On Jul 28, 2011, at 6:47 AM, Dave Gettier wrote:

> Can you elaborate on that? Not sure what "--method mapreduce" means.  
> 
> Some details:
> 
> - I have a CSV file. 
> - I programmatically create  namedVectors (a java program).
> - I run the kmeansdriver which (I thought) submits a number of mapreduce 
> jobs. 

Are you running KMeansDriver on trunk?  I'm pretty sure it has two options for 
method:  --method sequential or mapreduce

Not sure if that is the same driver as used in Mahout in Action.

Are you actually running on a Hadoop cluster or just locally?

> - The output is another sequence file, which I then turn back into CSV.
> - All sequence files are stored locally. 
> 
> 
> Side note: 
>       I am overriding the distance method. 
>       I can which distance method via a config file. 
>       I can run Euclidian, or my distance function.
> 
> -----Original Message-----
> From: Grant Ingersoll [mailto:[email protected]] 
> Sent: Thursday, July 28, 2011 9:23 AM
> To: [email protected]
> Subject: Re: Kmeans runs successfully, but no map/reduce jobs
> 
> Do you need --method mapreduce passed in?
> 
> On Jul 27, 2011, at 4:20 PM, Dave Gettier wrote:
> 
>> 
>> I am running a kmeans application which was adapted from  example 7.2 of 
>> Mahout in Action.  The java program runs successfully,  giving me the 
>> expected results;  however, there are no map/reduce jobs being kicked off.   
>>  My understanding was that KMeansCluster runs locally, but KMeansDriver run 
>> on the cluster.  How does one point the job to run on the cluster? Or am I 
>> missing something?
>> 
>> KMeansDriver.run(conf,
>>              new Path(cp.getsDataDir() +  "/points"),
>>              new Path(cp.getsDataDir()  + "/clusters"),
>>              new Path(cp.getsDataDir()  +  "/outputs"),
>>              new EuclideanDistanceMeasure(),
>>              .001,10, true, true);
>> 
>> Thanks in advance!
>> 
>> -DG
>> 
> 
> --------------------------------------------
> Grant Ingersoll
> 
> 

--------------------------------------------
Grant Ingersoll


Reply via email to