mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ted Dunning <ted.dunn...@gmail.com>
Subject Re: ** Using mahout : stochastic gradient descent **
Date Tue, 09 Oct 2012 15:31:56 GMT
On Tue, Oct 9, 2012 at 3:18 PM, Rajesh Nikam <rajeshnikam@gmail.com> wrote:

> Any idea of mahout - stochastic gradient descent(sgd) ? Model generated is
> similar to SVM.
>

Heh?


> I am using "mahout org.apache.mahout.classifier.sgd.TrainLogistic"  and
> "mahout runlogistic".
>

These are likely to be sub-optimal.  They are specific to a certain kind of
data.  Whether your data fits is anybody's guess since you don't say
anything about your data.


> Including all attributes as predictors I get
>
> AUC = 0.50
> confusion: [[1252978.0, 23003.0], [0.0, 0.0]]
> entropy: [[-0.0, -0.0], [-46.1, -0.8]]
>

This model is complete garbage.

Tried using this generated model use for classify instances however its
> classifies everythings as class 1.
>

No need to try that.  We already know that it won't work.


> One thing is I am using 1.2 million instances on class 1 against 20
> thousand instances from class 2.
>
> Does that affects training ?
>

It might.  It is more likely that you encoded your data poorly, but I can't
tell from what you say.

If you provide more information in your question, you are more likely to
get an answer you can use.  Given the likely time zone difference we have,
you can save days by just asking better questions.

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message