Due to recent updates, all users are required to create an Altair One account to login to the RapidMiner community. Click the Register button to create your account using the same email that you have previously used to login to the RapidMiner community. This will ensure that any previously created content will be synced to your Altair One account. Once you login, you will be asked to provide a username that identifies you to other Community users. Email us at Community with questions.
"Classification performance of one-class SVM"
TobiasMalbrecht
Moderator, Employee, Member Posts: 295 RM Product Management
Posted in SourceForge forum at http://sourceforge.net/forum/forum.php?thread_id=2185902&forum_id=390413:
Does anybody have any example/project they are willing to share showing the
use of one-class libSVM; I've been playing around with rapidminer for the past
day (which is amazing btw-- tnx!) but using the one-class and modifying any
of the parameters (gamma, coef, C, nu, etc.) does not affect the classification
result which is 100%. I've even modified the data on purpose (to mess it up)
and no luck on seeing any changes.
I've also exported the data (as an arff file) and imported it in to Weka and
the results were more reasonable (changing gamma from 0.01 to 0.07 changed the
classification correctness to ~88%). (This might be off topic, but if anybody
knows how to do PCA on one-class data set in Weka that would be also helpful)
Here is the simplified model xml file:
Does anybody have any example/project they are willing to share showing the
use of one-class libSVM; I've been playing around with rapidminer for the past
day (which is amazing btw-- tnx!) but using the one-class and modifying any
of the parameters (gamma, coef, C, nu, etc.) does not affect the classification
result which is 100%. I've even modified the data on purpose (to mess it up)
and no luck on seeing any changes.
I've also exported the data (as an arff file) and imported it in to Weka and
the results were more reasonable (changing gamma from 0.01 to 0.07 changed the
classification correctness to ~88%). (This might be off topic, but if anybody
knows how to do PCA on one-class data set in Weka that would be also helpful)
Here is the simplified model xml file:
Any help is very much appreciated! Thanks!
<operator name="Root" class="Process" expanded="yes">
<parameter key="logfile" value="/home/d/rm_workspace/log.log"/>
<parameter key="resultfile" value="/home/d/rm_workspace/results.res"/>
<operator name="ArffExampleSource" class="ArffExampleSource">
<parameter key="data_file" value="file.arff"/>
<parameter key="label_attribute" value="class"/>
</operator>
<operator name="Normalization" class="Normalization">
<parameter key="min" value="-1.0"/>
</operator>
<operator name="PCA" class="PCA">
</operator>
<operator name="XValidation" class="XValidation" expanded="yes">
<operator name="LibSVMLearner" class="LibSVMLearner">
<parameter key="cache_size" value="150"/>
<list key="class_weights">
</list>
<parameter key="coef0" value="0.03125"/>
<parameter key="confidence_for_multiclass" value="false"/>
<parameter key="degree" value="1"/>
<parameter key="gamma" value="0.01"/>
<parameter key="nu" value="9.765625E-4"/>
<parameter key="svm_type" value="one-class"/>
</operator>
<operator name="OperatorChain" class="OperatorChain" expanded="yes">
<operator name="ModelApplier (2)" class="ModelApplier">
<list key="application_parameters">
</list>
</operator>
<operator name="Performance" class="Performance">
</operator>
</operator>
</operator>
</operator>
Tagged:
0
Answers
To transform the data you need to add ModelApplier immediately after the PCA step.
in any case I also tried with and even simpler example: where the arff file is: and the class prediction is 100%
[attachment deleted by admin]