NMI（Normalized Mutual Information）用于衡量通过特征选择得到的子集与原始数据类别之间的关联度python代码完整示例

在Python中，你可以使用`scikit-learn`库中的`normalized_mutual_info_score`函数来计算NMI。以下是一个完整的示例，假设我们有一个特征矩阵和对应的目标标签：

from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import mutual_info_classif
from sklearn.preprocessing import OneHotEncoder
from sklearn.metrics import normalized_mutual_info_score
import pandas as pd

# 假设你有特征数据（DataFrame）
data = pd.DataFrame(..., columns=['feature1', 'feature2', 'feature3'])  # 各列是特征名
# 和对应的类别标签（一维数组）
labels = ...  # 类别标签列表，例如 ['class_0', 'class_1', 'class_2']

# 将类别标签转换为one-hot编码
encoder = OneHotEncoder(sparse=False)
labels_encoded = encoder.fit_transform(labels.reshape(-1, 1)).astype(int)

# 使用SelectKBest方法选择K个特征
selector = SelectKBest(mutual_info_classif, k=5)  # 这里使用了mutual_info_classif作为评分函数
features_selected = selector.fit_transform(data, labels)

# 计算NMI
selected_features_nmi = normalized_mutual_info_score(labels_encoded, selector.get_support())

print(f"NMI between selected features and original classes: {selected_features_nmi}")