You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Robert Hall <go...@gmail.com> on 2012/07/12 21:30:00 UTC
From example to job
Greetings.
I'm trying to jump from the examples in mahout to a practical job of my
very own. First, I'm very new to mahout but I do have some experience with
machine learning, clustering, and classifications.
My goal: To get KMeans clusters of time-based use from structured data
Example Input:
John Doe,1324,1233,2234,1267,1456,1745,1212
There's a name and a variable series of numbers that correspond to time in
seconds to complete an operation. The times are pre-filtered > 1200 and
built by date/time (pivoted into nameless columns) of the operation, but
the date/time is not relevant to my goal.
Can someone point me toward any resources that explain, not how to run an
example, but how the examples were put together?
If not a resource, how about a high-level description on what mahout is
looking for and how it does, say a KMeans cluster analysis.
Finally, can someone describe a mahout vector and vector file? A
description plus the actual format of a vector row/file.
--
Robert Hall
golsombe@gmail.com
Re: From example to job
Posted by Lance Norskog <go...@gmail.com>.
A Mahout vector is not one format- it is a family of data structures
optimized for various tasks.
A Mahout vector file is a Hadoop sequencefile of 0 or more entries of
Writable duples (Writable, VectorWritable).
Several programs require the first writable to be an int row number.
There may be single-process programs which require the row numbers in
sequence. There should not be any Hadoop-friendly programs which
require this.
The "rowid" you will notice referred to a lot is a pair of programs
that replace the Writable with a unique integer, and save the Writable
out to a dictionary of int->writable sequencefile.
On Thu, Jul 12, 2012 at 12:30 PM, Robert Hall <go...@gmail.com> wrote:
> Greetings.
>
> I'm trying to jump from the examples in mahout to a practical job of my
> very own. First, I'm very new to mahout but I do have some experience with
> machine learning, clustering, and classifications.
>
> My goal: To get KMeans clusters of time-based use from structured data
>
> Example Input:
> John Doe,1324,1233,2234,1267,1456,1745,1212
>
> There's a name and a variable series of numbers that correspond to time in
> seconds to complete an operation. The times are pre-filtered > 1200 and
> built by date/time (pivoted into nameless columns) of the operation, but
> the date/time is not relevant to my goal.
>
> Can someone point me toward any resources that explain, not how to run an
> example, but how the examples were put together?
>
> If not a resource, how about a high-level description on what mahout is
> looking for and how it does, say a KMeans cluster analysis.
>
> Finally, can someone describe a mahout vector and vector file? A
> description plus the actual format of a vector row/file.
>
>
>
> --
> Robert Hall
> golsombe@gmail.com
--
Lance Norskog
goksron@gmail.com