Logistic regression with L1 regularization is generally at least as good as
SVM. The problem with SVM is that it uses radially symmetric
regularization which doesn't learn sparse solutions very well. L1
regularization is much better for that.
On Tue, Dec 24, 2013 at 10:06 AM, Steven Bourke wro
Just test out libsvm against log regression on a sample of your data to get an
understanding of upside downside for your particular problem
Sent from my iPhone
> On 24 Dec 2013, at 15:55, Tharindu Rusira wrote:
>
> Thanks all for the words of wisdom :) ,
>
> @Ted, I'm coming from a text mini
Thanks all for the words of wisdom :) ,
@Ted, I'm coming from a text mining background. Many text books recommend
SVM because of its impressive performance with vectors having a larger
cardinality which is the usual case when dealing with text documents. Do
you think logistic regression would perf
You can paralize svm using same equations (which has slight difference)
explained in
http://books.google.co.in/books/about/DATA_MINING.html?id=IYc2muhCbmEC&redir_esc=y
But i dont gaurentee about the performance. for some 100 MB data it takes
10 min to train the data.
On Tue, Dec 24, 2013 at 3:30
someone tried to implement SVM in a summer google code but it turns out map
reduced version of svm is too difficult to implement and they dropped the
project.
I bet you can train via libsvm and use just classification part with map
reduce but if I have a choice I prefer logistic regression too
~--
You might try logistic regression with regularization for a very similar
result.
On Mon, Dec 23, 2013 at 11:57 PM, Sebastian Schelter <
ssc.o...@googlemail.com> wrote:
> Hi Tharindu,
>
> There is no SVM implementation in an official release.
>
> --sebastian
>
> On 24.12.2013 08:02, Tharindu Rusi
Hi Tharindu,
There is no SVM implementation in an official release.
--sebastian
On 24.12.2013 08:02, Tharindu Rusira wrote:
> Hi all,
> Do we have a SVM implementation in Mahout(either sequential or mapreduce)?
> I was searching in JIRA and MAHOUT-14[1] proposes a SVM implementation and
> also M
On Wed, Jun 20, 2012 at 6:19 AM, Sebastian Schelter wrote:
> I'm just asking because I review a paper where Mahout's SVM support on
> M/R is mentioned :)
>
Interesting. We don't have anything release, I don't think.
> As far as I recall All-Reduce is map + an aggregation tree that feeds
> bac
I'm just asking because I review a paper where Mahout's SVM support on
M/R is mentioned :)
As far as I recall All-Reduce is map + an aggregation tree that feeds
back the result?
Giraph supports aggregators where every worker instance preaggregates
the data, the master computes the final aggregati
No.
But L1 regularization for hinge loss should give essentially identical results.
We currently support logit loss but this is pluggable. This is in the sgd
framework which currently lacks a map reduce implementation.
Speaking of that, does anybody offhand know whether giraph supports an
a
10 matches
Mail list logo
