EHN accept one-vs-all encoding for labels (#410)

Author: glemaitre

doc/whats_new/v0.0.4.rst

@@ -15,6 +15,9 @@ Enhancement
 - Document the metrics to evaluate models on imbalanced dataset. :issue:`367`
   by :user:`Guillaume Lemaitre <glemaitre>`.
 
+- Add support for one-vs-all encoded target to support keras. :issue:`409` by
+  :user:`Guillaume Lemaitre <glemaitre>`.
+
 Bug fixes
 .........
 

imblearn/base.py

@@ -9,8 +9,11 @@
 import logging
 from abc import ABCMeta, abstractmethod
 
+import numpy as np
+
 from sklearn.base import BaseEstimator
 from sklearn.externals import six
+from sklearn.preprocessing import label_binarize
 from sklearn.utils import check_X_y
 from sklearn.utils.validation import check_is_fitted
 
@@ -54,14 +57,23 @@ def sample(self, X, y):
             The corresponding label of `X_resampled`
 
         """
-
         # Check the consistency of X and y
+        y, binarize_y = check_target_type(y, indicate_one_vs_all=True)
         X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
 
         check_is_fitted(self, 'ratio_')
         self._check_X_y(X, y)
 
-        return self._sample(X, y)
+        output = self._sample(X, y)
+
+        if binarize_y:
+            y_sampled = label_binarize(output[1], np.unique(y))
+            if len(output) == 2:
+                return output[0], y_sampled
+            else:
+                return output[0], y_sampled, output[2]
+        else:
+            return output
 
     def fit_sample(self, X, y):
         """Fit the statistics and resample the data directly.
@@ -152,8 +164,8 @@ def fit(self, X, y):
             Return self.
 
         """
-        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
         y = check_target_type(y)
+        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
         self.X_hash_, self.y_hash_ = hash_X_y(X, y)
         # self.sampling_type is already checked in check_ratio
         self.ratio_ = check_ratio(self.ratio, y, self._sampling_type)
@@ -232,25 +244,20 @@ def __init__(self, func=None, accept_sparse=True, kw_args=None):
         self.kw_args = kw_args
         self.logger = logging.getLogger(__name__)
 
-    def _check_X_y(self, X, y):
-        if self.accept_sparse:
-            X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
-        else:
-            X, y = check_X_y(X, y, accept_sparse=False)
-        y = check_target_type(y)
-
-        return X, y
-
     def fit(self, X, y):
-        X, y = self._check_X_y(X, y)
+        y = check_target_type(y)
+        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc']
+                         if self.accept_sparse else False)
         self.X_hash_, self.y_hash_ = hash_X_y(X, y)
         # when using a sampler, ratio_ is supposed to exist after fit
         self.ratio_ = 'is_fitted'
 
         return self
 
     def _sample(self, X, y, func=None, kw_args=None):
-        X, y = self._check_X_y(X, y)
+        y, binarize_y = check_target_type(y, indicate_one_vs_all=True)
+        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc']
+                         if self.accept_sparse else False)
         check_is_fitted(self, 'ratio_')
         X_hash, y_hash = hash_X_y(X, y)
         if self.X_hash_ != X_hash or self.y_hash_ != y_hash:
@@ -259,7 +266,16 @@ def _sample(self, X, y, func=None, kw_args=None):
         if func is None:
             func = _identity
 
-        return func(X, y, **(kw_args if self.kw_args else {}))
+        output = func(X, y, **(kw_args if self.kw_args else {}))
+
+        if binarize_y:
+            y_sampled = label_binarize(output[1], np.unique(y))
+            if len(output) == 2:
+                return output[0], y_sampled
+            else:
+                return output[0], y_sampled, output[2]
+        else:
+            return output
 
     def sample(self, X, y):
         return self._sample(X, y, func=self.func, kw_args=self.kw_args)

imblearn/combine/smote_enn.py

@@ -144,8 +144,8 @@ def fit(self, X, y):
             Return self.
 
         """
-        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
         y = check_target_type(y)
+        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
         self.ratio_ = self.ratio
         self.X_hash_, self.y_hash_ = hash_X_y(X, y)
 

imblearn/combine/smote_tomek.py

@@ -8,7 +8,6 @@
 from __future__ import division
 
 import logging
-import warnings
 
 from sklearn.utils import check_X_y
 
@@ -153,8 +152,8 @@ def fit(self, X, y):
             Return self.
 
         """
-        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
         y = check_target_type(y)
+        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
         self.ratio_ = self.ratio
         self.X_hash_, self.y_hash_ = hash_X_y(X, y)
 

imblearn/ensemble/balance_cascade.py

@@ -14,7 +14,7 @@
 from sklearn.model_selection import cross_val_predict
 
 from .base import BaseEnsembleSampler
-from ..utils import check_ratio
+from ..utils import check_ratio, check_target_type
 
 
 class BalanceCascade(BaseEnsembleSampler):
@@ -137,6 +137,7 @@ def fit(self, X, y):
 
         """
         super(BalanceCascade, self).fit(X, y)
+        y = check_target_type(y)
         self.ratio_ = check_ratio(self.ratio, y, 'under-sampling')
         return self
 

imblearn/ensemble/base.py

@@ -4,7 +4,14 @@
 # Authors: Guillaume Lemaitre <g.lemaitre58@gmail.com>
 # License: MIT
 
+import numpy as np
+
+from sklearn.preprocessing import label_binarize
+from sklearn.utils import check_X_y
+from sklearn.utils.validation import check_is_fitted
+
 from ..base import BaseSampler
+from ..utils import check_target_type
 
 
 class BaseEnsembleSampler(BaseSampler):
@@ -15,3 +22,46 @@ class BaseEnsembleSampler(BaseSampler):
     """
 
     _sampling_type = 'ensemble'
+
+    def sample(self, X, y):
+        """Resample the dataset.
+
+        Parameters
+        ----------
+        X : {array-like, sparse matrix}, shape (n_samples, n_features)
+            Matrix containing the data which have to be sampled.
+
+        y : array-like, shape (n_samples,)
+            Corresponding label for each sample in X.
+
+        Returns
+        -------
+        X_resampled : {ndarray, sparse matrix}, shape \
+(n_subset, n_samples_new, n_features)
+            The array containing the resampled data.
+
+        y_resampled : ndarray, shape (n_subset, n_samples_new)
+            The corresponding label of `X_resampled`
+
+        """
+        # Ensemble are a bit specific since they are returning an array of
+        # resampled arrays.
+        y, binarize_y = check_target_type(y, indicate_one_vs_all=True)
+        X, y = check_X_y(X, y, accept_sparse=['csr', 'csc'])
+
+        check_is_fitted(self, 'ratio_')
+        self._check_X_y(X, y)
+
+        output = self._sample(X, y)
+
+        if binarize_y:
+            y_resampled = output[1]
+            classes = np.unique(y)
+            y_resampled_encoded = np.array([label_binarize(batch_y, classes)
+                                            for batch_y in y_resampled])
+            if len(output) == 2:
+                return output[0], y_resampled_encoded
+            else:
+                return output[0], y_resampled_encoded, output[2]
+        else:
+            return output

imblearn/utils/estimator_checks.py

@@ -19,11 +19,13 @@
 
 from sklearn.datasets import make_classification
 from sklearn.cluster import KMeans
+from sklearn.preprocessing import label_binarize
 from sklearn.utils.estimator_checks import check_estimator \
     as sklearn_check_estimator, check_parameters_default_constructible
 from sklearn.exceptions import NotFittedError
 from sklearn.utils.testing import assert_allclose
 from sklearn.utils.testing import set_random_state
+from sklearn.utils.multiclass import type_of_target
 
 from imblearn.over_sampling.base import BaseOverSampler
 from imblearn.under_sampling.base import BaseCleaningSampler, BaseUnderSampler
@@ -44,6 +46,7 @@ def _yield_sampler_checks(name, Estimator):
     yield check_samplers_ratio_fit_sample
     yield check_samplers_sparse
     yield check_samplers_pandas
+    yield check_samplers_multiclass_ova
 
 
 def _yield_all_checks(name, estimator):
@@ -253,3 +256,23 @@ def check_samplers_pandas(name, Sampler):
         X_res, y_res = sampler.fit_sample(X, y)
         assert_allclose(X_res_pd, X_res)
         assert_allclose(y_res_pd, y_res)
+
+
+def check_samplers_multiclass_ova(name, Sampler):
+    # Check that multiclass target lead to the same results than OVA encoding
+    X, y = make_classification(n_samples=1000, n_classes=3,
+                               n_informative=4, weights=[0.2, 0.3, 0.5],
+                               random_state=0)
+    y_ova = label_binarize(y, np.unique(y))
+    sampler = Sampler()
+    set_random_state(sampler)
+    X_res, y_res = sampler.fit_sample(X, y)
+    X_res_ova, y_res_ova = sampler.fit_sample(X, y_ova)
+    assert_allclose(X_res, X_res_ova)
+    if issubclass(Sampler, BaseEnsembleSampler):
+        for batch_y, batch_y_ova in zip(y_res, y_res_ova):
+            assert type_of_target(batch_y_ova) == type_of_target(y_ova)
+            assert_allclose(batch_y, batch_y_ova.argmax(axis=1))
+    else:
+        assert type_of_target(y_res_ova) == type_of_target(y_ova)
+        assert_allclose(y_res, y_res_ova.argmax(axis=1))

imblearn/utils/tests/test_validation.py

@@ -6,17 +6,20 @@
 from collections import Counter
 
 import numpy as np
+import pytest
 from pytest import raises
 
 from sklearn.neighbors.base import KNeighborsMixin
 from sklearn.neighbors import NearestNeighbors
 from sklearn.utils import check_random_state
 from sklearn.externals import joblib
+from sklearn.utils.testing import assert_array_equal
 
 from imblearn.utils.testing import warns
 from imblearn.utils import check_neighbors_object
 from imblearn.utils import check_ratio
 from imblearn.utils import hash_X_y
+from imblearn.utils import check_target_type
 
 
 def test_check_neighbors_object():
@@ -35,6 +38,36 @@ def test_check_neighbors_object():
         check_neighbors_object(name, n_neighbors)
 
 
+@pytest.mark.parametrize(
+    "target, output_target",
+    [(np.array([0, 1, 1]), np.array([0, 1, 1])),
+     (np.array([0, 1, 2]), np.array([0, 1, 2])),
+     (np.array([[0, 1], [1, 0]]), np.array([1, 0]))]
+)
+def test_check_target_type(target, output_target):
+    converted_target = check_target_type(target.astype(int))
+    assert_array_equal(converted_target, output_target.astype(int))
+
+
+@pytest.mark.parametrize(
+    "target, output_target, is_ova",
+    [(np.array([0, 1, 1]), np.array([0, 1, 1]), False),
+     (np.array([0, 1, 2]), np.array([0, 1, 2]), False),
+     (np.array([[0, 1], [1, 0]]), np.array([1, 0]), True)]
+)
+def test_check_target_type_ova(target, output_target, is_ova):
+    converted_target, binarize_target = check_target_type(
+        target.astype(int), indicate_one_vs_all=True)
+    assert_array_equal(converted_target, output_target.astype(int))
+    assert binarize_target == is_ova
+
+
+def test_check_target_warning():
+    target = np.arange(4).reshape((2, 2))
+    with pytest.warns(UserWarning, match='should be of types'):
+        check_target_type(target)
+
+
 def test_check_ratio_error():
     with raises(ValueError, match="'sampling_type' should be one of"):
         check_ratio('auto', np.array([1, 2, 3]), 'rnd')

imblearn/utils/validation.py

@@ -19,7 +19,7 @@
 
 SAMPLING_KIND = ('over-sampling', 'under-sampling', 'clean-sampling',
                  'ensemble')
-TARGET_KIND = ('binary', 'multiclass')
+TARGET_KIND = ('binary', 'multiclass', 'multilabel-indicator')
 
 
 def check_neighbors_object(nn_name, nn_object, additional_neighbor=0):
@@ -54,29 +54,42 @@ def check_neighbors_object(nn_name, nn_object, additional_neighbor=0):
         raise_isinstance_error(nn_name, [int, KNeighborsMixin], nn_object)
 
 
-def check_target_type(y):
+def check_target_type(y, indicate_one_vs_all=False):
     """Check the target types to be conform to the current samplers.
 
-    The current samplers should be compatible with ``'binary'`` and
-    ``'multiclass'`` targets only.
+    The current samplers should be compatible with ``'binary'``,
+    ``'multilabel-indicator'`` and ``'multiclass'`` targets only.
 
     Parameters
     ----------
     y : ndarray,
-        The array containing the target
+        The array containing the target.
+
+    indicate_one_vs_all : bool, optional
+        Either to indicate if the targets are encoded in a one-vs-all fashion.
 
     Returns
     -------
     y : ndarray,
         The returned target.
 
+    is_one_vs_all : bool, optional
+        Indicate if the target was originally encoded in a one-vs-all fashion.
+        Only returned if ``indicate_multilabel=True``.
+
     """
-    if type_of_target(y) not in TARGET_KIND:
+    type_y = type_of_target(y)
+    if type_y not in TARGET_KIND:
         # FIXME: perfectly we should raise an error but the sklearn API does
         # not allow for it
         warnings.warn("'y' should be of types {} only. Got {} instead.".format(
             TARGET_KIND, type_of_target(y)))
-    return y
+
+    if indicate_one_vs_all:
+        return (y.argmax(axis=1) if type_y == 'multilabel-indicator' else y,
+                type_y == 'multilabel-indicator')
+    else:
+        return y.argmax(axis=1) if type_y == 'multilabel-indicator' else y
 
 
 def hash_X_y(X, y, n_samples=10, n_features=5):