DDR爱好者之家 Design By 杰米

因为需要一个html形式的数据统计界面,所以做了一个基于pyecharts包的可视化程序,当然matplotlib还是常用的数据可视化包,只不过各有优劣;基本功能概述就是读取csv文件数据,对每列进行数据统计并可视化,最后形成html动态界面,选择pyecharts的最主要原因就是这个动态界面简直非常炫酷。

先上成品图:

利用pyecharts读取csv并进行数据统计可视化的实现

数据读取和数据分析模块:

#导入csv模块
import csv
#导入可视化模块
from matplotlib import pyplot as plt
from pylab import mpl
import numpy as np
import random
from pyecharts import Line,Pie,Grid,Bar,WordCloud
#指定文件名,然后使用 with open() as 打开

python_file = 'haiyang.csv'
#filename = 'release/111.csv'
#python3 LieCharts.py test_chart --python_file 'haiyang.csv'
with open(python_file) as f:
    #创建一个阅读器:将f传给csv.reader
    reader = csv.reader(f)
    #使用csv的next函数,将reader传给next,将返回文件的下一行
    header_row = next(reader)

    for index, column_header in enumerate(header_row):
        print(index, column_header)

    #读取置信度
    #创建置信度的列表
    confidences =[]
    #创建风险等级数组
    highRisk = []
    middleRisk = []
    lowRisk = []
    noRisk = []
    person = []
    #创建时间点
    timePoint = []
    #文件信息
    fileInformation = []


    #遍历reader的余下的所有行(next读取了第一行,reader每次读取后将返回下一行)
    for row in reader:

    # 下面就是对某一列数据进行遍历,因为项目保密,就不列出具体代码了,其实就是各种循环语句,大家根据自己的数据简单写一下就行
            
    fileInformation.append('某某某某')
    fileInformation.append(row[0])
    fileInformation.append(row[1])
    fileInformation.append(row[2])
    fileInformation.append(len(confidences))
    int_confidences = []
    for i in confidences:
  # 同上
    len_noRisk = len(noRisk)
    len_lowRisk = len(lowRisk)
    len_middleRisk = len(middleRisk)
    len_highRisk = len(highRisk)
    len_person = len(person)

    total = int(len_person+len_highRisk+len_middleRisk+len_lowRisk+len_noRisk)
    if (len_highRisk > total/2):
  # 同上

数据可视化模块:

pie_title = Pie('某某某分析报表', "", title_pos='center',title_top="1%",title_text_size=42,subtitle_text_size=20)

value=[10000,6181,4386,4055,4000]
wordcloud=WordCloud(width=30,height=12,title="某某某某信息",title_pos="22%",title_top="12%",title_text_size=32)
wordcloud1=WordCloud(width=30,height=12,title="某某:"+fileInformation[1],title_pos="22%",title_top="22%",title_text_size=26)
wordcloud2=WordCloud(width=30,height=12,title="某某:"+fileInformation[2],title_pos="22%",title_top="30%",title_text_size=26)
#wordcloud3=WordCloud(width=30,height=12,title="音频采样率:"+fileInformation[3],title_pos="22%",title_top="38%",title_text_size=26)
#wordcloud4=WordCloud(width=30,height=12,title="总时长/s:"+fileInformation[4],title_pos="22%",title_top="36%",title_text_size=32)

# wordcloud.add("",fileInformation,value,word_size_range=[20,100],rotate_step=3
#        ,xaxis_pos=200,grid_left="1%",grid_bottom="50%",grid_top="5%",grid_right="80%")
#折线图
line=Line("某某某某某走势图",title_pos='center',title_top="51%",title_text_size=32,width=600,height = 20)
attr=timePoint
line.add("某某某某某",attr,int_confidences,legend_pos="85%",legend_top="54%",
    mark_point=["max","min"],mark_line=["average"])
#饼图
attr=["某某某某", "某某某某", "某某某某", "某某某"]
v1=[len_highRisk, len_middleRisk, len_lowRisk,len_noRisk]
pie=Pie("某某某某某某某",title_pos="65%",title_top="12%",title_text_size=32,width=100,height = 100)
pie.add("",attr,v1,radius=[0,30],center=[71,35],
    legend_pos="85%",legend_top="20%" ,legend_orient="vertical")
grid=Grid(width = 1800 ,height= 900)#调整画布大小

grid.add(line,grid_left="5%",grid_bottom="2%",grid_top="60%")
grid.add(pie_title,grid_bottom="10%")
grid.add(wordcloud,grid_left="1%",grid_bottom="50%",grid_top="5%",grid_right="80%")
grid.add(wordcloud1,grid_left="1%",grid_bottom="50%",grid_top="5%",grid_right="80%")
grid.add(wordcloud2,grid_left="1%",grid_bottom="50%",grid_top="5%",grid_right="80%")
#grid.add(wordcloud3,grid_left="1%",grid_bottom="50%",grid_top="5%",grid_right="80%")
#grid.add(wordcloud4,grid_left="1%",grid_bottom="50%",grid_top="5%",grid_right="80%")
grid.add(pie,grid_left="50%",grid_bottom="50%")


#grid.render()
grid.render(path='./release/XXXX.html')

根据需求这个还可以跨平台跨语言调用,比如C++程序调用python进行数据分析。

DDR爱好者之家 Design By 杰米
广告合作:本站广告合作请联系QQ:858582 申请时备注:广告合作(否则不回)
免责声明:本站资源来自互联网收集,仅供用于学习和交流,请遵循相关法律法规,本站一切资源不代表本站立场,如有侵权、后门、不妥请联系本站删除!
DDR爱好者之家 Design By 杰米