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 13:45:33 GMT
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