You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mahout.apache.org by "Herta, Christian" <Ch...@htw-berlin.de> on 2012/02/07 11:15:15 UTC
Multi Layer Perceptron for Mahout
Is there anybody else working on a Multilayer Perceptron (Neural Network) for
Mahout? If not, I'm going to implement it.There is a gradient machine for
classification. But I think the gradient machine is to restricted. Also, the
implementation is stochastic gradient descent only. Therefore it's not possible
to distribute the learning.
What's the procedure to start? Do I have to create a jira issue with a
description and a road map?
Re: Multi Layer Perceptron for Mahout
Posted by "Herta, Christian" <Ch...@htw-berlin.de>.
Hello Atul,
thanks for your interest. Yes, I created a jira issue(MAHOUT-976
[https://issues.apache.org/jira/browse/MAHOUT-976] ) and started with the
implementation. I will upload a patch in the next days. Then you can get an
impression what's going on. I am interested in your feedback.
The most complicated part is not to be confused by the indices and the notation.
If it's necessary I will write the math in latex document in "the same" notation
I used in the code.
Cheers
Christian
Atul Kulkarni <at...@gmail.com> hat am 9. Februar 2012 um 19:25
geschrieben:
> Have you created a JIRA for this? I would like to help.
>
> On Tue, Feb 7, 2012 at 9:43 AM, Ted Dunning <te...@gmail.com> wrote:
>
> > The JIRA will be the best place to talk about your strategy for
> > implementation. Parallelization of NN training isn't a trivial problem.
> > We should learn from recent experience of others.
> >
> > On Tue, Feb 7, 2012 at 2:43 AM, Paritosh Ranjan <pr...@xebia.com> wrote:
> >
> > > I'll suggest to create a jira, mark it for 0.7 and submit the patch.
> > > That's how I did it, and it worked.
> > >
> > >
> > > Herta, Christian wrote:
> > >
> > >> Is there anybody else working on a Multilayer Perceptron (Neural
> > Network)
> > >> for
> > >> Mahout? If not, I'm going to implement it.There is a gradient machine
> > for
> > >> classification. But I think the gradient machine is to restricted. Also,
> > >> the
> > >> implementation is stochastic gradient descent only. Therefore it's not
> > >> possible
> > >> to distribute the learning.
> > >>
> > >> What's the procedure to start? Do I have to create a jira issue with a
> > >> description and a road map?
> > >>
> > >
> > >
> >
>
>
>
> --
> Regards,
> Atul Kulkarni
Re: Multi Layer Perceptron for Mahout
Posted by Atul Kulkarni <at...@gmail.com>.
Have you created a JIRA for this? I would like to help.
On Tue, Feb 7, 2012 at 9:43 AM, Ted Dunning <te...@gmail.com> wrote:
> The JIRA will be the best place to talk about your strategy for
> implementation. Parallelization of NN training isn't a trivial problem.
> We should learn from recent experience of others.
>
> On Tue, Feb 7, 2012 at 2:43 AM, Paritosh Ranjan <pr...@xebia.com> wrote:
>
> > I'll suggest to create a jira, mark it for 0.7 and submit the patch.
> > That's how I did it, and it worked.
> >
> >
> > Herta, Christian wrote:
> >
> >> Is there anybody else working on a Multilayer Perceptron (Neural
> Network)
> >> for
> >> Mahout? If not, I'm going to implement it.There is a gradient machine
> for
> >> classification. But I think the gradient machine is to restricted. Also,
> >> the
> >> implementation is stochastic gradient descent only. Therefore it's not
> >> possible
> >> to distribute the learning.
> >>
> >> What's the procedure to start? Do I have to create a jira issue with a
> >> description and a road map?
> >>
> >
> >
>
--
Regards,
Atul Kulkarni
Re: Multi Layer Perceptron for Mahout
Posted by Ted Dunning <te...@gmail.com>.
The JIRA will be the best place to talk about your strategy for
implementation. Parallelization of NN training isn't a trivial problem.
We should learn from recent experience of others.
On Tue, Feb 7, 2012 at 2:43 AM, Paritosh Ranjan <pr...@xebia.com> wrote:
> I'll suggest to create a jira, mark it for 0.7 and submit the patch.
> That's how I did it, and it worked.
>
>
> Herta, Christian wrote:
>
>> Is there anybody else working on a Multilayer Perceptron (Neural Network)
>> for
>> Mahout? If not, I'm going to implement it.There is a gradient machine for
>> classification. But I think the gradient machine is to restricted. Also,
>> the
>> implementation is stochastic gradient descent only. Therefore it's not
>> possible
>> to distribute the learning.
>>
>> What's the procedure to start? Do I have to create a jira issue with a
>> description and a road map?
>>
>
>
Re: Multi Layer Perceptron for Mahout
Posted by Paritosh Ranjan <pr...@xebia.com>.
I'll suggest to create a jira, mark it for 0.7 and submit the patch.
That's how I did it, and it worked.
Herta, Christian wrote:
> Is there anybody else working on a Multilayer Perceptron (Neural Network) for
> Mahout? If not, I'm going to implement it.There is a gradient machine for
> classification. But I think the gradient machine is to restricted. Also, the
> implementation is stochastic gradient descent only. Therefore it's not possible
> to distribute the learning.
>
>
> What's the procedure to start? Do I have to create a jira issue with a
> description and a road map?
>
>
>
>