Source code for imblearn.under_sampling._prototype_selection._edited_nearest_neighbours

"""Class to perform under-sampling based on the edited nearest neighbour
method."""

# Authors: Guillaume Lemaitre <g.lemaitre58@gmail.com>
#          Dayvid Oliveira
#          Christos Aridas
# License: MIT

from __future__ import division

from collections import Counter

import numpy as np
from scipy.stats import mode

from sklearn.utils import safe_indexing

from ..base import BaseCleaningSampler
from ...utils import check_neighbors_object
from ...utils import Substitution
from ...utils.deprecation import deprecate_parameter
from ...utils._docstring import _random_state_docstring

SEL_KIND = ('all', 'mode')


[docs]@Substitution( sampling_strategy=BaseCleaningSampler._sampling_strategy_docstring, random_state=_random_state_docstring) class EditedNearestNeighbours(BaseCleaningSampler): """Class to perform under-sampling based on the edited nearest neighbour method. Read more in the :ref:`User Guide <edited_nearest_neighbors>`. Parameters ---------- {sampling_strategy} return_indices : bool, optional (default=False) Whether or not to return the indices of the samples randomly selected. .. deprecated:: 0.4 ``return_indices`` is deprecated. Use the attribute ``sample_indices_`` instead. {random_state} .. deprecated:: 0.4 ``random_state`` is deprecated in 0.4 and will be removed in 0.6. n_neighbors : int or object, optional (default=3) If ``int``, size of the neighbourhood to consider to compute the nearest neighbors. If object, an estimator that inherits from :class:`sklearn.neighbors.base.KNeighborsMixin` that will be used to find the nearest-neighbors. kind_sel : str, optional (default='all') Strategy to use in order to exclude samples. - If ``'all'``, all neighbours will have to agree with the samples of interest to not be excluded. - If ``'mode'``, the majority vote of the neighbours will be used in order to exclude a sample. n_jobs : int, optional (default=1) The number of threads to open if possible. ratio : str, dict, or callable .. deprecated:: 0.4 Use the parameter ``sampling_strategy`` instead. It will be removed in 0.6. Attributes ---------- sample_indices_ : ndarray, shape (n_new_samples) Indices of the samples selected. .. versionadded:: 0.4 ``sample_indices_`` used instead of ``return_indices=True``. Notes ----- The method is based on [1]_. Supports multi-class resampling. A one-vs.-rest scheme is used when sampling a class as proposed in [1]_. See also -------- CondensedNearestNeighbour, RepeatedEditedNearestNeighbours, AllKNN References ---------- .. [1] D. Wilson, Asymptotic" Properties of Nearest Neighbor Rules Using Edited Data," In IEEE Transactions on Systems, Man, and Cybernetrics, vol. 2 (3), pp. 408-421, 1972. Examples -------- >>> from collections import Counter >>> from sklearn.datasets import make_classification >>> from imblearn.under_sampling import \ EditedNearestNeighbours # doctest: +NORMALIZE_WHITESPACE >>> X, y = make_classification(n_classes=2, class_sep=2, ... weights=[0.1, 0.9], n_informative=3, n_redundant=1, flip_y=0, ... n_features=20, n_clusters_per_class=1, n_samples=1000, random_state=10) >>> print('Original dataset shape %s' % Counter(y)) Original dataset shape Counter({{1: 900, 0: 100}}) >>> enn = EditedNearestNeighbours() >>> X_res, y_res = enn.fit_resample(X, y) >>> print('Resampled dataset shape %s' % Counter(y_res)) Resampled dataset shape Counter({{1: 887, 0: 100}}) """
[docs] def __init__(self, sampling_strategy='auto', return_indices=False, random_state=None, n_neighbors=3, kind_sel='all', n_jobs=1, ratio=None): super(EditedNearestNeighbours, self).__init__( sampling_strategy=sampling_strategy, ratio=ratio) self.random_state = random_state self.return_indices = return_indices self.n_neighbors = n_neighbors self.kind_sel = kind_sel self.n_jobs = n_jobs
def _validate_estimator(self): """Validate the estimator created in the ENN.""" # check for deprecated random_state if self.random_state is not None: deprecate_parameter(self, '0.4', 'random_state') self.nn_ = check_neighbors_object( 'n_neighbors', self.n_neighbors, additional_neighbor=1) self.nn_.set_params(**{'n_jobs': self.n_jobs}) if self.kind_sel not in SEL_KIND: raise NotImplementedError def _fit_resample(self, X, y): if self.return_indices: deprecate_parameter(self, '0.4', 'return_indices', 'sample_indices_') self._validate_estimator() idx_under = np.empty((0, ), dtype=int) self.nn_.fit(X) for target_class in np.unique(y): if target_class in self.sampling_strategy_.keys(): target_class_indices = np.flatnonzero(y == target_class) X_class = safe_indexing(X, target_class_indices) y_class = safe_indexing(y, target_class_indices) nnhood_idx = self.nn_.kneighbors( X_class, return_distance=False)[:, 1:] nnhood_label = y[nnhood_idx] if self.kind_sel == 'mode': nnhood_label, _ = mode(nnhood_label, axis=1) nnhood_bool = np.ravel(nnhood_label) == y_class elif self.kind_sel == 'all': nnhood_label = nnhood_label == target_class nnhood_bool = np.all(nnhood_label, axis=1) index_target_class = np.flatnonzero(nnhood_bool) else: index_target_class = slice(None) idx_under = np.concatenate( (idx_under, np.flatnonzero(y == target_class)[index_target_class]), axis=0) self.sample_indices_ = idx_under if self.return_indices: return (safe_indexing(X, idx_under), safe_indexing(y, idx_under), idx_under) return safe_indexing(X, idx_under), safe_indexing(y, idx_under)
[docs]@Substitution( sampling_strategy=BaseCleaningSampler._sampling_strategy_docstring, random_state=_random_state_docstring) class RepeatedEditedNearestNeighbours(BaseCleaningSampler): """Class to perform under-sampling based on the repeated edited nearest neighbour method. Read more in the :ref:`User Guide <edited_nearest_neighbors>`. Parameters ---------- {sampling_strategy} return_indices : bool, optional (default=False) Whether or not to return the indices of the samples randomly selected. .. deprecated:: 0.4 ``return_indices`` is deprecated. Use the attribute ``sample_indices_`` instead. {random_state} .. deprecated:: 0.4 ``random_state`` is deprecated in 0.4 and will be removed in 0.6. n_neighbors : int or object, optional (default=3) If ``int``, size of the neighbourhood to consider to compute the nearest neighbors. If object, an estimator that inherits from :class:`sklearn.neighbors.base.KNeighborsMixin` that will be used to find the nearest-neighbors. max_iter : int, optional (default=100) Maximum number of iterations of the edited nearest neighbours algorithm for a single run. kind_sel : str, optional (default='all') Strategy to use in order to exclude samples. - If ``'all'``, all neighbours will have to agree with the samples of interest to not be excluded. - If ``'mode'``, the majority vote of the neighbours will be used in order to exclude a sample. n_jobs : int, optional (default=1) The number of thread to open when it is possible. ratio : str, dict, or callable .. deprecated:: 0.4 Use the parameter ``sampling_strategy`` instead. It will be removed in 0.6. Attributes ---------- sample_indices_ : ndarray, shape (n_new_samples) Indices of the samples selected. .. versionadded:: 0.4 ``sample_indices_`` used instead of ``return_indices=True``. Notes ----- The method is based on [1]_. A one-vs.-rest scheme is used when sampling a class as proposed in [1]_. Supports multi-class resampling. See also -------- CondensedNearestNeighbour, EditedNearestNeighbours, AllKNN References ---------- .. [1] I. Tomek, "An Experiment with the Edited Nearest-Neighbor Rule," IEEE Transactions on Systems, Man, and Cybernetics, vol. 6(6), pp. 448-452, June 1976. Examples -------- >>> from collections import Counter >>> from sklearn.datasets import make_classification >>> from imblearn.under_sampling import \ RepeatedEditedNearestNeighbours # doctest : +NORMALIZE_WHITESPACE >>> X, y = make_classification(n_classes=2, class_sep=2, ... weights=[0.1, 0.9], n_informative=3, n_redundant=1, flip_y=0, ... n_features=20, n_clusters_per_class=1, n_samples=1000, random_state=10) >>> print('Original dataset shape %s' % Counter(y)) Original dataset shape Counter({{1: 900, 0: 100}}) >>> renn = RepeatedEditedNearestNeighbours() >>> X_res, y_res = renn.fit_resample(X, y) >>> print('Resampled dataset shape %s' % Counter(y_res)) Resampled dataset shape Counter({{1: 887, 0: 100}}) """
[docs] def __init__(self, sampling_strategy='auto', return_indices=False, random_state=None, n_neighbors=3, max_iter=100, kind_sel='all', n_jobs=1, ratio=None): super(RepeatedEditedNearestNeighbours, self).__init__( sampling_strategy=sampling_strategy, ratio=ratio) self.random_state = random_state self.return_indices = return_indices self.n_neighbors = n_neighbors self.kind_sel = kind_sel self.n_jobs = n_jobs self.max_iter = max_iter
def _validate_estimator(self): """Private function to create the NN estimator""" # check for deprecated random_state if self.random_state is not None: deprecate_parameter(self, '0.4', 'random_state') if self.max_iter < 2: raise ValueError('max_iter must be greater than 1.' ' Got {} instead.'.format(type(self.max_iter))) self.nn_ = check_neighbors_object( 'n_neighbors', self.n_neighbors, additional_neighbor=1) self.enn_ = EditedNearestNeighbours( sampling_strategy=self.sampling_strategy, return_indices=False, n_neighbors=self.nn_, kind_sel=self.kind_sel, n_jobs=self.n_jobs, ratio=self.ratio) def _fit_resample(self, X, y): if self.return_indices: deprecate_parameter(self, '0.4', 'return_indices', 'sample_indices_') self._validate_estimator() X_, y_ = X, y self.sample_indices_ = np.arange(X.shape[0], dtype=int) target_stats = Counter(y) class_minority = min(target_stats, key=target_stats.get) for n_iter in range(self.max_iter): prev_len = y_.shape[0] X_enn, y_enn = self.enn_.fit_resample(X_, y_) # Check the stopping criterion # 1. If there is no changes for the vector y # 2. If the number of samples in the other class become inferior to # the number of samples in the majority class # 3. If one of the class is disappearing # Case 1 b_conv = (prev_len == y_enn.shape[0]) # Case 2 stats_enn = Counter(y_enn) count_non_min = np.array([ val for val, key in zip(stats_enn.values(), stats_enn.keys()) if key != class_minority ]) b_min_bec_maj = np.any( count_non_min < target_stats[class_minority]) # Case 3 b_remove_maj_class = (len(stats_enn) < len(target_stats)) X_, y_, = X_enn, y_enn self.sample_indices_ = self.sample_indices_[ self.enn_.sample_indices_] if b_conv or b_min_bec_maj or b_remove_maj_class: if b_conv: X_, y_, = X_enn, y_enn self.sample_indices_ = self.sample_indices_[ self.enn_.sample_indices_] break X_resampled, y_resampled = X_, y_ if self.return_indices: return X_resampled, y_resampled, self.sample_indices_ return X_resampled, y_resampled
[docs]@Substitution( sampling_strategy=BaseCleaningSampler._sampling_strategy_docstring, random_state=_random_state_docstring) class AllKNN(BaseCleaningSampler): """Class to perform under-sampling based on the AllKNN method. Read more in the :ref:`User Guide <edited_nearest_neighbors>`. Parameters ---------- {sampling_strategy} return_indices : bool, optional (default=False) Whether or not to return the indices of the samples randomly selected. .. deprecated:: 0.4 ``return_indices`` is deprecated. Use the attribute ``sample_indices_`` instead. {random_state} .. deprecated:: 0.4 ``random_state`` is deprecated in 0.4 and will be removed in 0.6. n_neighbors : int or object, optional (default=3) If ``int``, size of the neighbourhood to consider to compute the nearest neighbors. If object, an estimator that inherits from :class:`sklearn.neighbors.base.KNeighborsMixin` that will be used to find the nearest-neighbors. kind_sel : str, optional (default='all') Strategy to use in order to exclude samples. - If ``'all'``, all neighbours will have to agree with the samples of interest to not be excluded. - If ``'mode'``, the majority vote of the neighbours will be used in order to exclude a sample. allow_minority : bool, optional (default=False) If ``True``, it allows the majority classes to become the minority class without early stopping. .. versionadded:: 0.3 n_jobs : int, optional (default=1) The number of thread to open when it is possible. ratio : str, dict, or callable .. deprecated:: 0.4 Use the parameter ``sampling_strategy`` instead. It will be removed in 0.6. Attributes ---------- sample_indices_ : ndarray, shape (n_new_samples) Indices of the samples selected. .. versionadded:: 0.4 ``sample_indices_`` used instead of ``return_indices=True``. Notes ----- The method is based on [1]_. Supports multi-class resampling. A one-vs.-rest scheme is used when sampling a class as proposed in [1]_. See also -------- CondensedNearestNeighbour, EditedNearestNeighbours, RepeatedEditedNearestNeighbours References ---------- .. [1] I. Tomek, "An Experiment with the Edited Nearest-Neighbor Rule," IEEE Transactions on Systems, Man, and Cybernetics, vol. 6(6), pp. 448-452, June 1976. Examples -------- >>> from collections import Counter >>> from sklearn.datasets import make_classification >>> from imblearn.under_sampling import \ AllKNN # doctest: +NORMALIZE_WHITESPACE >>> X, y = make_classification(n_classes=2, class_sep=2, ... weights=[0.1, 0.9], n_informative=3, n_redundant=1, flip_y=0, ... n_features=20, n_clusters_per_class=1, n_samples=1000, random_state=10) >>> print('Original dataset shape %s' % Counter(y)) Original dataset shape Counter({{1: 900, 0: 100}}) >>> allknn = AllKNN() >>> X_res, y_res = allknn.fit_resample(X, y) >>> print('Resampled dataset shape %s' % Counter(y_res)) Resampled dataset shape Counter({{1: 887, 0: 100}}) """
[docs] def __init__(self, sampling_strategy='auto', return_indices=False, random_state=None, n_neighbors=3, kind_sel='all', allow_minority=False, n_jobs=1, ratio=None): super(AllKNN, self).__init__( sampling_strategy=sampling_strategy, ratio=ratio) self.random_state = random_state self.return_indices = return_indices self.n_neighbors = n_neighbors self.kind_sel = kind_sel self.allow_minority = allow_minority self.n_jobs = n_jobs
def _validate_estimator(self): """Create objects required by AllKNN""" # check for deprecated random_state if self.random_state is not None: deprecate_parameter(self, '0.4', 'random_state') if self.kind_sel not in SEL_KIND: raise NotImplementedError self.nn_ = check_neighbors_object( 'n_neighbors', self.n_neighbors, additional_neighbor=1) self.enn_ = EditedNearestNeighbours( sampling_strategy=self.sampling_strategy, return_indices=False, n_neighbors=self.nn_, kind_sel=self.kind_sel, n_jobs=self.n_jobs, ratio=self.ratio) def _fit_resample(self, X, y): if self.return_indices: deprecate_parameter(self, '0.4', 'return_indices', 'sample_indices_') self._validate_estimator() X_, y_ = X, y target_stats = Counter(y) class_minority = min(target_stats, key=target_stats.get) self.sample_indices_ = np.arange(X.shape[0], dtype=int) for curr_size_ngh in range(1, self.nn_.n_neighbors): self.enn_.n_neighbors = curr_size_ngh X_enn, y_enn = self.enn_.fit_resample(X_, y_) # Check the stopping criterion # 1. If the number of samples in the other class become inferior to # the number of samples in the majority class # 2. If one of the class is disappearing # Case 1else: stats_enn = Counter(y_enn) count_non_min = np.array([ val for val, key in zip(stats_enn.values(), stats_enn.keys()) if key != class_minority ]) b_min_bec_maj = np.any( count_non_min < target_stats[class_minority]) if self.allow_minority: # overwrite b_min_bec_maj b_min_bec_maj = False # Case 2 b_remove_maj_class = (len(stats_enn) < len(target_stats)) X_, y_, = X_enn, y_enn self.sample_indices_ = self.sample_indices_[ self.enn_.sample_indices_] if b_min_bec_maj or b_remove_maj_class: break X_resampled, y_resampled = X_, y_ if self.return_indices: return X_resampled, y_resampled, self.sample_indices_ return X_resampled, y_resampled