Hi Yimin, Thanks for pointing this out. I've just committed a patch to correct it.
On 6/1/12 3:13 PM, Yimin Wu wrote:
Unlike KMeansUtil, the FuzzyKMeansUtil does not take a Configuration as a parameter. While performing deserialization, it create a new Configuration. This is a bug since it wouldn't work on Hadoop files. It will calles FuzzyMapper setup throw an exception complainning the file can not be found. I verified that the issue would gone after passing in conf like KMeansUtil. Bascially, this code patern "new Configuration()" could be problematic as long as the code would be used in a mapreduce job. I suggest developer team review all the places where a Configuration object is used. Whenever a new Configuration object is created instead of passed in. Thanks,Yimin Wu Predixion Software
