码迷,mamicode.com
首页 > 其他好文 > 详细

利用朴素贝叶斯对名字进行性别预测

时间:2019-01-30 15:51:13      阅读:146      评论:0      收藏:0      [点我收藏+]

标签:use   lap   ora   利用   imp   turn   loaddata   count   完整   

完整代码

#-*-coding:utf-8-*-
import pandas as pd
import math
from collections import defaultdict

# load the data and preprocess the data

train = pd.read_csv("./data/train.txt")
test = pd.read_csv("./data/test.txt")
def loadData():
    # divide the data into two parts female and male
    names_male = train[train[‘gender‘] == 0]
    names_female = train[train[‘gender‘] == 1]

    totals = {
        ‘f‘:len(names_female),
        ‘m‘:len(names_male),
    }

    # use total to storage the oss
    return names_male,names_female,totals

# cal the posibilitied of the word in the name 

def calFreq(names_male,names_female,totals):
    # the word appereanced in female‘s name
    freq_list_f = defaultdict(int)
    for name in names_female :
        for char in name:
            freq_list_f[char] += 1.0 / totals[‘f‘]

        # the word appereanced in female‘s name
    freq_list_m = defaultdict(int)
    for name in names_male :
        for char in name:
            freq_list_f[char] += 1.0 / totals[‘m‘]

    return freq_list_m, freq_list_f     

# to avoid some word not disapperenced in the train data
def LaplaceSmooth(char, freq_list,total,alpha=1.0):
    count = freq_list[char * total]
    distinct_chars = len(freq_list)
    freq_smooth = (count+alpha)/(total+ distinct_chars * alpha)
    return freq_smooth

## ??

def GetLogProb(char, frequency_list, total):
    freq_smooth = LaplaceSmooth(char, frequency_list, total)
    return math.log(freq_smooth) - math.log(1 - freq_smooth)

def getBase(freq_list_m,freq_list_f,train):
    base_f = math.log(1 - train[‘gender‘].mean())
    base_f += sum([math.log(1 - freq_list_f[char]) for char in freq_list_f])
    base_m = math.log(train[‘gender‘].mean())
    base_m += sum([math.log(1 - freq_list_m[char]) for char in freq_list_m])
    bases = {‘f‘: base_f, ‘m‘: base_m}
    return bases

def calLogProb(name, bases,totals, freq_list_m,freq_list_f):
    logprob_m = bases[‘m‘]
    logprob_f = bases[‘f‘]
    for char in name:
        logprob_m += GetLogProb(char,freq_list_m,totals[‘m‘])
        logprob_f += GetLogProb(char,freq_list_f,totals[‘f‘])
    return {‘male‘:logprob_m,‘female‘:logprob_f}

def getGender(logProbs):
    return logProbs[‘male‘] > logProbs[‘female‘]

def getResult(bases, totals, freq_list_m, freq_list_f):
    result = []
    for name in test[‘name‘]:
        LogProbs = calLogProb(name, bases, totals, freq_list_m, freq_list_f)
        gender = getGender(LogProbs)
        result.append(int(gender))
    test[‘pred‘] = result
    print(test.head(20))
    return result
def main():
    names_male,names_female,totals = loadData()
    freq_list_m, freq_list_f = calFreq(names_male,names_female,totals)
    base = getBase(freq_list_m,freq_list_f,train)
    result = getResult(base, totals, freq_list_m, freq_list_f)


main()


利用朴素贝叶斯对名字进行性别预测

标签:use   lap   ora   利用   imp   turn   loaddata   count   完整   

原文地址:https://www.cnblogs.com/Mr0wang/p/10337916.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!