数据不均衡时的过采样代码实现----SMOTE

首先，要明确数据不均衡是指正例：反例差别大于10倍时及以上
其次，本次代码实现的是简单的SMOTE模型，不涉及SMOTE优化
下面是python代码实现：
数据使用的是系统自带的数据（load_iris）

##首先，创建类
class SMOTE():
    """
    K_neighbors：查找目标对象的几个邻近值
    N_need：将数据扩充几倍
    random_state：随机因子，使每次随机结果相同
    """
    def __init__(self, 
                 K_neighbors = 5,
                 N_need = 3,
                 random_state = 42):
        self.K_neighbors = K_neighbors
        self.N_need = N_need
        self.random_state = 42
 
 
    def div_data(self, x_data, y_label):
        """
        将数据分为少数样本，多数样本，少数样本对应label值，多数样本对应label值，注意样本数据包含label标签值
        """
#         print('y_label',y_label)
        tp = set(y_label)
#         print('tp',tp)
        #找出结果值出现次数较少的一类
        tp_less = [a for a in tp if sum(y_label == a) < sum(y_label != a)][0]
#         print('tp_less',tp_less)
        data_less = x_data.iloc[y_label == tp_less,:]
#         print('tp',tp)
#         print('data_less',data_less)
        data_more = x_data.iloc[y_label != tp_less, :]
        tp.remove(tp_less)
        return data_less, data_more, tp_less, list(tp)[0]

    def get_SMOTE_sample(self, x_data, y_label):
        """
        复制少数样本，并追加到少数样本数据中
        """
        data_less, data_more, tp_less, tp_more = self.div_data(x_data, y_label)

        n_integ = self.N_need
        data_add = copy.deepcopy(data_less)
#         print('data_less', data_less.shape)
        if n_integ == 0 :
            print('WARNING: PLEASE RE-ENTER N_need')
        else:
            for i in range(1,n_integ):##扩充少数类的倍数
                data_less = data_less.append(data_add)
        data_less.reset_index(inplace = True, drop = True)
#         print('data_out', data_less.shape)
        return data_less, tp_less
    
    
    def over_sample(self, x_data, y_label):
        """
        SMOTE算法简单计算公式：
        new_sample = sample[i] + random*(sample[i]-sample.any.neighbor)
        """
        sample, tp_less = self.get_SMOTE_sample(x_data, y_label)        
        knn = NearestNeighbors(n_neighbors = self.K_neighbors ,n_jobs = -1).fit(sample)
        
        n_atters = x_data.shape[1]
        label_out = copy.deepcopy(y_label)
        new = pd.DataFrame(columns = x_data.columns)
#         print('new', new)

        for i in range(len(sample)): # 1. 选择一个正样本
            # 2.选择少数类中最近的K个样本
            k_sample_index = knn.kneighbors(np.array(sample.iloc[i, :]).reshape(1, -1),
                                            n_neighbors = self.K_neighbors + 1,
                                            return_distance = False)
#             print('k_sample_index',type(k_sample_index),k_sample_index)
#             print('np.array(sample.iloc[i, :]).reshape(1, -1)', np.array(sample.iloc[i, :]).reshape(1, -1))
 
            # 计算插值样本
            # 3.随机选取K中的一个样本
            np.random.seed(self.random_state)
            choice_all = k_sample_index.flatten()
#             print('choice_all',choice_all)
#             print('choice_all[choice_all != 0]',choice_all != 0)
            choosed = np.random.choice(choice_all[choice_all != 0])
#             print('choosed',choosed)

            # 4. 在正样本和随机样本之间选出一个点
            diff = sample.iloc[choosed,] - sample.iloc[i,]
#             print('diff',type(diff), diff)
            gap = np.random.rand(1, n_atters)
#             print('gap', gap)
#             print('sample.iloc[i,]', sample.iloc[i,])
            new.loc[i] = [x for x in sample.iloc[i,] + gap.flatten() * diff]
#             print('new',new)
            
            label_out = np.r_[label_out, tp_less]##给新增加的一行数据添加label标签
        print('new', new)
        new_sample = pd.concat([x_data, new])
        new_sample.reset_index(inplace = True, drop = True)
        return new_sample, label_out

from sklearn.neighbors import NearestNeighbors
import numpy as np 
import pandas as pd 
import copy
from sklearn.datasets import load_iris

##其次，将数据data处理成多行*5列, label处理成多行*1列形式。
iris= load_iris()
data = pd.DataFrame(data = iris.data, columns = iris.feature_names)
data['label'] = iris.target
data0 = data.loc[data['label'] == 0,]
data1 = data.loc[data['label'] == 1,][:10]
data = pd.concat([data0, data1],axis = 0)
data.reset_index(inplace=True,drop=True)
label = np.r_[label[label == 0], label[label == 1][:10]]##np.r_函数作用是将两个数组首尾相连，类似df.append操作
##最后，使用类
smt = SMOTE()###实例化类
new_sample, new_label = smt.over_sample(data, label)