[MAHOUT-1263] Serialise/Deserialise Lambda value for OnlineLogisticRegression

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.7
• Fix Version/s: 0.8
• Component/s: Classification
• Labels: patch
• Assignee: Suneel Marthi
• Reporter: Mike Davy
• Created: 14/Jun/13 13:18
• Updated: 27/Jul/13 04:23
• Resolved: 15/Jun/13 04:40

Description

The value for Lambda in OnlineLogisticRegression seems not the be serialised/deserialised correctly.

If I train a model with a specific lambda value, serialise it, then read it back in, the value of Lambda goes back to the default value (1.0e-5).

I've created a patch that adds the lambda value into the write/readFields (org.apache.hadoop.io.Writable). Patch includes a unit test that checks the values after serialising/deserialising to/from a ByteArray.

I think this is correct, unless I'm missing something obvious?

Note: this patch is not backwards compatible - can easily adapt to be more backwards compatible if required.