sklearn学习：make_multilabel_classification——多标签数据集方法

Posted 2020-10-04

Generate a random multilabel classification problem.

For each sample, the generative process is:

• pick the number of labels: n ~ Poisson(n_labels)：选取标签的数目
• n times, choose a class c: c ~ Multinomial(theta) ：n次,选取类别C:多项式
• pick the document length: k ~ Poisson(length) ：选取文档长度
• k times, choose a word: w ~ Multinomial(theta_c)：k次,选取一个单词

In the above process, rejection sampling is used to make sure that n is never zero or more than n_classes, and that the document length is never zero. Likewise, we reject classes which have already been chosen.
在上面的过程中，为确保n不为0或不超过变量n_classes，且文本长度不为0，采用拒绝抽样的方法。同样的，我们拒绝已经选择的类。

Parameters:

n_samples : int, optional (default=100) The number of samples.【生成样本数】 n_features : int, optional (default=20) The total number of features.【每个样本特征数】 n_classes : int, optional (default=5) The number of classes of the classification problem.【分类问题类或标签总数】 n_labels : int, optional (default=2) The average number of labels per instance. More precisely, the number of labels per sample is drawn from a Poisson distribution with n_labels as its expected value, but samples are bounded (usingrejection sampling) by n_classes, and must be nonzero if allow_unlabeled is False. 【每个样本的平均标签数量。更准确地说，每个样本的标签数量是以泊松分布绘制的，其中n_labels为其预期值，但样本是由n_classes限定（使用注射采样），如果allow_unlabeled为False，那么它们必须非零。】 length : int, optional (default=50) The sum of the features (number of words if documents) is drawn from a Poisson distribution with this expected value.【特征的总和（如果是文档，则为单词的数量），从具有该预期值的泊松分布绘制。】 allow_unlabeled : bool, optional (default=True) If True, some instances might not belong to any class.【如果为True，一些样例可能就不属于任何一类】 sparse : bool, optional (default=False) If True, return a sparse feature matrix【如果为True，返回一个稀疏的特征矩阵】 New in version 0.17: parameter to allow sparse output. return_indicator : ‘dense’ (default) | ‘sparse’ | False If dense returnY in the dense binary indicator format. If‘sparse‘ returnY in the sparse binary indicator format.False returns a list of lists of labels. return_distributions : bool, optional (default=False) If True, return the prior class probability and conditional probabilities of features given classes, from which the data wasdrawn.【如果为True，则返回先前的类概率和给定类的特征的条件概率，从中提取数据。】 random_state : int, RandomState instance or None, optional (default=None) If int, random_state is the seed used by the random number generator;If RandomState instance, random_state is the random number generator;If None, the random number generator is the RandomState instance usedbynp.random.【如果数字，random_state是随机数生成器使用的种子；如果是随机状态实例，random_state是随机数生成器；如果为None，则随机数生成器是np.random使用的随机状态实例。】

Returns:

X : array of shape [n_samples, n_features] The generated samples.【返回n_samples行n_features列的训练集】 Y : array or sparse CSR matrix of shape [n_samples, n_classes] The label sets.【n_samples行n_classes列的数组或稀疏CSR阵】 p_c : array, shape [n_classes] The probability of each class being drawn. Only returned if return_distributions=True. p_w_c : array, shape [n_features, n_classes] The probability of each feature being drawn given each class.Only returned if return_distributions=True.

官网教程：

"""
==============================================
Plot randomly generated multilabel dataset【绘制随机生成的多标签数据集】
==============================================
This illustrates the `datasets.make_multilabel_classification` dataset generator. Each sample consists of counts of two features (up to 50 in total), which are differently distributed in each of two classes.Points are labeled as follows, where Y means the class is present:
【数据集生成器“datasets.make_multilabel_classification”说明：】

    =====  =====  =====  ======
      1      2      3    Color
    =====  =====  =====  ======
      Y      N      N    Red
      N      Y      N    Blue
      N      N      Y    Yellow
      Y      Y      N    Purple
      Y      N      Y    Orange
      Y      Y      N    Green
      Y      Y      Y    Brown
    =====  =====  =====  ======
A star marks the expected sample for each class; its size reflects the probability of selecting that class label.【一颗星星标志着每个类标签的预期样本，它的大小反映了
选择该类标签的概率。】
The left and right examples highlight the ``n_labels`` parameter: more of the samples in the right plot have 2 or 3 labels.Note that this two-dimensional example is very degenerate:generally the number of features would be much greater than the "document length", while here we have much larger documents than vocabulary.
Similarly, with ``n_classes > n_features``, it is much less likely that a feature distinguishes a particular class.

【左右两幅图显示“n_labels”的参数；右边的大多数样本有2到3个标签。注意，这个二维的样本是非常退化的：通常，特征的总数比“文本”的总数要多，但是在这里，我们的文本长度大于词汇数。类似地，因为``n_classes(3)> n_features(2)``，特征不太可能区分特定的类】

"""