基于信息熵和逻辑回归的特征提取分类python源码.zip

""" -*- coding: utf-8 -*- Desc: 信息熵算法特征提取 Intro：使用信息熵算法对数据集进行特征提取，选出最具有代表性的特征，从而达到降维的目的，便于之后的分类操作 Version：v1.0 Time：2023.7.6 """ import pandas as pd # 引用pandas库 import numpy as np # 引用numpy库 # 数据处理 def main_processing(data, m, n): data_feature = np.zeros(shape=(6, n)) # 初始化特征二维矩阵 for i in range(n): # 对数据集中的每列 data_ave = np.mean(data[:, i]) # 求均值 data_std = np.std(data[:, i]) # 求标准差 data_max = np.max(data[:, i]) # 求最大值 data_min = np.min(data[:, i]) # 求最小值 data_energy = np.sum(np.abs(data[:, i])) # 求能量（数据绝对值之和） data_normal = (data[:, i] - data_min) / (data_max - data_min) # 数据归一化(0,1) segment = int(0.5 * m) data_entropy = entropy(data_normal, segment) # 求信息熵 data_feature[:, i] = [data_ave, data_std, data_max, data_min, data_energy, data_entropy] # 特征二维数组 return data_feature # 结果写入表格 def output_to_file(data_feature): data_f = pd.DataFrame(data_feature) # 写入数据 data_f.index = ['ave', 'std', 'max', 'min', 'energy', 'entropy'] # 行标题 writer = pd.ExcelWriter('./result/SCZ/SCZ_f21.xlsx') # 写入路径 data_f.to_excel(writer, 'data_feature', float_format='%.20f') # data_feature为表格sheet名，float_format为数值精度 writer.save() # 保存 # 初始化 def initial(): csv_file_path = './test/SCZ/s01_block021_SCZ.csv' # 导入数据地址 raw_data = pd.read_csv(csv_file_path) # 导入数据 height, width = raw_data.shape # 获得导入表格的高度和宽度 raw_data = raw_data.values # 取值，且转化为二维数组 data = np.array(raw_data) # 转化为二维数组(矩阵)，转化为numpy数据标准 ,不改变维度 (m, n) = raw_data.shape # 矩阵的行数m列数n，返回一个元组 print(height, width, type(raw_data)) return [data, m, n] # 求信息熵 def entropy(vector, segment): # 自定义一个求解信息熵的函数，vector为向量，segment分段数值 x_min = np.min(vector) x_max = np.max(vector) x_dis = np.abs(x_max - x_min) x_lower = x_min seg = 1.0 / segment internal = x_dis * seg basic_list = [] full_list = [] # for i in range(len(vector)): if vector[i] >= x_lower + internal: basic_list.append(vector[i]) length_list = len(basic_list) full_list.append(length_list) # for j in range(1, segment): basic_list = [] for i in range(len(vector)): if x_lower + j * internal <= vector[i] < x_lower + (j + 1) * internal: basic_list.append(vector[i]) length_list = len(basic_list) full_list.append(length_list) # basic_list = [] for i in range(len(vector)): if vector[i] >= x_lower + (segment - 1) * internal: basic_list.append(vector[i]) length_list = len(basic_list) full_list.append(length_list) full_list = full_list / np.sum(full_list) # y = 0 full_y = [] for i in range(segment): if full_list[i] == 0: y = 0 full_y.append(y) else: y = -full_list[i] * np.log2(full_list[i]) full_y.append(y) result = np.sum(full_y) return result # 主函数 if __name__ == '__main__': # 导入数据及初始化 main_data, data_height, data_width = initial() # 求信息熵等主要处理 features = main_processing(main_data, data_height, data_width) # 写出数据到表格文件 output_to_file(features)