zoukankan      html  css  js  c++  java
  • RDD编程

    一、词频统计:

    1.读文本文件生成RDD lines

    lines = sc.textFile('file:///home/hadoop/word.txt')

    2.将一行一行的文本分割成单词 words flatmap()

    words = lines.flatMap(lambda line:line.split())
    words.collect()

    3.全部转换为小写 lower()

    words = lines.flatMap(lambda line:line.lower().split())
    words.collect()

    4.去掉长度小于3的单词 filter()

    words = words.filter(lambda word:len(word)>3)
    words.collect()

    5.去掉停用词

    with open('/home/hadoop/stopwords.txt') as f:
         stops = f.read().split()
    
    words = words.filter(lambda word:word not in stops)
    words.count()
    words.collect()

    6.转换成键值对 map()

    words = words.map(lambda word:(word,1))

    7.统计词频 reduceByKey()

    words = words.reduceByKey(lambda a,b:a+b)

    8.按字母顺序排序 sortBy(f)

    words = words.sortBy(lambda word:word[0])
    words.collect()

    9.按词频排序 sortByKey()

    words=words.sortByKey()
    words.collect()
    # 或者用sortBy(f)
    words = words.sortBy(lambda word:word[1])
    words.collect()

    10.结果文件保存 saveAsTextFile(out_url)

    words.saveAsTextFile("file:///home/hadoop/out.txt")

     11.词频结果可视化charts.WordCloud()

    #11.词频结果可视化charts.WordCloud()
    from pyecharts.charts import WordCloud
    url='D:/1342-0.txt'
    with open(r'D:/stopwords.txt') as f:
        stops=f.read().split()
    wc=sc.textFile(url).flatMap(lambda line:line.lower().replace(',','').split()).filter(lambda word:word not in stops).filter(lambda word:len(word)>2).map(lambda word:(word,1)).reduceByKey(lambda a,b:a+b).sortBy(lambda x:x[1],False).take(100)
    
    mywordcloud=WordCloud()
    mywordcloud.add("",wc,shape='circle')
    mywordcloud.render()

    二、学生课程分数案例

    lines = sc.textFile('file:///home/hadoop/chapter4-data01.txt')
    lines.take(5)

    1.总共有多少学生?map(), distinct(), count()

    lines.map(lambda line : line.split(',')[0]).distinct().count()

    2.开设了多少门课程?

    lines.map(lambda line : line.split(',')[1]).distinct().count()

    3.每个学生选修了多少门课?map(), countByKey()

    lines.map(lambda line : line.split(',')).map(lambda line:(line[0],(line[1],line[2]))).countByKey()
    # 或者 map(),countByValue()

    4.每门课程有多少个学生选?map(), countByValue()

    lines.map(lambda line : line.split(',')).map(lambda line : (line[1])).countByValue()
    # 或者 map(),countByKey()
    lines.map(lambda line:line.split(',')).map(lambda line:(line[1],(line[0],line[2]))).countByKey()

    5.Les选修了几门课?每门课多少分?filter(), map() RDD

    lines.filter(lambda line:"Les" in line).map(lambda line:line.split(',')).collect()

    6.Les选修了几门课?每门课多少分?map(),lookup()  list

    lines.map(lambda line:line.split(',')).map(lambda line:(line[0],line[1])).lookup("Les")
    lines.map(lambda line:line.split(',')).map(lambda line:(line[0],line[2])).lookup("Les")

    7.Les的成绩按分数大小排序。filter(), map(), sortBy()

    lines.filter(lambda line:"Les" in line).map(lambda line:line.split(',')).sortBy(lambda line:(line[2])).collect()

    8.Les的平均分。map(),lookup(),mean()

    import numpy as np
    meanlist=lines.map(lambda line:line.split(',')).map(lambda line:(line[0],line[2])).lookup("Les")
    np.mean([int(x) for x in meanlist])

     

    9.生成(课程,分数)RDD,观察keys(),values()

    lines = sc.textFile('file:///home/hadoop/chapter4-data01.txt')
    words = lines.map(lambda line:line.split(',')).map(lambda line:(line[1],line[2]))
    words.keys().take(5)
    words.values().take(5)

    10.每个分数+5分。mapValues(func)

    words.mapValues(lambda x:int(x)+5).foreach(print)

    11.求每门课的选修人数及所有人的总分。combineByKey()

    course = words.combineByKey(lambda v:(int(v),1),lambda c,v:(c[0]+int(v),c[1]+1),lambda c1,c2:(c1[0]+c2[0],c1[1]+c2[1]))

    12.求每门课的选修人数及平均分,精确到2位小数。map(),round()

    course.map(lambda x:(x[0],x[1][1],round(x[1][0]/x[1][1],2))).collect()

    13.求每门课的选修人数及平均分。用reduceByKey()实现,并比较与combineByKey()的异同。

    lines.map(lambda line:line.split(',')).map(lambda x:(x[1],(int(x[2]),1))).reduceByKey(lambda a,b:(a[0]+b[0],a[1]+b[1])).foreach(print)

     

     14.结果可视化。charts,Bar()

    from pyecharts.charts import Bar
    from pyecharts import options as opts
    
    bar = Bar()
    bar.add_xaxis(cs.keys().collect())
    bar.add_yaxis('avg',cs.map(lambda x:x[2]).collect())
    bar.set_global_opts()
    bar.render_notebook()

    from pyecharts.charts import Bar
    from pyecharts import options as opts
    from pyecharts.globals import ThemeType
    bar = Bar(init_opts=opts.InitOpts(theme=ThemeType.PURPLE_PASSION))
    bar.add_xaxis(cs.keys().collect())
    bar.add_yaxis('rs',cs.map(lambda x:x[1]).collect())
    bar.add_yaxis('avg',cs.map(lambda x:x[2]).collect())
    
    bar.set_global_opts(title_opts=opts.TitleOpts(title="课程", subtitle="选修人数,平均分"),
                        xaxis_opts=opts.AxisOpts(axislabel_opts=opts.LabelOpts(rotate=-15)),
                        yaxis_opts=opts.AxisOpts(max_=150)) 
    
    bar.render_notebook()

  • 相关阅读:
    Flask学习 1创建第一个页面
    Flask教程
    微信服务号认证和不认证的区别
    python保存selenium的cookies写入和读出
    centos 无界面安装selenium+chrome+chromedirver的设置
    win7系统下安装Splash。
    ubuntu mysql5.6二进制安装
    Python socket实现处理多个连接
    Python 简单soket例子
    Python socket网络编程(通信介绍)
  • 原文地址:https://www.cnblogs.com/shawncs/p/14673296.html
Copyright © 2011-2022 走看看