mahout-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Lance Norskog (JIRA)" <>
Subject [jira] Commented: (MAHOUT-586) Redo RecommenderEvaluator for modularity
Date Tue, 15 Mar 2011 08:52:29 GMT


Lance Norskog commented on MAHOUT-586:

Patch walkthrough:

There are now old and new apis on RecommenderEvaluator.
The one ones are as before. The new ones match the modular toolkit.
The evaluator can be run on a datamodel v.s. a recommender or even between datamodels.
I have datamodels which by nature supply recommendations; this structure makes them easy to
work with.

Patch Structure:
Abstract...Evaluator is as before.
It has two subclasses to supply statistics tracking.
Its API is tagged obsolete.

The newer APIs support a more modular approach where the tool can implement several strategies
and the user can supply your choice of statistical tracking.

The SamplingDataModel is a component that lets this new code implement separating data for
training v.s. test.This implementation is sparse but not very scalable.

> Redo RecommenderEvaluator for modularity
> ----------------------------------------
>                 Key: MAHOUT-586
>                 URL:
>             Project: Mahout
>          Issue Type: Improvement
>          Components: Collaborative Filtering
>            Reporter: Lance Norskog
>            Assignee: Sean Owen
>             Fix For: 0.5
>         Attachments: MAHOUT-559.patch, MAHOUT-586.patch, MAHOUT-586.patch
> The RecommenderEvaluator implementation is hard-coded around one algorithm.
> This is a more flexible, modular rewrite.

This message is automatically generated by JIRA.
For more information on JIRA, see:

View raw message