Node Editor — Properties
Context
- The Properties tab in the Node Editor allows you to define and modify a range of characteristics of a node.
Usage
- In the Node Editor, select the Properties tab.
- Within this tab, you can select and deselect, as well as edit node properties.
- Please see the following topics for detailed description of each property.
Thispanelallowseditingfourpropertiesofthenode.Theeditionofeachpropertyisalsoavailablefrom thenode's contextual menu.
- The color:allowsdisplayingacoloredtagonthenode,itscommentanditsmonitor. Checkingthe optionorclickingonthepreviewrectangleallowsdisplayingthecolorchooserdialogbox.Oncethe colorselected,itisdisplayedinsidetherectangleandcanbemodifiedagainbyclickingontherectangle. Toremovethecolor,youwillhavetounchecktheoption.ThePropagate color to the classes button displaysadialogboxallowingselectingthe**classes,associatedtothenode,onwhichwewantto propagatethecolor.Inthiscase,allthenodesbelongingtothesechosenclasseswilldisplaythe same selected color or any if there is no selection.
- The image:allowsdisplayinganimageinsteadofthenode'sdefaultrepresentation.Checkingthe optionorclickingonthepreviewrectangleallowsdisplayingthefilechooserdialogboxinorderto choosethewantedimage.Thedisplaysizeis3030.Ifanimageisbigger,itwillbereduced,ifitis smaller,itwillbecentered.Theimagewillbesavedinthenetwork'sfile.Toremovetheimage,you willhavetounchecktheoption.ThePropagateimagetotheclassesbuttondisplaysadialogbox allowingselectingthe**classes,associatedtothenode,onwhich wewanttopropagatetheimage.In thiscase,allthenodesbelongingtothesechosenclasseswilldisplaythesameselectedimageor any if there is no selection.
- The temporalindex:allowsassociatingatemporalindextothenode.Thisindexisapositiveornull integer.Itallowsindicatingatemporalorderbetweenthenodesthatistakenintoaccountbythe learningalgorithms.Anodewithatemporalindexgreaterthanthetemporalindexofanothernode cannotbeitsancestor.Toremovetheindex,youwillhavetounchecktheoption.ThePropagateindex to the classes buttondisplaysadialogboxallowingselectingthe**classes,associatedtothenode, onwhichwewanttopropagatetheindex.Inthiscase,allthenodesbelongingtothesechosenclasses will have the same index or any if there is no index.
- The cost:allowsassociatingacosttothenode.This costisareal number greater orequalto1representing thecostofanobservationuponanode.Thecostisusedintheadaptive questionnaire.Itispossible tomakeanode not observable byuncheckingtheoption.Inthiscase,thenodewillnotbeproposed in the adaptive questionnaire.It is also possible to use the "not observable" cost to ignore the values ofthenodethatarereadinadatabase(cf.Interactiveinference**,Batchlabeling**,Batchjointprobability),toindicatethenodetoupdate(cfInteractiveBayesianupdating),ortoindicatethe nodeforwhichonewantstocomputetheposteriorprobabilitydistributionforeachcasedescribedin adatabase(cf Batchinference**).ThePropagatecosttotheclassesbuttondisplaysadialogbox allowingselectingthe classes,associatedtothenode,onwhichwewanttopropagatethecost.In thiscase,allthenodesbelongingtothesechosenclasseswillhavethesamecostorwillbenotobservable if there is no cost.
- The state virtual number:allowsreplacingtherealnumberofstatesduringthelearningwiththe MDLscore.Thenode'sstatenumberhasanimportantimpactontheMDLscorecomputedduring thestructurallearning.Thisallowsinfluencingthenetwork'sstructuralcomplexitylocallytothenode. Moreanodehasstatesthelessithaschanceofhavinglinkedparentsduringthelearningandvice versa.DecreasingthisparameterdecreasestheMDLscoreofthenodeandviceversa.The Propagate state virtual number buttondisplaysadialogboxallowingselectingtheclasses,associatedtothenode,onwhichwewanttopropagatethestatevirtualnumber.Inthiscase,allthenodes belonging to these chosen classes will have the same state virtual number or any if there is no one.
- The local structural coefficient:thisparameteractslikethenetwork'sglobal structural coefficientbutispropertoeachnode.Itcanincreaseordecreasethestructuralcomplexityofthenetwork atthenode.ThisparameteractsonthewholeMDLscoreofthenodecontrarytothestatevirtual number.MoreanodehasahighMDLscorethelessithaschanceofhavinglinkedparentsduring learningandviceversa.Decreasingthisparameterdecreasesthenode'sMDLscoreandviceversa. The Propagate local structural coefficient buttondisplaysadialogboxallowingselectingthe**classes,associatedtothenode,onwhichwewanttopropagatethelocalstructuralcoefficient.In thiscase,allthenodesbelongingtothesechosenclasseswillhavethesamelocalstructuralcoefficient or any if there is no one.
- The exclusion:anodecanbeexcludedduringstructurallearning,meaningthelearningalgorithm won'taddanyarchavingthisnodeasextremity.Thisisparticularlyusefulifyouwishtolearnthe structureofanetworkonasubsetofnodes.ThePropagate exclusionbuttondisplaysadialogbox allowingselectingtheclasses,associatedtothenode,onwhichwewanttopropagatetheexclusion status.In this case, all the nodes belonging to these chosen classes will have been excluded or not.