[jira] [Updated] (MAHOUT-1083) CIReducer in kmeans doesn't work well

["Jeff Eastman (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/MAHOUT-1083?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jeff Eastman updated MAHOUT-1083:
---------------------------------

    Attachment: MAHOUT-1083.patch

Oh Yikes! This is another example of the Hadoop platform reusing a single instance of the Writable every time another one is read from the input stream. Every time iter.next() is called, the same object (cw == first) will be returned. This does not show up in unit tests because only a single reducer is ever used. It explains exactly the odd clustering behavior you have observed.

Here's a patch that *may* fix the problem. I believe that ClusterWritable will create a new instance for its value instVar even if the same ClusterWritable is used.

Please try this and see if it resolves the issue.
                
> CIReducer in kmeans doesn't work well
> -------------------------------------
>
>                 Key: MAHOUT-1083
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-1083
>             Project: Mahout
>          Issue Type: Bug
>         Environment: hadoop-2.0.0-alpha: pseudo cluster and single node cluster    hadoop-1.0.3: pseudo cluster    hadoop-0.20.2:pseudo cluster mahout:    mahout-0.7 os:    ubuntu 11.04 jdk:    jdk1.6.0_27
>            Reporter: liutengfei
>         Attachments: MAHOUT-1083.patch
>
>
> the function "reduce" in mahout-0.7-kmeans-CIReducer.java doesn't work well as it looks like.
>   protected void reduce(IntWritable key, Iterable<ClusterWritable> values, Context context) throws IOException,
>       InterruptedException {
>     Iterator<ClusterWritable> iter = values.iterator();
>     ClusterWritable first = null;
>     while (iter.hasNext()) {
>       ClusterWritable cw = iter.next();
>       if (first == null) {
>         first = cw;
>       } else {
>         first.getValue().observe(cw.getValue());
>       }
>     }
>     List<Cluster> models = new ArrayList<Cluster>();
>     models.add(first.getValue());
>     classifier = new ClusterClassifier(models, policy);
>     classifier.close();
>     context.write(key, first);
>   }
> Apparently， the variable "first" will collect all output data of maps. Actually but, the value of "first" will change after the code "ClusterWritable cw = iter.next();", same with this new variable "cw"! I don't why but running result shows that the code runs looks like this:"ClusterWritable cw = first = iter.next();".
> is "cw" a reference a to "iter"?
> is "iter.next" just change the value of "iter" itself to the next?

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira