[OTDev] Validation: classification statistics for non-binary class values
Martin Guetlein martin.guetlein at googlemail.comThu Dec 10 12:41:45 CET 2009
- Previous message: [OTDev] Validation: classification statistics for non-binary class values
- Next message: [OTDev] Java Examples
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Hello All, I changed the ClassificationStatistics class (former name was ClassificationInfo) to take the non-binary classification task into account. You can find an updated version of the opentox.owl on the webpage: http://www.opentox.org/data/documents/development/RDF%20files/OpenToxOntology/view Best regards, Martin On Wed, Dec 9, 2009 at 1:45 PM, Martin Guetlein <martin.guetlein at googlemail.com> wrote: > Hi Nina, > > On Wed, Dec 9, 2009 at 12:38 PM, Nina Jeliazkova <nina at acad.bg> wrote: >> Hi Martin, >> >> Martin Guetlein wrote: >> >> Hi Nina, All, >> >> very good Point. Here is how it could look like: >> >> [[ >> default:confusionmatrix >> a ot:ConfusionMatrix ; >> >> # contains numClassValues**2 entries like the following >> >> ot:confusionMatrixValue >> [ >> a ot:ConfusionMatrixValue ; >> dc:value "25"^^xsd:int ; >> ot:confusionMatrixCoordinates ; >> [ >> a ot:ConfusionMatrixCoordinate ; >> dc:predictedValue "active"^^xsd:String ; >> dc:actualValue "moderately_active"^^xsd:String ; >> ] >> ] >> ... >> ]] >> >> I think we will end up with quite a lot of Classes in our opentox.owl. >> >> >> Having large number of classes should be fine, provided we are not >> replicating things under different names. >> >> Here is an idea how we can reuse some of the existing classes : >> >> #This is a cell in a confusion matrix >> <owl:Class rdf:ID="ConfusionMatrixCell"> >> <rdfs:subClassOf rdf:resource="#OpentoxResource"/> >> </owl:Class> >> #the cell is linked to the Feature and the actual value via FeatureValue >> class >> <owl:ObjectProperty rdf:ID="confusionMatrixActual"> >> <rdf:type rdf:resource="&owl;FunctionalProperty"/> >> <rdfs:domain rdf:resource="#ConfusionMatrixCell"/> >> <rdfs:range rdf:resource="#FeatureValue"/> >> </owl:ObjectProperty> >> >> #the cell is linked to the Feature and the predicted value via FeatureValue >> class >> <owl:ObjectProperty rdf:ID="confusionMatrixPredicted"> >> <rdf:type rdf:resource="&owl;FunctionalProperty"/> >> <rdfs:domain rdf:resource="#ConfusionMatrixCell"/> >> <rdfs:range rdf:resource="#FeatureValue"/> >> </owl:ObjectProperty> >> #and the numeric value itself >> <owl:DatatypeProperty rdf:ID="confusionMatrixValue"> >> <rdf:type rdf:resource="&owl;FunctionalProperty"/> >> <rdfs:domain rdf:resource="#ConfusionMatrixCell"/> >> <rdfs:range rdf:resource="&xsd;int"/> >> </owl:DatatypeProperty> >> >> #theabove is to be added in opentox.owl >> >> #instances elsewhere (generated by services) >> <ConfusionMatrixCell rdf:ID="ConfusionMatrixCell_7"> >> <confusionMatrixActual rdf:resource="#FeatureValue_8"/> >> <confusionMatrixPredicted rdf:resource="#FeatureValue_9"/> >> <confusionMatrixValue >> rdf:datatype="&xsd;int">25</confusionMatrixValue> >> </ConfusionMatrixCell> >> >> <FeatureValue rdf:ID="FeatureValue_8"> >> <feature rdf:resource="#Feature_10"/> >> <value rdf:datatype="&xsd;string">active</value> >> </FeatureValue> >> <FeatureValue rdf:ID="FeatureValue_9"> >> <feature rdf:resource="#Feature_10"/> >> <value rdf:datatype="&xsd;string">moderate</value> >> </FeatureValue> >> >> (as a side effect, visualising confusion matrix with relevant links for >> predicted/actual will be straightforward :) >> >> #and using ConfusionMatrixCell to denote a ConfusionMatrix as in your >> proposal. >> >> Any comments? >> > > Looks good, linking to feature values really makes sense. > I will try to integrate this into the opentox ontology today. > > Best regards, > Martin > > >> Best regards, >> Nina >> >> Best Regards, >> Martin >> >> >> >> On Tue, Dec 8, 2009 at 12:57 PM, Nina Jeliazkova <nina at acad.bg> wrote: >> >> >> Hi Martin, >> >> Do we have confusion matrix somewhere in the classification statistics? >> It provides more information than just true positives. >> >> Best regards, >> Nina >> >> >> Martin Guetlein wrote: >> >> >> Hello All, >> >> as Harry noted in one of the last meetings, the classification >> statistics in the validation object only take binary classification >> into account so far. There can of course be more than one class value >> (e.g. inacitve, moderately-active, active). >> Hence, some classification results (e.g. numTruePositives) are now >> available multiple times (once for each class-value). >> >> As collections are not allowed in OWL-DL, I had to create >> intermediate classes (following the scheme Nina proposed for the >> dataset). Here is how an example of the Classification Statistics >> Object may look like: >> >> [[ >> default:thisClassificationStatistics >> a ot:classificationStatistics ; >> >> ot:accuracy "99.0"^^xsd:float ; # accuracy is only available once >> ot:numberUnclassified "26"^^xsd:int ; >> ... >> >> ot:classStatisticEntry >> [ a ot:classStatisticEntry ; >> ot:classValue "moderately_active"^^String ; >> ot:classStatisticValue >> [ a ot:ClassStatisticValue ; >> ot:classStatistic default:areaUnderRocCurve ; >> ot:value "0.77"^^:xsd:float ; >> ] ; >> ot:classStatisticValue >> [ a ot:ClassStatisticValue ; >> ot:classStatistic default:numTruePositives ; >> ot:value "123"^^:xsd:int ; >> ] ; >> ot:classStatisticValue >> ... >> >> ot:classStatisticEntry >> [ a ot:classStatisticEntry ; >> ot:classValue "intactive"^^String ; >> ... >> >> ot:classStatisticEntry >> [ a ot:classStatisticEntry ; >> ot:classValue "active"^^String ; >> ... >> ]] >> >> Here is the old classification statistics object (I renamed it from >> ClassifcationInformation to ClassificationStatistics): >> http://www.opentox.org/data/documents/development/RDF%20files/Validation/#-ot-classificationinfo-rdf >> >> Any comments, corrections before I add that to the opentox.owl? >> >> Best regards, >> Martin >> >> >> >> >> >> _______________________________________________ >> Development mailing list >> Development at opentox.org >> http://www.opentox.org/mailman/listinfo/development >> >> >> >> >> > > > > -- > Dipl-Inf. Martin Gütlein > Phone: > +49 (0)761 203 8442 (office) > +49 (0)177 623 9499 (mobile) > Email: > guetlein at informatik.uni-freiburg.de > -- Dipl-Inf. Martin Gütlein Phone: +49 (0)761 203 8442 (office) +49 (0)177 623 9499 (mobile) Email: guetlein at informatik.uni-freiburg.de
- Previous message: [OTDev] Validation: classification statistics for non-binary class values
- Next message: [OTDev] Java Examples
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Development mailing list