mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mat Kelcey <matthew.kel...@gmail.com>
Subject Re: SGD: Logistic regression package in Mahout
Date Wed, 31 Oct 2012 14:45:02 GMT
Rajesh, Ted has added the test case code already
https://issues.apache.org/jira/browse/MAHOUT-1107

On 31 October 2012 05:14, Rajesh Nikam <rajeshnikam@gmail.com> wrote:

> Hi Ted,
>
> Please update once JIRA and test case is uploaded.
>
> Looking forward for your reply.
>
> Thanks
> Rajesh
>
> On Wed, Oct 31, 2012 at 11:00 AM, Rajesh Nikam <rajeshnikam@gmail.com
> >wrote:
>
> > Hi Ted,
> >
> > Thanks for reply. I will wait for JIRA and hope to get rid of any
> encoding
> > issue.
> >
> > Thanks,
> > Rajesh
> > On Oct 31, 2012 5:24 AM, "Ted Dunning" <ted.dunning@gmail.com> wrote:
> >
> >> OK.  I am back up for air.
> >>
> >> Rajesh,
> >>
> >> As I am sure you know, most folks here contribute on their own time.  I
> >> have been busy with my day job and unable to help with this until just
> >> now.
> >>
> >> I just wrote a test case that looks at the Iris data set.  The results
> are
> >> categorically different from yours.
> >>
> >> That substantiates my original feeling that your encoding of the data is
> >> problematic.  I will file a JIRA and attach a test case that you can
> look
> >> at.  Then we can see what the differences are.
> >>
> >>
> >> On Tue, Oct 23, 2012 at 1:28 AM, Rajesh Nikam <rajeshnikam@gmail.com>
> >> wrote:
> >>
> >> > Hi,
> >> >
> >> > Is there development happening on fixing issue with SGD that generates
> >> > models which are as good as random prediction?
> >> >
> >> > I am not sure why such issue is not noticed and raised by others ?
> >> > May be this specific algo is not used in practical applications.
> >> >
> >> > Thanks,
> >> > Rajesh
> >> >
> >> >
> >> > >>
> >> > >> On Tue, Oct 16, 2012 at 10:23 PM, Ted Dunning <
> ted.dunning@gmail.com
> >> > >wrote:
> >> > >>
> >> > >>> Rajesh,
> >> > >>>
> >> > >>> In the testing that I did, I ran 100, 1000 and 10,000 passes
> through
> >> > the
> >> > >>> data.  All produced identical results.  Thus it isn't an issue
of
> >> SGD
> >> > >>> converging.
> >> > >>>
> >> > >>> I also did a parameter scan of lambda and saw no effect.
> >> > >>>
> >> > >>> I also did the standard thing in R with glm and got the expected
> >> > >>> (correct)
> >> > >>> results.
> >> > >>>
> >> > >>> I haven't looked yet in detail, but I really suspect that
the
> >> reading
> >> > of
> >> > >>> the data is horked.  This is exactly how that behaves.
> >> > >>>
> >> > >>> On Tue, Oct 16, 2012 at 4:49 AM, Rajesh Nikam <
> >> rajeshnikam@gmail.com>
> >> > >>> wrote:
> >> > >>>
> >> > >>> > Hi Ted,
> >> > >>> >
> >> > >>> > I was thinking, this might be due to having only 100
instances
> for
> >> > >>> > training.
> >> > >>> >
> >> > >>> > So I have created test set with two classes having ~49K
> instances,
> >> > >>> included
> >> > >>> > all features as predictors.
> >> > >>> > PFA sgd.grps.zip with test file.
> >> > >>> >
> >> > >>> > mahout trainlogistic --input
> >> /usr/local/mahout/trainme/sgd-grps.csv
> >> > >>> > --output /usr/local/mahout/trainme/sgd-grps.model --target
class
> >> > >>> > --categories 2 --features 128 --types n --predictors
a1 a2 a3 a4
> >> a5
> >> > a6
> >> > >>> a7
> >> > >>> > a8 a9 a10 a11 a12 a13 a14 a15 a16 a17 a18 a19 a20 a21
a22 a23
> a24
> >> a25
> >> > >>> a26
> >> > >>> > a27 a28 a29 a30 a31 a32 a33 a34 a35 a36 a37 a38 a39 a40
a41 a42
> >> a43
> >> > >>> a44 a45
> >> > >>> > a46 a47 a48 a49 a50 a51 a52 a53 a54 a55 a56 a57 a58 a59
a60 a61
> >> a62
> >> > >>> a63 a64
> >> > >>> > a65 a66 a67 a68 a69 a70 a71 a72 a73 a74 a75 a76 a77 a78
a79 a80
> >> a81
> >> > >>> a82 a83
> >> > >>> > a84 a85 a86 a87 a88 a89 a90 a91 a92 a93 a94 a95 a96 a97
a98 a99
> >> a100
> >> > >>> a101
> >> > >>> > a102 a103 a104 a105 a106 a107 a108 a109 a110 a111 a112
a113 a114
> >> a115
> >> > >>> a116
> >> > >>> > a117 a118 a119 a120 a121 a122 a123 a124 a125 a126 a127
> >> > >>> >
> >> > >>> >
> >> > >>> > mahout runlogistic --input
> /usr/local/mahout/trainme/sgd-grps.csv
> >> > >>> --model
> >> > >>> > /usr/local/mahout/trainme/sgd-grps.model --auc --confusion
> >> > >>> >
> >> > >>> > Still the results are similar, it classifies everything
as
> >> class_1.
> >> > >>> >
> >> > >>> > AUC = 0.50
> >> > >>> > confusion: [[*26563.0, 23006.0*], [0.0, 0.0]]
> >> > >>> > entropy: [[-0.0, -0.0], [-46.1, -21.4]]
> >> > >>> >
> >> > >>> > I am not sure why this is failing all the time.
> >> > >>> >
> >> > >>> > Looking forward for your reply.
> >> > >>> >
> >> > >>> > Thanks
> >> > >>> > Rajesh
> >> > >>> >
> >> > >>> >
> >> > >>> >
> >> > >>> > On Tue, Oct 16, 2012 at 3:57 AM, Ted Dunning <
> >> ted.dunning@gmail.com>
> >> > >>> > wrote:
> >> > >>> >
> >> > >>> > > I would love to help and will before long.  Just
can't do it
> in
> >> the
> >> > >>> first
> >> > >>> > > part of this week.
> >> > >>> > >
> >> > >>> > > On Mon, Oct 15, 2012 at 6:28 AM, Rajesh Nikam <
> >> > rajeshnikam@gmail.com
> >> > >>> >
> >> > >>> > > wrote:
> >> > >>> > >
> >> > >>> > > > Hello,
> >> > >>> > > >
> >> > >>> > > > I have asked below question on issue with using
sgd on
> mahout
> >> > >>> forum.
> >> > >>> > > >
> >> > >>> > > > Similar issue with sgd is reported by
> >> > >>> > > >
> >> > >>> > > >
> >> > >>> > >
> >> > >>> >
> >> > >>>
> >> >
> >>
> http://stackoverflow.com/questions/11221436/using-sgd-classifier-in-mahout
> >> > >>> > > >
> >> > >>> > > > Even below link has similar output:
> >> > >>> > > >
> >> > >>> > > > AUC = 0.57*confusion: [[27.0, 13.0], [0.0,
0.0]]*
> >> > >>> > > > entropy: [[-0.4, -0.3], [-1.2, -0.7]]
> >> > >>> > > >
> >> > >>> > > >
> >> > >>> > > >
> >> > >>> >
> >> > >>>
> >> >
> http://sujitpal.blogspot.in/2012/09/learning-mahout-classification.html
> >> > >>> > > >
> >> > >>> > > > I am still wannder confusion how then this
model works and
> >> used
> >> > by
> >> > >>> > many ?
> >> > >>> > > > Not able to get any points on how to use SGD
that generates
> >> > >>> effective
> >> > >>> > > > model.
> >> > >>> > > >
> >> > >>> > > > Could someone point out what is missing in
input file or
> >> provided
> >> > >>> > > > parameters.
> >> > >>> > > >
> >> > >>> > > > I appreciate your help.
> >> > >>> > > >
> >> > >>> > > > Below is description of steps that I followed.
> >> > >>> > > >
> >> > >>> > > > PF Attached uses input files for experiment.
> >> > >>> > > >
> >> > >>> > > > I am using Iris Plants Database from Michael
Marshall. PFA
> >> > >>> iris.arff.
> >> > >>> > > > Converted this to csv file just by updating
header:
> >> > >>> iris-3-classes.csv
> >> > >>> > > >
> >> > >>> > > > mahout org.apache.mahout.classifier.
> >> > >>> > > > sgd.TrainLogistic --input
> >> > >>> > > /usr/local/mahout/trunk/*iris-3-classes.csv*--features
4
> >> --output
> >> > >>> > > /usr/local/mahout/trunk/
> >> > >>> > > > *iris-3-classes.model* --target class *--categories
3*
> >> > --predictors
> >> > >>> > > > sepallength sepalwidth petallength petalwidth
--types n
> >> > >>> > > >
> >> > >>> > > > >> it gave following error.
> >> > >>> > > > Exception in thread "main"
> java.lang.IllegalArgumentException:
> >> > Can
> >> > >>> only
> >> > >>> > > > call classifyScalar with two categories
> >> > >>> > > >
> >> > >>> > > > Now created csv with only 2 classes. PFA iris-2-classes.csv
> >> > >>> > > >
> >> > >>> > > > >> trained iris-2-classes.csv with sgd
> >> > >>> > > >
> >> > >>> > > > mahout org.apache.mahout.classifier.sgd.TrainLogistic
> --input
> >> > >>> > > > /usr/local/mahout/trunk/*iris-2-classes.csv*
--features 4
> >> > --output
> >> > >>> > > > /usr/local/mahout/trunk/*iris-2-classes.mode*l
--target
> class
> >> > >>> > > *--categories
> >> > >>> > > > 2* --predictors sepallength sepalwidth petallength
> petalwidth
> >> > >>> --types n
> >> > >>> > > >
> >> > >>> > > > mahout runlogistic --input
> >> > >>> /usr/local/mahout/trunk/iris-2-classes.csv
> >> > >>> > > > --model /usr/local/mahout/trunk/iris-2-classes.model
--auc
> >> > >>> --confusion
> >> > >>> > > >
> >> > >>> > > > AUC = 0.14
> >> > >>> > > > confusion: [[50.0, 50.0], [0.0, 0.0]]
> >> > >>> > > > entropy: [[-0.6, -0.3], [-0.8, -0.4]]
> >> > >>> > > >
> >> > >>> > > > >> AUC seems to poor. Now changed --predictors
> >> > >>> > > >
> >> > >>> > > > mahout org.apache.mahout.classifier.sgd.TrainLogistic
> --input
> >> > >>> > > > /usr/local/mahout/trunk/*iris-2-classes.csv*
--features 4
> >> > --output
> >> > >>> > > > /usr/local/mahout/trunk/*iris-2-classes.mode*l
--target
> class
> >> > >>> > > *--categories
> >> > >>> > > > 2* --predictors sepalwidth petallength --types
n
> >> > >>> > > >
> >> > >>> > > > mahout runlogistic --input
> >> > >>> /usr/local/mahout/trunk/iris-2-classes.csv
> >> > >>> > > > --model /usr/local/mahout/trunk/iris-2-classes.model
--auc
> >> > >>> --confusion
> >> > >>> > > > --scores
> >> > >>> > > >
> >> > >>> > > > AUC = 0.80
> >> > >>> > > > *confusion: [[50.0, 50.0], [0.0, 0.0]]*
> >> > >>> > > > entropy: [[-0.7, -0.3], [-0.7, -0.4]]
> >> > >>> > > >
> >> > >>> > > > This model classifies everything as category
1 which of no
> >> use.
> >> > >>> > > >
> >> > >>> > > > Thanks
> >> > >>> > > > Rajesh
> >> > >>> > > >
> >> > >>> > > >
> >> > >>> > > >
> >> > >>> > > >
> >> > >>> > >
> >> > >>> >
> >> > >>>
> >> > >>
> >> > >>
> >> > >
> >> >
> >>
> >
>

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