hadoop-mapreduce-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Shahab Yunus <shahab.yu...@gmail.com>
Subject Re: cleanup() in hadoop results in aggregation of whole file/not
Date Sat, 28 Feb 2015 13:10:19 GMT
As far as I understand cleanup is called per task. In your case I.e.
per map task. To get an overall count or measure, you need to aggregate
it yourself after the job is done.

One way to do that is to use counters and then merge them programmatically
at the end of the job.

Regards,
Shahab

On Saturday, February 28, 2015, unmesha sreeveni <unmeshabiju@gmail.com>
wrote:

>
> ‚ÄčI am having an input file, which contains last column as class label
> 7.4 0.29 0.5 1.8 0.042 35 127 0.9937 3.45 0.5 10.2 7 1
> 10 0.41 0.45 6.2 0.071 6 14 0.99702 3.21 0.49 11.8 7 -1
> 7.8 0.26 0.27 1.9 0.051 52 195 0.9928 3.23 0.5 10.9 6 1
> 6.9 0.32 0.3 1.8 0.036 28 117 0.99269 3.24 0.48 11 6 1
> ...................
> I am trying to get the unique class label of the whole file. Inorder to
> get the same I am doing the below code.
>
> *public class MyMapper extends Mapper<LongWritable, Text, IntWritable,
> FourvalueWritable>{*
> *    Set<String> uniqueLabel = new HashSet();*
>
> *    public void map(LongWritable key,Text value,Context context){*
> *        //Last column of input is classlabel.*
> *         Vector<String> cls = CustomParam.customLabel(line, delimiter,
> classindex); // *
> *         uniqueLabel.add(cls.get(0));*
> *    }*
> *    public void cleanup(Context context) throws IOException{*
> *        //find min and max label*
> *
>  context.getCounter(UpdateCost.MINLABEL).setValue(Long.valueOf(minLabel));*
> *
>  context.getCounter(UpdateCost.MAXLABEL).setValue(Long.valueOf(maxLabel));*
> *}*
> Cleanup is only executed for once.
>
> And after each map whether "Set uniqueLabel = new HashSet();" the set get
> updated,Hope that set get updated for each map?
> Hope I am able to get the uniqueLabel of the whole file in cleanup
> Please suggest if I am wrong.
>
> Thanks in advance.
>
>
>

Mime
View raw message