Question

当我使用线性SVM处理scikit-learn中的分类问题时，我可以将自定义权重应用于每个训练样本，如下所示：

from sklearn.linear_model import SGDClassifier

X = [[0.0, 0.0], [1.0, 1.0]]
y = [0, 1]
sample_weight = [1.0, 0.5]
clf = SGDClassifier(loss="hinge")
clf.fit(X, y, sample_weight=sample_weight)

现在，当我有一个多标签分类任务时，我需要转换标签，而SGDClassifier必须包含在OneVsRestClassifier之类的元估算中：

from sklearn.preprocessing import MultiLabelBinarizer
from sklearn.multiclass import OneVsRestClassifier
from sklearn.linear_model import SGDClassifier

X = [[0.0, 0.0], [1.0, 1.0], [1.0, 0.0]]
y = [[0], [1], [0, 1]]
y_mlb = MultiLabelBinarizer().fit_transform(y)
sample_weight = [1.0, 0.5, 0.8]
clf = OneVsRestClassifier(SGDClassifier(loss="hinge"))
clf.fit(X, y_mlb)   # unable to pass `sample_weight`

但是，除了OneVsRestClassifier和fit之外，X不允许我将任何参数传递给y方法，因此我无法应用样本权重，就像我之前做的那样。如何在这种情况下应用我自己的样本权重？

Answer 1

而是尝试子类化OneVsRestClassifier来更改fit方法以允许传递sample_weight。您需要更改其中使用的fit（）和_fit_binary（）方法。

尝试将source from here编辑为：

import warnings
import numpy as np
from sklearn.externals.joblib import Parallel, delayed
from sklearn.base import clone
from sklearn.multiclass import _ConstantPredictor, OneVsRestClassifier
from sklearn.preprocessing import LabelBinarizer, MultiLabelBinarizer
from sklearn.linear_model import SGDClassifier

def _fit_binary_new(estimator, X, y, sample_weight, classes=None):
    unique_y = np.unique(y)
    if len(unique_y) == 1:
        if classes is not None:
            if y[0] == -1:
                c = 0
            else:
                c = y[0]
            warnings.warn("Label %s is present in all training examples." %
                          str(classes[c]))
        estimator = _ConstantPredictor().fit(X, unique_y)
    else:
        estimator = clone(estimator)

        # Only this changed
        estimator.fit(X, y, sample_weight=sample_weight)
    return estimator

class OneVsRestClassifierNew(OneVsRestClassifier):

    def fit(self, X, y, sample_weight=None):
        self.label_binarizer_ = LabelBinarizer(sparse_output=True)
        Y = self.label_binarizer_.fit_transform(y)
        Y = Y.tocsc()
        self.classes_ = self.label_binarizer_.classes_
        columns = (col.toarray().ravel() for col in Y.T)
        self.estimators_ = Parallel(n_jobs=self.n_jobs)(delayed(_fit_binary_new)(
            self.estimator, X, column, sample_weight, classes=[
                "not %s" % self.label_binarizer_.classes_[i],
                self.label_binarizer_.classes_[i]])
            for i, column in enumerate(columns))

        return self


X = [[0.0, 0.0], [1.0, 1.0], [1.0, 0.0]]
y = [[0], [1], [0, 1]]
y_mlb = MultiLabelBinarizer().fit_transform(y)
sample_weight = [1.0, 0.5, 0.8]
clf = OneVsRestClassifierNew(SGDClassifier(loss="hinge"))
clf.fit(X, y_mlb, sample_weight=sample_weight)

clf.predict(X)
# Output: array([[1, 0],
#                [0, 1],
#                [1, 1]])

注意：这只适用于那些在fit（）方法中定义了sample_weight的分类器，因为我没有检查_fit_binary_new()中是否存在。

在多标签分类中使用样品重量

1 个答案: