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: SGD: Logistic regression package in Mahout
Date Thu, 01 Nov 2012 16:11:35 GMT
The output will be in a report file under core/target, I think.  Look for a
file with OnlineLogisticRegressionTest in the name.  Saving to a separate
file is a fine approach as well.

On Thu, Nov 1, 2012 at 6:45 AM, Rajesh Nikam <rajeshnikam@gmail.com> wrote:

> Thanks Ted for providing testcase that helped me to look into details of
> the problem that I am facing.
>
> Got how to run test case using maven:
>
> mvn test
> -Dtest="org.apache.mahout.classifier.sgd.OnlineLogisticRegressionTest"
>
> However I could not see printf output spitted on console, so I have saved
> output to file.
>
> Now I will look at the results and update in case of any issue.
>
> Thanks
> Rajesh
>
>
> On Thu, Nov 1, 2012 at 1:05 PM, Rajesh Nikam <rajeshnikam@gmail.com>
> wrote:
>
> > 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