ReliefF算法代码，并展示结果

ReliefF算法是一种特征选择方法，它通过迭代的方式评估每个特征对于样本分类的重要性。这个算法基于信息增益和距离的概念，旨在找出那些能够最好地区分正负类别的特征。以下是 ReliefF 算法的一个简单Python实现：

import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

class ReliefF:
    def __init__(self, n_features_to_select=10):
        self.n_features_to_select = n_features_to_select
        self.distance_metric = "euclidean"  # 可选的距离度量，如“euclidean”或“manhattan”

    def fit(self, X, y):
        self.X_train, self.y_train = X, y
        self.feature_importances_ = np.zeros(X.shape[1])
        self.iterate()

    def iterate(self):
        for _ in range(self.n_features_to_select * len(self.X_train)):
            sample = self.get_random_sample()
            neighbor_classes = self.find_neighbors(sample)
            self.update_feature_importance(sample, neighbor_classes)

    def get_random_sample(self):
        return np.random.choice(self.X_train, replace=True)

    def find_neighbors(self, sample):
        distances = [self.distance_metric(sample, x) for x in self.X_train]
        indices = np.argsort(distances)[:len(distances)//2] + len(distances)//2
        return [self.y_train[i] for i in indices]

    def update_feature_importance(self, sample, classes):
        weights = [self.compute_weight(sample, c) for c in classes]
        delta_weights = [weights[0] - w for w in weights]
        feature_importances = self.feature_importances_
        feature_importances += delta_weights / (len(classes) * 2)
        
    def compute_weight(self, sample, class_label):
        # 计算正误样本的差异权重
        pos_distance = min([self.distance_metric(sample, s) for s in self.X_train if self.y_train[sample_idx] != class_label])
        neg_distance = max([self.distance_metric(sample, s) for s in self.X_train if self.y_train[sample_idx] == class_label])
        return np.exp(-(pos_distance + neg_distance))

# 示例使用
iris = load_iris()
X_train, X_test, y_train, y_test = train_test_split(iris.data, iris.target, test_size=0.2, random_state=42)
relieff = ReliefF(n_features_to_select=2)
relieff.fit(X_train, y_train)
top_features = np.argsort(relieff.feature_importances_)[-2:][::-1]
print("Top 2 features based on ReliefF:")
for i, feature in enumerate(iris.feature_names[top_features]):
    print(f"{i+1}. {feature} ({relieff.feature_importances_[top_features[i]]:.2f})")