[
https://issues.apache.org/jira/browse/MAHOUT-808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13104374#comment-13104374
]
Phil commented on MAHOUT-808:
-----------------------------
Agree with U!
And turn off the combiner does work! That's the easiest way, even though this
may hurt the efficiency, but not serious.
> logical error with term counting in
> org.apache.mahout.vectorizer.DictionaryVectorizer
> -------------------------------------------------------------------------------------
>
> Key: MAHOUT-808
> URL: https://issues.apache.org/jira/browse/MAHOUT-808
> Project: Mahout
> Issue Type: Bug
> Affects Versions: 0.5
> Reporter: Phil
> Priority: Critical
> Labels: DictionaryVectorizer, TermCountReducer
> Fix For: 0.6
>
> Original Estimate: 0.25h
> Remaining Estimate: 0.25h
>
> when using mahout lda for topic modeling, creating vectors from SequenceFile
> is essential, (refer to
> https://cwiki.apache.org/confluence/display/MAHOUT/Creating+Vectors+from+Text)
> but when the --minSupport was set a little bit larger, I found the term
> counting not right --- there is a logical error at
> org.apache.mahout.vectorizer.DictionaryVectorizer.java:line 335
> job.setCombinerClass(TermCountReducer.class);
> Now turn to line 41 at org.apache.mahout.vectorizer.term.TermCountReducer.java
> if (sum >= minSupport) {
> context.write(key, new LongWritable(sum));
> }
> so some terms would be filtered at Combiner even though they actually could
> pass through, absolutely this is not what we've expected.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
