SVM and logistic regression in Mahout

[Patrick Diviacco <pa...@gmail.com>]

Hello,

I'm new to Mahout and I need some information to decide if I can use it for
my thesis experiment.

I've been suggested to use WEKA, but I would like more to program in Java
and I enjoyed using another Apache product such as Lucene.

In the specific, I need to run the following classifiers:

- support vector machines (weka sequential minimal optimization
implementation)

and

- logistic regression

Could you tell me if they are available, and if you know Weka, if they have
the same implementation, or if there is something I should know.. etc..

thanks in advance,
Patrick

Re: SVM and logistic regression in Mahout

[vineet yadav <vi...@gmail.com>]

Hi Patrick,
Currently Mahout don't support support vector machine. But SVM was
part of future plan and check out concern  issues for SVM
implementation(https://issues.apache.org/jira/browse/MAHOUT-14 and
https://issues.apache.org/jira/browse/MAHOUT-334). But I think support
vector machine idea was dropped since SVM is scalable.
You can use mahout  SGD algorithm for logistic
regression(https://cwiki.apache.org/confluence/display/MAHOUT/Logistic+Regression).
Thanks
Vineet Yadav

On Sat, Mar 26, 2011 at 7:08 PM, Patrick Diviacco
<pa...@gmail.com> wrote:
> Hello,
>
> I'm new to Mahout and I need some information to decide if I can use it for
> my thesis experiment.
>
> I've been suggested to use WEKA, but I would like more to program in Java
> and I enjoyed using another Apache product such as Lucene.
>
> In the specific, I need to run the following classifiers:
>
> - support vector machines (weka sequential minimal optimization
> implementation)
>
> and
>
> - logistic regression
>
> Could you tell me if they are available, and if you know Weka, if they have
> the same implementation, or if there is something I should know.. etc..
>
> thanks in advance,
> Patrick
>

Re: SVM and logistic regression in Mahout

[Mat Kelcey <ma...@gmail.com>]

Weka is more than just the gui. Everything is exposed as apis and it's all
java.
Mat

On Mar 26, 2011 6:39 AM, "Patrick Diviacco" <pa...@gmail.com>
wrote:

Hello,

I'm new to Mahout and I need some information to decide if I can use it for
my thesis experiment.

I've been suggested to use WEKA, but I would like more to program in Java
and I enjoyed using another Apache product such as Lucene.

In the specific, I need to run the following classifiers:

- support vector machines (weka sequential minimal optimization
implementation)

and

- logistic regression

Could you tell me if they are available, and if you know Weka, if they have
the same implementation, or if there is something I should know.. etc..

thanks in advance,
Patrick

Re: SVM and logistic regression in Mahout

[Ted Dunning <te...@gmail.com>]

Others have told you about Mahout's lack of SVM and ability with logistic
regression and about the fact that weka is Java.

One question still pending is how large is your data set?  If it is less
than hundreds of thousands of rows, I would actually recommend using R
instead of either Weka or Mahout.  Mahout's purpose with classifiers is to
handle very large problems.  The problems we have handled best so far are
sparse in addition to being large.

If your problem is not sparse or not large, then you might be much more
productive with more traditional tools.

On Sat, Mar 26, 2011 at 6:38 AM, Patrick Diviacco <
patrick.diviacco@gmail.com> wrote:

> Hello,
>
> I'm new to Mahout and I need some information to decide if I can use it for
> my thesis experiment.
>
> I've been suggested to use WEKA, but I would like more to program in Java
> and I enjoyed using another Apache product such as Lucene.
>
> In the specific, I need to run the following classifiers:
>
> - support vector machines (weka sequential minimal optimization
> implementation)
>
> and
>
> - logistic regression
>
> Could you tell me if they are available, and if you know Weka, if they have
> the same implementation, or if there is something I should know.. etc..
>
> thanks in advance,
> Patrick
>