[MAHOUT-920] Remove a mapreduce job from parallel FPGrowth workflow

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 0.6
• Fix Version/s: 0.6
• Assignee: Robin Anil
• Reporter: Tom Pierce
• Created: 10/Dec/11 20:55
• Updated: 31/Mar/15 22:49
• Resolved: 15/Jan/12 16:19

Description

The transaction sorting job could have been made map-only, and another mapreduce job follows, so it made sense to combine them. It would have been possible to use a ChainMapper approach, but it seemed simpler to fold the functionality into the following mapper (ParallelFPGrowthMapper). The attached patch does just that.

Issue Links

• is related to: MAHOUT-890 Performance issue in FPGrowth
• relates to: MAHOUT-921 FPG uses a lot of boxed primitives - this patch eliminates a bunch of List<Integer>
• relates to: MAHOUT-927 FPG saves a mapping from from feature to mining group, when this can be calculated on the fly