[
https://issues.apache.org/jira/browse/MAHOUT-1421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13909209#comment-13909209
]
jay vyas commented on MAHOUT-1421:
----------------------------------
>From mailinglist "Great idea. Hard to do well. Would it be possible for you
>to try to build a picture of all the pieces
that need to be connected before you start building connectors and
converters? " -- ted dunning.
Sure ted. Will see if i can visualize all the different input types.
> Adapter package for all mahout tools
> ------------------------------------
>
> Key: MAHOUT-1421
> URL: https://issues.apache.org/jira/browse/MAHOUT-1421
> Project: Mahout
> Issue Type: Improvement
> Reporter: jay vyas
>
> Hi mahout. I'd like to create an umbrella JIRA for allowing more runtime
> flexibility for reading different types of input formats for all mahout
> tasks.
> Specifically, I'd like to start with the FreeTextRecommenderAdapeter, which
> typically requires:
> 1) Hashing text entries into numbers
> 2) Saving the large transformed file on disk
> 3) Feeding it into classifieer
> Instead, we could build adapters into the classifier itself, so that the user
> 1) Specifies input file to recommender
> 2) Specifies transformation class which converts each record of input to 3
> column recommender format
> 3) Runs internal mahout recommender directly against the data
> And thus the user could easily run mahout against existing data without
> having to munge it to much.
> This package might be called something like "org.apache.mahout.adapters", and
> would over time provide flexible adapters to the core mahout algorithm
> implementations, so that folks wouldnt have to worry so much about
> vectors/csv transformers/etc...
> Any thoughts on this? If positive feedback I can submit an initial patch to
> get things started.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
