mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sebastian Schelter <...@apache.org>
Subject Re: Documentation for ParallelALSFactorizationJob
Date Tue, 13 Nov 2012 13:56:03 GMT
Hi Kris,

ALS factorizes your input matrix A (users x items) into two smaller,
dense matrices: U (users x features) and M (features x items).

Best,
Sebastian



On 13.11.2012 13:12, Kris Jack wrote:
> What are the equations to estimate how much RAM is required for the mappers
> and reducers in the ParallelALSFactorizationJob steps?
> 
> I don't think that I'm alone in asking having this kind of question.  It
> would probably be useful to include such estimates in the documents and in
> the code too as standard for jobs.
> 
> Best,
> Kris
> 
> 
> 
> 
> 
> 2012/11/12 Sebastian Schelter <ssc.open@googlemail.com>
> 
>> Hi Kris,
>>
>> There is no such code. You can partially built that from what we have.
>> You can use ItemSimilarityJob to compute item similarities, load the
>> results into the non-distributed recommenders using a FileDataModel and
>> do the evaluation there.
>>
>> Best,
>> Sebastian
>>
>> On 12.11.2012 16:42, Kris Jack wrote:
>>> Thanks guys, I've got that working now.  I was interested to find that
>>> there is code that helps to evaluate the results of AWS.  Out of
>> interest,
>>> is there similar code in Mahout that helps with evaluating matrix
>>> multiplication (e.g.
>> org.apache.mahout.cf.taste.hadoop.item.RecommenderJob)?
>>>
>>> Best,
>>> Kris
>>>
>>>
>>>
>>>
>>>
>>> 2012/10/18 Sebastian Schelter <ssc@apache.org>
>>>
>>>>> I don't know if there is code,
>>>>> probably not, but conceptually that is all that it involves.
>>>>
>>>> Once you factorized your interaction matrix, you can use
>>>>
>>>> org.apache.mahout.cf.taste.hadoop.als.RecommenderJob
>>>>
>>>> to compute recommendations in parallel.
>>>>
>>>> Best,
>>>> Sebastian
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>
>>
> 
> 


Mime
View raw message