mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rajesh Nikam <rajeshni...@gmail.com>
Subject Re: SGD: Logistic regression package in Mahout
Date Thu, 01 Nov 2012 07:35:03 GMT
Hi Mat,

Thanks for pointing out link for JIRA for this particular case.

Could you extend one more help:

I have not used maven for building and running java classes. I am looking
at
http://maven.apache.org/guides/getting-started/index.html

Could you please point out how to build & run any specific class like
OnlineLogisticRegressionTest.java from mahout.

Thanks
Rajesh

On Wed, Oct 31, 2012 at 8:15 PM, Mat Kelcey <matthew.kelcey@gmail.com>wrote:

> 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