org.apache.mahout.math.DiagonalMatrix

It even supports sparse values in the diagonal!

On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
<[EMAIL PROTECTED]> wrote:
>
>    [ https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670 ]
>
> Tamas Jambor commented on MAHOUT-737:
> -------------------------------------
>
> I think so. When will you be releasing 0.6? I am planning to do a bit more extensive testing in the next 2-3 weeks.
>
>> Implicit Alternating Least Squares SVD
>> --------------------------------------
>>
>>                 Key: MAHOUT-737
>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
>>             Project: Mahout
>>          Issue Type: New Feature
>>          Components: Collaborative Filtering
>>    Affects Versions: 0.6
>>            Reporter: Tamas Jambor
>>            Assignee: Sebastian Schelter
>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
>>
>>
>> I am sharing this Java implementation of mine that is based on the paper - Collaborative Filtering with Implicit Datasets. The implementation is multi-treading and can be easily extended to use it on Hadoop. In fact this approach would possibly work with non-implicit datasets, but further testing is needed. The algorithm is tried and tested on an implicit TV-viewing dataset, and the performance was pretty good (details to follow).
>
> --
> This message is automatically generated by JIRA.
> If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
> For more information on JIRA, see: http://www.atlassian.com/software/jira
>
>

--
Lance Norskog
[EMAIL PROTECTED]

I think Tamas referred to matrices that are symmetric (only the upper
triangular half would need to be stored) not diagonal matrices.
On 14.01.2012 05:25, Lance Norskog wrote:
> org.apache.mahout.math.DiagonalMatrix
>
> It even supports sparse values in the diagonal!
>
> On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
> <[EMAIL PROTECTED]> wrote:
>>
>>    [ https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670 ]
>>
>> Tamas Jambor commented on MAHOUT-737:
>> -------------------------------------
>>
>> I think so. When will you be releasing 0.6? I am planning to do a bit more extensive testing in the next 2-3 weeks.
>>
>>> Implicit Alternating Least Squares SVD
>>> --------------------------------------
>>>
>>>                 Key: MAHOUT-737
>>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
>>>             Project: Mahout
>>>          Issue Type: New Feature
>>>          Components: Collaborative Filtering
>>>    Affects Versions: 0.6
>>>            Reporter: Tamas Jambor
>>>            Assignee: Sebastian Schelter
>>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
>>>
>>>
>>> I am sharing this Java implementation of mine that is based on the paper - Collaborative Filtering with Implicit Datasets. The implementation is multi-treading and can be easily extended to use it on Hadoop. In fact this approach would possibly work with non-implicit datasets, but further testing is needed. The algorithm is tried and tested on an implicit TV-viewing dataset, and the performance was pretty good (details to follow).
>>
>> --
>> This message is automatically generated by JIRA.
>> If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
>> For more information on JIRA, see: http://www.atlassian.com/software/jira
>>
>>
>
>
>

thanks, ideally I would need both symmetric and diagonal.

On Sat, Jan 14, 2012 at 8:26 AM, Sebastian Schelter <[EMAIL PROTECTED]> wrote:

> I think Tamas referred to matrices that are symmetric (only the upper
> triangular half would need to be stored) not diagonal matrices.
>
>
> On 14.01.2012 05:25, Lance Norskog wrote:
> > org.apache.mahout.math.DiagonalMatrix
> >
> > It even supports sparse values in the diagonal!
> >
> > On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
> > <[EMAIL PROTECTED]> wrote:
> >>
> >>    [
> https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670]
> >>
> >> Tamas Jambor commented on MAHOUT-737:
> >> -------------------------------------
> >>
> >> I think so. When will you be releasing 0.6? I am planning to do a bit
> more extensive testing in the next 2-3 weeks.
> >>
> >>> Implicit Alternating Least Squares SVD
> >>> --------------------------------------
> >>>
> >>>                 Key: MAHOUT-737
> >>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
> >>>             Project: Mahout
> >>>          Issue Type: New Feature
> >>>          Components: Collaborative Filtering
> >>>    Affects Versions: 0.6
> >>>            Reporter: Tamas Jambor
> >>>            Assignee: Sebastian Schelter
> >>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch,
> MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
> >>>
> >>>
> >>> I am sharing this Java implementation of mine that is based on the
> paper - Collaborative Filtering with Implicit Datasets. The implementation
> is multi-treading and can be easily extended to use it on Hadoop. In fact
> this approach would possibly work with non-implicit datasets, but further
> testing is needed. The algorithm is tried and tested on an implicit
> TV-viewing dataset, and the performance was pretty good (details to follow).
> >>
> >> --
> >> This message is automatically generated by JIRA.
> >> If you think it was sent incorrectly, please contact your JIRA
> administrators:
> https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
> >> For more information on JIRA, see:
> http://www.atlassian.com/software/jira
> >>
> >>
> >
> >
> >
>
>

There is a packed symmetric matrix impl in the Stochastic SVD stuff,
but it is hard-coded to a packed implementation.
org.apache.mahout.math.hadoop.stochasticsvd.UpperTriangular - mahout/math

You could recode this to use the Vector class for storage. Be sure to
run all of the Matrix unit tests if you do this; Matrix has a lot of
things that can go wrong.

On Sat, Jan 14, 2012 at 2:03 AM, Tamas Jambor <[EMAIL PROTECTED]> wrote:
> thanks, ideally I would need both symmetric and diagonal.
>
> On Sat, Jan 14, 2012 at 8:26 AM, Sebastian Schelter <[EMAIL PROTECTED]> wrote:
>
>> I think Tamas referred to matrices that are symmetric (only the upper
>> triangular half would need to be stored) not diagonal matrices.
>>
>>
>> On 14.01.2012 05:25, Lance Norskog wrote:
>> > org.apache.mahout.math.DiagonalMatrix
>> >
>> > It even supports sparse values in the diagonal!
>> >
>> > On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
>> > <[EMAIL PROTECTED]> wrote:
>> >>
>> >>    [
>> https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670]
>> >>
>> >> Tamas Jambor commented on MAHOUT-737:
>> >> -------------------------------------
>> >>
>> >> I think so. When will you be releasing 0.6? I am planning to do a bit
>> more extensive testing in the next 2-3 weeks.
>> >>
>> >>> Implicit Alternating Least Squares SVD
>> >>> --------------------------------------
>> >>>
>> >>>                 Key: MAHOUT-737
>> >>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
>> >>>             Project: Mahout
>> >>>          Issue Type: New Feature
>> >>>          Components: Collaborative Filtering
>> >>>    Affects Versions: 0.6
>> >>>            Reporter: Tamas Jambor
>> >>>            Assignee: Sebastian Schelter
>> >>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch,
>> MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
>> >>>
>> >>>
>> >>> I am sharing this Java implementation of mine that is based on the
>> paper - Collaborative Filtering with Implicit Datasets. The implementation
>> is multi-treading and can be easily extended to use it on Hadoop. In fact
>> this approach would possibly work with non-implicit datasets, but further
>> testing is needed. The algorithm is tried and tested on an implicit
>> TV-viewing dataset, and the performance was pretty good (details to follow).
>> >>
>> >> --
>> >> This message is automatically generated by JIRA.
>> >> If you think it was sent incorrectly, please contact your JIRA
>> administrators:
>> https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
>> >> For more information on JIRA, see:
>> http://www.atlassian.com/software/jira
>> >>
>> >>
>> >
>> >
>> >
>>
>>

--
Lance Norskog
[EMAIL PROTECTED]

I Would recommend not worrying about having a special symmetric matrix for now.  It won't make a huge difference and it will be a pain to convert the one that ssvd uses.

The diagonal matrix could make a much bigger difference.

Sent from my iPhone

On Jan 14, 2012, at 15:06, Lance Norskog <[EMAIL PROTECTED]> wrote:

> There is a packed symmetric matrix impl in the Stochastic SVD stuff,
> but it is hard-coded to a packed implementation.
> org.apache.mahout.math.hadoop.stochasticsvd.UpperTriangular - mahout/math
>
> You could recode this to use the Vector class for storage. Be sure to
> run all of the Matrix unit tests if you do this; Matrix has a lot of
> things that can go wrong.
>
> On Sat, Jan 14, 2012 at 2:03 AM, Tamas Jambor <[EMAIL PROTECTED]> wrote:
>> thanks, ideally I would need both symmetric and diagonal.
>>
>> On Sat, Jan 14, 2012 at 8:26 AM, Sebastian Schelter <[EMAIL PROTECTED]> wrote:
>>
>>> I think Tamas referred to matrices that are symmetric (only the upper
>>> triangular half would need to be stored) not diagonal matrices.
>>>
>>>
>>> On 14.01.2012 05:25, Lance Norskog wrote:
>>>> org.apache.mahout.math.DiagonalMatrix
>>>>
>>>> It even supports sparse values in the diagonal!
>>>>
>>>> On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
>>>> <[EMAIL PROTECTED]> wrote:
>>>>>
>>>>>    [
>>> https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670]
>>>>>
>>>>> Tamas Jambor commented on MAHOUT-737:
>>>>> -------------------------------------
>>>>>
>>>>> I think so. When will you be releasing 0.6? I am planning to do a bit
>>> more extensive testing in the next 2-3 weeks.
>>>>>
>>>>>> Implicit Alternating Least Squares SVD
>>>>>> --------------------------------------
>>>>>>
>>>>>>                 Key: MAHOUT-737
>>>>>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
>>>>>>             Project: Mahout
>>>>>>          Issue Type: New Feature
>>>>>>          Components: Collaborative Filtering
>>>>>>    Affects Versions: 0.6
>>>>>>            Reporter: Tamas Jambor
>>>>>>            Assignee: Sebastian Schelter
>>>>>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch,
>>> MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
>>>>>>
>>>>>>
>>>>>> I am sharing this Java implementation of mine that is based on the
>>> paper - Collaborative Filtering with Implicit Datasets. The implementation
>>> is multi-treading and can be easily extended to use it on Hadoop. In fact
>>> this approach would possibly work with non-implicit datasets, but further
>>> testing is needed. The algorithm is tried and tested on an implicit
>>> TV-viewing dataset, and the performance was pretty good (details to follow).
>>>>>
>>>>> --
>>>>> This message is automatically generated by JIRA.
>>>>> If you think it was sent incorrectly, please contact your JIRA
>>> administrators:
>>> https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
>>>>> For more information on JIRA, see:
>>> http://www.atlassian.com/software/jira
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>
>>>
>
>
>
> --
> Lance Norskog
> [EMAIL PROTECTED]

yes, the UpperTriangular in SSVD doesn't support some of the Matrix
operations. On top of it, it is built with some assumptions that
Mahout support in general should not have, or have an alternative for,
for efficiency sake of ssvd:

1) it assumes the matrix is dense.
2) most importantly, it assumes the matrix is sufficiently small (it
packs it into a single vector). that means it may experience problems
when dense allocation exceeds  2G contiguous heap chunk.
3) consequently, it only serializes into a dense vector.

for diagonal matrices, the same ssvd uses vector form. It does mean
that you don't have a nice single method to implement multiplication;
but it is just one vector-scalar loop instead, but oh well.

-d

On Sat, Jan 14, 2012 at 2:03 AM, Tamas Jambor <[EMAIL PROTECTED]> wrote:
> thanks, ideally I would need both symmetric and diagonal.
>
> On Sat, Jan 14, 2012 at 8:26 AM, Sebastian Schelter <[EMAIL PROTECTED]> wrote:
>
>> I think Tamas referred to matrices that are symmetric (only the upper
>> triangular half would need to be stored) not diagonal matrices.
>>
>>
>> On 14.01.2012 05:25, Lance Norskog wrote:
>> > org.apache.mahout.math.DiagonalMatrix
>> >
>> > It even supports sparse values in the diagonal!
>> >
>> > On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
>> > <[EMAIL PROTECTED]> wrote:
>> >>
>> >>    [
>> https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670]
>> >>
>> >> Tamas Jambor commented on MAHOUT-737:
>> >> -------------------------------------
>> >>
>> >> I think so. When will you be releasing 0.6? I am planning to do a bit
>> more extensive testing in the next 2-3 weeks.
>> >>
>> >>> Implicit Alternating Least Squares SVD
>> >>> --------------------------------------
>> >>>
>> >>>                 Key: MAHOUT-737
>> >>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
>> >>>             Project: Mahout
>> >>>          Issue Type: New Feature
>> >>>          Components: Collaborative Filtering
>> >>>    Affects Versions: 0.6
>> >>>            Reporter: Tamas Jambor
>> >>>            Assignee: Sebastian Schelter
>> >>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch,
>> MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
>> >>>
>> >>>
>> >>> I am sharing this Java implementation of mine that is based on the
>> paper - Collaborative Filtering with Implicit Datasets. The implementation
>> is multi-treading and can be easily extended to use it on Hadoop. In fact
>> this approach would possibly work with non-implicit datasets, but further
>> testing is needed. The algorithm is tried and tested on an implicit
>> TV-viewing dataset, and the performance was pretty good (details to follow).
>> >>
>> >> --
>> >> This message is automatically generated by JIRA.
>> >> If you think it was sent incorrectly, please contact your JIRA
>> administrators:
>> https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
>> >> For more information on JIRA, see:
>> http://www.atlassian.com/software/jira
>> >>
>> >>
>> >
>> >
>> >
>>
>>

btw I think Mahout sparse matrix should be fairly efficient when
handling diagonal and triangular or Hessenberg like stuff.
On Mon, Jan 16, 2012 at 11:18 AM, Dmitriy Lyubimov <[EMAIL PROTECTED]> wrote:
> yes, the UpperTriangular in SSVD doesn't support some of the Matrix
> operations. On top of it, it is built with some assumptions that
> Mahout support in general should not have, or have an alternative for,
> for efficiency sake of ssvd:
>
> 1) it assumes the matrix is dense.
> 2) most importantly, it assumes the matrix is sufficiently small (it
> packs it into a single vector). that means it may experience problems
> when dense allocation exceeds  2G contiguous heap chunk.
> 3) consequently, it only serializes into a dense vector.
>
> for diagonal matrices, the same ssvd uses vector form. It does mean
> that you don't have a nice single method to implement multiplication;
> but it is just one vector-scalar loop instead, but oh well.
>
> -d
>
> On Sat, Jan 14, 2012 at 2:03 AM, Tamas Jambor <[EMAIL PROTECTED]> wrote:
>> thanks, ideally I would need both symmetric and diagonal.
>>
>> On Sat, Jan 14, 2012 at 8:26 AM, Sebastian Schelter <[EMAIL PROTECTED]> wrote:
>>
>>> I think Tamas referred to matrices that are symmetric (only the upper
>>> triangular half would need to be stored) not diagonal matrices.
>>>
>>>
>>> On 14.01.2012 05:25, Lance Norskog wrote:
>>> > org.apache.mahout.math.DiagonalMatrix
>>> >
>>> > It even supports sparse values in the diagonal!
>>> >
>>> > On Fri, Jan 13, 2012 at 9:00 AM, Tamas Jambor (Commented) (JIRA)
>>> > <[EMAIL PROTECTED]> wrote:
>>> >>
>>> >>    [
>>> https://issues.apache.org/jira/browse/MAHOUT-737?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185670#comment-13185670]
>>> >>
>>> >> Tamas Jambor commented on MAHOUT-737:
>>> >> -------------------------------------
>>> >>
>>> >> I think so. When will you be releasing 0.6? I am planning to do a bit
>>> more extensive testing in the next 2-3 weeks.
>>> >>
>>> >>> Implicit Alternating Least Squares SVD
>>> >>> --------------------------------------
>>> >>>
>>> >>>                 Key: MAHOUT-737
>>> >>>                 URL: https://issues.apache.org/jira/browse/MAHOUT-737
>>> >>>             Project: Mahout
>>> >>>          Issue Type: New Feature
>>> >>>          Components: Collaborative Filtering
>>> >>>    Affects Versions: 0.6
>>> >>>            Reporter: Tamas Jambor
>>> >>>            Assignee: Sebastian Schelter
>>> >>>         Attachments: MAHOUT-737-2.patch, MAHOUT-737.patch,
>>> MAHOUT-737.patch, MAHOUT-737.patch, MAHOUT-737.patch
>>> >>>
>>> >>>
>>> >>> I am sharing this Java implementation of mine that is based on the
>>> paper - Collaborative Filtering with Implicit Datasets. The implementation
>>> is multi-treading and can be easily extended to use it on Hadoop. In fact
>>> this approach would possibly work with non-implicit datasets, but further
>>> testing is needed. The algorithm is tried and tested on an implicit
>>> TV-viewing dataset, and the performance was pretty good (details to follow).
>>> >>
>>> >> --
>>> >> This message is automatically generated by JIRA.
>>> >> If you think it was sent incorrectly, please contact your JIRA
>>> administrators:
>>> https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
>>> >> For more information on JIRA, see:
>>> http://www.atlassian.com/software/jira
>>> >>
>>> >>
>>> >
>>> >
>>> >
>>>
>>>