基于上一篇文章，python实现TF-IDF算法//www.greatytc.com/p/d9c303c29a67
- 将POI的类别抽象为文档中的词
- 划分的区域 抽象为 文档
则可利用TFIDF的值 计算 与 人口密度的协方差 ，计算出二者的相关性，相关系数越大，则该类别对人口的影响越大 ，后续则可利用得出的结论 研究分析更多的问题。
下面将给出用python实现TFIDF与人口密度的相关系数
image.png
TLij为相应的TFIDF值
PDj为第j区域的人口密度值
用python实现该公式：
def relative(tf_idf,idf,area):
#N是第i类所占的分区数
rou = {}
for m in idf:
#tl_pd 是（第i类所在分区的tf_idf值）与（第i类所在分区的栅格值）的乘积 后的求和
#sum_2 是 （第i类所在分区的tf_idf值）的平方的求和
#sum 是 第 （第i类所在分区的tf_idf值）的求和
#PD 是第i类所在的分区的栅格值的求和
#PD_2 是第i类所在的分区的栅格值的平方的求和
N = 0
tl_pd = 0
sum_2 = 0
sum = 0
PD = 0
PD_2 = 0
for l in tf_idf:
if m in tf_idf[l].keys():
N += 1
for l in tf_idf:
if m in tf_idf[l].keys():
sum_2+=tf_idf[l][m]*tf_idf[l][m]
tl_pd+=tf_idf[l][m]*area[l]
sum+=tf_idf[l][m]
for l in tf_idf:
if m in tf_idf[l].keys():
PD += area[l]
PD_2 += area[l]*area[l]
fenzi = N*tl_pd-sum*PD
fenmu = (N*sum_2-sum*sum)**0.5- (N*PD_2-PD*PD)**0.5
poi_md = fenzi/fenmu
rou[m] = poi_md
return rou
完整的代码如下：
输入数据为：
data1.json
{
"id"：POI的编号
"type":POI的类别
"number":POI所属区域的区域编号
}
area.json
{
"区域编号"：区域人口值
}
输出为：
[
["h", 232.7468753744591],
["f",511.4907105256037],
["e",614.9901766893532],
["g",697.3614562757045],
["d",812.2365216229458],
["b",878.0307964717691],
["a",1137.981560137959],
["c",1448.4753152430358]
]
a b c d e f g h 分别代表不同类别的POI
import csv
import json
import math
filename = "data1.json"
# 处理基础数据 转换成字典
# key为分区代码 相应的value为嵌套字典
# 键为某一分区内 POI的类型代码 相应的值为该类POI出现的次数
def data_parse(filename):
number_counts = {}
with open(filename,'r',encoding='utf-8') as f:
data = json.load(f)
for l in data:
if l["number"] not in number_counts.keys():
number_counts[l["number"]] = {}
if l["type"] not in number_counts[l["number"]].keys():
number_counts[l["number"]][l["type"]] = 1
else:
if l["type"] not in number_counts[l["number"]].keys():
number_counts[l["number"]][l["type"]] = 1
else:
number_counts[l["number"]][l["type"]] += 1
return number_counts
#计算TF值（TF=当前分区含有的I类POI数目/当前分区含有的POI数目）
def tf(number_counts):
for m in number_counts:
fenmu = 0
for j in number_counts[m]:
fenmu += number_counts[m][j]
for k in number_counts[m]:
number_counts[m][k] = number_counts[m][k]/fenmu
return number_counts
#计算IDF的值
def idf(number_counts):
idf = {"a":0,"b":0,"c":0,"d":0,"e":0,"f":0,"g":0,"h":0}
for l in idf:
count = 0
D = 0
for m in number_counts:
D += 1
if l in number_counts[m].keys():
count+=1
idf[l] = math.log(D/count)
return idf
#计算TF-IDF的值
def TFIDF(tf,idf):
for m in tf:
for k in idf:
if k in tf[m].keys():
tf[m][k] = tf[m][k]*idf[k]
# for j in tf[m]:
# print(m,k)
# tf[m][j] = idf[k]*tf[m][j]
# print(tf[m][j])
return tf
#将最终的TF值写入成json文件
def writetojson(number_counts,filename):
with open(filename,'a') as json_file:
json.dump(number_counts, json_file, indent=2,ensure_ascii=False)
def read_area(filename):
with open(filename,'r',encoding='utf-8') as f:
area = json.load(f)
return area
def relative(tf_idf,idf,area):
#N是第i类所占的分区数
#tl_pd 是（第i类所在分区的tf_idf值）与（第i类所在分区的栅格值）的乘积 后的求和
#sum_2 是 （第i类所在分区的tf_idf值）的平方的求和
#sum 是 第 （第i类所在分区的tf_idf值）的求和
#PD 是第i类所在的分区的栅格值的求和
#PD_2 是第i类所在的分区的栅格值的平方的求和
rou = {}
for m in idf:
N = 0
tl_pd = 0
sum_2 = 0
sum = 0
PD = 0
PD_2 = 0
for l in tf_idf:
if m in tf_idf[l].keys():
N += 1
if m in tf_idf[l].keys():
sum_2+=tf_idf[l][m]*tf_idf[l][m]
tl_pd+=tf_idf[l][m]*area[l]
sum+=tf_idf[l][m]
for l in tf_idf:
if m in tf_idf[l].keys():
PD += area[l]
PD_2 += area[l]*area[l]
fenzi = N*tl_pd-sum*PD
fenmu = (N*sum_2-sum*sum)**0.5- (N*PD_2-PD*PD)**0.5
poi_md = fenzi/fenmu
rou[m] = poi_md
return rou
if __name__ == '__main__':
filename = "data1.json"
number_counts = data_parse(filename)
tf = tf(number_counts)
idf = idf(number_counts)
tf_idf = TFIDF(tf,idf)
print(tf_idf)
area = read_area('area.json',)
rou = relative(tf_idf,idf,area)
rou=sorted(rou.items(),key=lambda x:x[1],reverse=False)
writetojson(rou,"rou.json")
