大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计

环境:centos7+hadoop2.5.2

1.使用ECLIPS具打包运行WORDCOUNT实例,统计莎士比亚文集各单词计数(文件SHAKESPEARE.TXT)。

①WorldCount.java 中的main函数修改如下:

public static void main(String[] args) throws Exception {
Configuration conf = new Configuration();
Job job = new Job(conf, "word count");
job.setJarByClass(WordCount.class);
job.setMapperClass(TokenizerMapper.class);
job.setCombinerClass(IntSumReducer.class);
job.setReducerClass(IntSumReducer.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(IntWritable.class);
//设置输入文本路径
FileInputFormat.addInputPath(job, new Path("/input"));
//设置mp结果输出路径
FileOutputFormat.setOutputPath(job, new Path("/output/wordcount"));    System.exit(job.waitForCompletion(true) ? 0 : 1);
}


②导出WordCount的jar包:
  export->jar file->next->next->Main class里面选择WordCount->Finish。
③使用scp将wc.jar拷贝到node1机器,创建目录:hadoop fs –mkdir /input,将shakespeare.txt上传到hdfs上,运行wc.jar文件:hadoop jar wc.jar
④使用hadoop fs -cat /output/wordcount/part-r-00000 grep|head -n 30 查看前30条输出结果:大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计_第1张图片

2.对于SOGOU_500W_UTF文件,完成下列程序设计内容:

(1) 统计每个用户搜索的关键字总长度

Mapreduce程序:

public class sougou3 {
public static class Sougou3Map extends
Mapper {
public void map(Object key, Text value, Context context)
throws IOException, InterruptedException {
  String line = value.toString();
  String[] vals = line.split("\t");
  String uid = vals[1];
  String search = vals[2];
  context.write(new Text(uid), new Text(search+"|"+search.length()));
}
}
public static class Sougou3Reduce extends
Reducer {
public void reduce(Text key, Iterable values,
Context context) throws IOException, InterruptedException {
  String result = "";
  for (Text value : values) {
    String strVal = value.toString();
    result += (strVal+" ");
  }
  context.write(new Text(key + "\t"), new Text(result));
  }
  }
}

输出结果:

大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计_第2张图片(2) 统计2011年12月30日1点到2点之间,搜索过的UID有哪些?

Mapreduce程序:

public class sougou1 {

    public static class Sougou1Map extends
            Mapper {

        public void map(Object key, Text value, Context context)
                throws IOException, InterruptedException {
            String line = value.toString();
            String[] vals = line.split("\t");
            String time = vals[0];
            String uid = vals[1];
            //2008-07-10 19:20:00
            String formatTime = time.substring(0,4)+"-"+time.substring(4,6)+"-"+time.substring(6,8)+" "
                    +time.substring(8,10)+":"+time.substring(10,12)+":"+time.substring(12,14);
            SimpleDateFormat sdf = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");
            Date date;
            try {
                date = sdf.parse(formatTime);
                Date date1 = sdf.parse("2011-12-30 01:00:00");
                Date date2 = sdf.parse("2011-12-30 02:00:00");
                //日期在范围区间上
                if (date.getTime() > date1.getTime() && date.getTime() < date2.getTime()){
                    context.write(new Text(uid), new Text(formatTime));
                }
            } catch (ParseException e) {
                e.printStackTrace();
            }
        }
    }
    public static class Sougou1Reduce extends
            Reducer {
        public void reduce(Text key, Iterable values,
                Context context) throws IOException, InterruptedException {
                String result = "";
                for (Text value : values) {
                    result += value.toString()+"|";
                }
                context.write(key, new Text(result));
        }
    }
}

 

输出结果:
左边是用户id,右边分别是时间,以“|”作为分割。

大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计_第3张图片

 

(3) 统计搜索过‘仙剑奇侠’的每个UID搜索该关键词的次数。

Mapreduce程序:

public class sougou2 {
    public static class Sougou2Map extends
            Mapper {
        public void map(Object key, Text value, Context context)
                throws IOException, InterruptedException {
            String line = value.toString();
            String[] vals = line.split("\t");
            String uid = vals[1];
            String search = vals[2];
            if (search.equals("仙剑奇侠")){
                context.write(new Text(uid), new IntWritable(1));
            }
        }
    }
    public static class Sougou2Reduce extends
            Reducer {
        public void reduce(Text key, Iterable values,
                Context context) throws IOException, InterruptedException {
                int result = 0;
                for (IntWritable value : values) {
                    result += value.get();
                }
                context.write(new Text(key+"\t"), new IntWritable(result));
        }
    }
}

 


输出结果:
UID为:6856e6e003a05cc912bfe13ebcea8a04的用户搜索过“仙剑奇侠”共1次。

3.使用MAPREDUCE程序设计实现对文件中下列数据的排序操作78 11 56 87 25 63 19 22 55

Mapreduce程序:

public class Sort {
    //map将输入中的value化成IntWritable类型,作为输出的key    
    public static class Map extends Mapper{
        private static IntWritable data=new IntWritable();
        //实现map函数
        public void map(Object key,Text value,Context context)
                throws IOException,InterruptedException{
            String line=value.toString();
            data.set(Integer.parseInt(line));
            context.write(data, NullWritable.get());
        }
    }
   
    //reduce将输入中的key复制到输出数据的key上,
    //然后根据输入的value-list中元素的个数决定key的输出次数
    //用全局linenum来代表key的位次
    public static class Reduce extends
            Reducer{
       
       
        //实现reduce函数
        public void reduce(IntWritable key,Iterable values,Context context)
                throws IOException,InterruptedException{
            for(NullWritable val:values){
                context.write(key, NullWritable.get());
            }
        }
 
    }

}

 

输出内容为:
大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计_第4张图片

4.学生成绩文件TXT内容(字段用TAB键分隔)如下,使用MAPREDUCE计算每个学生的平均成绩

李平 87 89 98 75
张三 66 78 69 70
李四 96 82 78 90
王五 82 77 74 86
赵六 88 72 81 76

Mapreduce 程序:

public class Score {

    public static class ScoreMap extends
            Mapper {

        public void map(Object key, Text value, Context context)
                throws IOException, InterruptedException {
            context.write(value, NullWritable.get());
        }

    }

    public static class ScoreReduce extends
            Reducer {
        public void reduce(Text key, Iterable values,
                Context context) throws IOException, InterruptedException {
            for (NullWritable nullWritable : values) {
                String line = key.toString();
                String[] vals = line.split("\t");
                String name = vals[0];
                int val1 = Integer.parseInt(vals[1]);
                int val2 = Integer.parseInt(vals[2]);
                int val3 = Integer.parseInt(vals[3]);
                int average = (val1 + val2 + val3) / 3;
                context.write(new Text(name), new IntWritable(average));
            }
        }
    }
}

 

输出结果为

 大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计_第5张图片

相关资料:

链接:http://pan.baidu.com/s/1dFD7mdr 密码:xwu8

转载于:https://www.cnblogs.com/tongkey/p/7854266.html

你可能感兴趣的:(大数据(1):基于sogou.500w.utf8数据的MapReduce程序设计)