kmeans-oversampling', 'random-oversampler')
# Oversamplers and classifiers
oversamplers = [
('NO OVERSAMPLING', None),
('RANDOM OVERSAMPLING', RandomOverSampler(random_state=0)),
('K-MEANS RANDOM OVERSAMPLING', RandomOverSampler(clusterer=KMeans(random_state=1, n_init=1), distributor=DensityDistributor(), random_state=0), {
'clusterer__n_clusters': [0.0, 0.25, 0.5, 0.75, 1.0],
'distributor__distances_exponent': [0, 1, 2],
'distributor__filtering_threshold': [0.5, 1.0]
}
)
]
classifiers = [
('LR', LogisticRegression()),
('KNN', KNeighborsClassifier(), {'n_neighbors': [3, 5]}),
('DT', DecisionTreeClassifier(random_state=2), {'max_depth': [3, 6]}),
('GBC', GradientBoostingClassifier(random_state=3), {'max_depth': [3, 6], 'n_estimators': [50, 100]})
]
# Load datasets
imbalanced_datasets = read_csv_dir(datasets_path)
import pytest
import numpy as np
from sklearnext.cluster import KMeans
from ...over_sampling import RandomOverSampler, SMOTE, GeometricSMOTE
from ...cluster import SOM
from ...utils.validation import _TrivialOversampler
from ..distribution import DensityDistributor
X = np.array(list(product(range(5), range(4))))
y = np.array([0] * 10 + [1] * 6 + [2] * 4)
LABELS = np.array([0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 3, 3, 3, 0, 3, 3, 3])
NEIGHBORS = [(0, 1), (0, 2), (0, 3), (1, 2), (2, 3)]
@pytest.mark.parametrize('clusterer', [None, KMeans(), SOM()])
def test_fit(clusterer):
"""Test the fit method of the extended base oversampler."""
oversampler = _TrivialOversampler(clusterer=clusterer).fit(X, y)
assert oversampler.sampling_strategy_ == OrderedDict({1: 4, 2: 6})
@pytest.mark.parametrize('clusterer', [None, KMeans(), SOM()])
def test_fit(clusterer):
"""Test the fit and resample method of the extended base oversampler."""
oversampler = _TrivialOversampler(clusterer=clusterer)
X_resampled, y_resampled = oversampler.fit_resample(X, y)
assert hasattr(oversampler, 'distributor_')
assert hasattr(oversampler.distributor_, 'intra_distribution_')
assert hasattr(oversampler.distributor_, 'inter_distribution_')
if isinstance(clusterer, SOM):
from sklearnext.tools import evaluate_binary_imbalanced_experiments, read_csv_dir, summarize_binary_datasets
from sklearnext.over_sampling import SMOTE, DensityDistributor
# Paths
datasets_path = join(dirname(__file__), '..', '..', 'data',
'binary-numerical-imbalanced')
results_path = join(dirname(__file__), '..', '..', 'data', 'results',
'kmeans-oversampling', 'smote')
# Oversamplers and classifiers
oversamplers = [('NO OVERSAMPLING', None),
('SMOTE', SMOTE(random_state=0), {
'k_neighbors': [3, 5]
}),
('K-MEANS SMOTE',
SMOTE(clusterer=KMeans(random_state=1, n_init=1),
distributor=DensityDistributor(),
random_state=0), {
'k_neighbors': [3, 5],
'clusterer__n_clusters':
[0.0, 0.25, 0.5, 0.75, 1.0],
'distributor__distances_exponent': [0, 1, 2],
'distributor__filtering_threshold': [0.5, 1.0]
})]
classifiers = [('LR', LogisticRegression()),
('KNN', KNeighborsClassifier(), {
'n_neighbors': [3, 5]
}),
('DT', DecisionTreeClassifier(random_state=2), {
'max_depth': [3, 6]
}),
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors.classification import KNeighborsClassifier
from sklearn.ensemble import GradientBoostingClassifier
from sklearnext.cluster import KMeans
from sklearnext.tools import evaluate_binary_imbalanced_experiments, read_csv_dir, summarize_binary_datasets
from sklearnext.over_sampling import SMOTE, DensityDistributor
# Paths
datasets_path = join(dirname(__file__), '..', '..', 'data', 'binary-numerical-imbalanced')
results_path = join(dirname(__file__), '..', '..', 'data', 'results', 'kmeans-oversampling', 'borderline-smote')
# Oversamplers and classifiers
oversamplers = [
('NO OVERSAMPLING', None),
('BORDERLINE-SMOTE', SMOTE(random_state=0, kind='borderline1'), {'k_neighbors': [3, 5]}),
('K-MEANS BORDERLINE-SMOTE', SMOTE(clusterer=KMeans(random_state=1, n_init=1), distributor=DensityDistributor(), random_state=0, kind='borderline1'), {
'k_neighbors': [3, 5],
'clusterer__n_clusters': [0.0, 0.25, 0.5, 0.75, 1.0],
'distributor__distances_exponent': [0, 1, 2],
'distributor__filtering_threshold': [0.5, 1.0]
}
)
]
classifiers = [
('LR', LogisticRegression()),
('KNN', KNeighborsClassifier(), {'n_neighbors': [3, 5]}),
('DT', DecisionTreeClassifier(random_state=2), {'max_depth': [3, 6]}),
('GBC', GradientBoostingClassifier(random_state=3), {'max_depth':[3, 6], 'n_estimators': [50, 100]})
]
# Load datasets
from sklearnext.over_sampling import GeometricSMOTE, DensityDistributor
# Paths
datasets_path = join(dirname(__file__), '..', '..', 'data', 'binary-numerical-imbalanced')
results_path = join(dirname(__file__), '..', '..', 'data', 'results', 'kmeans-oversampling', 'gsmote')
# Oversamplers and classifiers
oversamplers = [
('NO OVERSAMPLING', None),
('G-SMOTE', GeometricSMOTE(random_state=0), {
'k_neighbors': [3, 5],
'truncation_factor': [-1.0, 0.0, 1.0],
'deformation_factor': [0.0, 0.5, 1.0]
}
),
('K-MEANS G-SMOTE', GeometricSMOTE(clusterer=KMeans(random_state=1, n_init=1), distributor=DensityDistributor(), random_state=0), {
'k_neighbors': [3, 5],
'truncation_factor': [-1.0, 0.0, 1.0],
'deformation_factor': [0.0, 0.5, 1.0],
'clusterer__n_clusters': [0.0, 0.25, 0.5, 0.75, 1.0],
'distributor__distances_exponent': [0, 1, 2],
'distributor__filtering_threshold': [0.5, 1.0]
}
)
]
classifiers = [
('LR', LogisticRegression()),
('KNN', KNeighborsClassifier(), {'n_neighbors': [3, 5]}),
('DT', DecisionTreeClassifier(random_state=2), {'max_depth': [3, 6]}),
('GBC', GradientBoostingClassifier(random_state=3), {'max_depth': [3, 6], 'n_estimators': [50, 100]})
]