one-hot encoding 对于一个特征包含多个特征id的一种处理方法

from sklearn.preprocessing import OneHotEncoder
import numpy
onehot = OneHotEncoder()
#建立一个映射，将多个特征共存的情况定义为单个数表示比如有特征{a，b}，a用1,b用2,ab用3----（1）
import pandas
data = pandas.read_csv("userFeature_part_bat.csv")

#列出所有特征值，遍历,uid是用户唯一标识，不应该算作特征

userfeature = ["age","carrier","consumptionAbility","ct","education","gender","house","interest1","interest2","interest3","interest4","interest5","kw1","kw2","kw3","marriageStatus","os","topic1","topic2","topic3","LBS","appIdAction","appIdInstall"]
#for index in data[feature] :
userdata = []
for index in range(len(data["uid"])):
    feature_li = []
    for feature in userfeature:
        # a = data[feature]
        # print(a[0],type(a[index]),isinstance(a[0],(numpy.int64)))

        if isinstance(data[feature][index],numpy.int64):
            feature_li.append(int(data[feature][index]))
        elif isinstance(data[feature][index],numpy.float64):
            feature_li.append(0)#缺失值用0填充，这是不合理的，有待改进
        elif isinstance(data[feature][index], numpy.float):
            feature_li.append(0)
        else :
            trans = data[feature][index].strip().split(" ")
            trans = map(int,trans)
            trans = sorted(trans)
            #print(trans)
            s = 0
            for num in trans :
                s += num
            feature_li.append(s)

    print(feature_li)
    userdata.append(feature_li)
userdata = numpy.array(userdata)
print("--------------------------------------------------------------------")
print(userdata)
'''
第一步要把所有特征值id提取，
特征特征值id进行排序，组合成一个特征值，这样就实现了（1）的目标
多特征值id时我们把特征值排序再进行onehot编码，带来的缺点：
例如 特征a有特征id1,2,3 b有特征id1，2他们显然是相近的，我的方法onehot
编码后他们不再相关;我想应该有更好的方法可以解决，比如：word2vec
'''