mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dmitriy Lyubimov <dlie...@gmail.com>
Subject Re: SVD, how are the missing values treated?
Date Wed, 11 Sep 2013 00:58:56 GMT
On Tue, Sep 10, 2013 at 5:48 PM, Yang <teddyyyy123@gmail.com> wrote:
> in the simple equation describing SVD:
>
> A = USV
>
> I guess the original matrix A has to have every value filled, so that
> mathematics will be able to carry out the calculation, right?

No. A may be sparse, where 0 elements are omitted. But, logically,
every element has a value, yes.

>
> but the mahout package described here:
> https://cwiki.apache.org/confluence/display/MAHOUT/Dimensional+Reduction
>
> https://cwiki.apache.org/confluence/display/MAHOUT/SVD+-+Singular+Value+Decomposition
>
>
> allows for input to be sparse, so most elements of A are missing values. so
> I wonder how mahout takes care of the missing values? this paper:
> http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA439541
> fills missing values with some sort of averages, which sounds rather
> arbitrary.

in strict SVD, "missing" values are always 0s. I put "missing" into
quotes because they are not really missing, this is just a way
compress input.

>
>
> thanks
> Yang

Mime
View raw message