[MAHOUT-419] Convert decomposer code to Hadoop 0.20 API

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 0.3, 0.4
• Component/s: classic
• Labels: deprecated hadoop job
• Assignee: Jake Mannix
• Reporter: Danny Leshem
• Created: 20/Jun/10 14:38
• Updated: 31/Jan/24 22:16
• Resolved: 22/Sep/10 07:09

Description

org.apache.mahout.math.hadoop classes (MatrixMultiplicationJob, TimesSquaredJob, TransposeJob) all use the deprecated Hadoop API. In the spirit of MAHOUT-167 and MAHOUT-143, I suggest converting them to Hadoop's 0.20 API.

The reason I'm raising this now is that this code no longer runs on my Hadoop 0.22-SNAPSHOT cluster (not sure why really - it was running fine about a month ago, but after updating to the latest Mahout trunk a few days ago the code throws "java.lang.RuntimeException: Error in configuring object" at MapTask.runOldMapper).

Also, the documentation at https://cwiki.apache.org/MAHOUT/dimensionalreduction.html is no longer accurate - the command line parameters have changed (even without the new arguments from MAHOUT-308). This is partly due to using the new argument parser which receives the input/output directories differently.