VehicleIDPO Tutorial kbl 120304 v7 - George Mason Universitymason.gmu.edu/~klaskey/GraphicalModels/VehicleIDPO Tutorial_kbl_120304... · Vehicle’IDPO’Tutorial’ ’ 4’ Spring’2012’

Vehicle ID PO Tutorial Spring 2012 1

PROBABILISTIC ONTOLOGY DEVELOPMENT OF THE VEHICLE ID PROBLEM

Spring 2012

This tutorial will generally follow a Probabilistic Ontology Development Methodology (PODM). The model will be used to infer vehicle type from MTI, imaging sensor reports, weather reports and GIS reports.

Background:

1. Vehicles exist in an area of interest with some prior probability. A MEBN model is used to determine the type of vehicle that is approaching.

2. A military commander receives input from some or all of his available sensors, including MTI, Imaging sensors, and GIS reports. A current weather forecast may be available.

3. The class diagram (Figure 1) shows relationships of features that immediately affect the likelihood of a determination of vehicle type. There are three basic types of entity: contacts (objects such as vehicles that we wish to identify), reports (evidence about contacts derived from sensors), and regions (places where contacts can be located). The diagram shows associations between these entity types: vehicles can be located in regions (hasLocation association) and reports provide evidence about contacts (hasReportedEntity association). There are also entities to represent the terrain type of a region, the vehicle type of a contact, and the content of a report. These entity types and associations are also shown on the diagram: hasTerrainType connects a region with its associated terrain type; hasVehicleType connects a vehicle contact with the type of vehicle; and hasImagingDetection connects an image report with its content.

Figure 1 – Vehicle ID Class Diagram


4. Figure 2 is a Bayesian network (BN) model created in Netica for a more extensive vehicle identification problem. This BN model can be used to test the probabilistic ontology created in Multi-‐Entity Bayesian Network (MEBN) for logic errors. The Conditional Probability Tables (CPT) populated for this model will be used to populate the Local Probability Distributions (LPD) in the MEBN model.

5. We will be creating a probabilistic ontology (PO) for a part of this model. The part covered by the tutorial is outlined in red. We will note how to extend the PO to include the full set of RVs shown in Figure 2.

Figure 2 -‐ Vehicle ID Bayesian Network

Ontology Development:

1. Create and Save the Ontology in UnBBayes i. Start UnBBayes ii. Select Multi-entity Bayesian Network button (2nd from left) iii. Select Save net button (4th from left) iv. Navigate to the directory of your choice and type VehicleIDPO.ubf and Save v. In the I/O extension conflict pane, select UnBBayes File with PR-OWL 2.0

and OK vi. At the OK dialogue click OK

Close the file to reset the ontology


2. Create Ontology Classes i. On the Toolbar, click the Open net button (3rd from left) to open the MEBN ii. In the Open dialogue, navigate to the location of your saved VehicleIDPO.ubf file,

highlight the file, and select Open iii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK iv. On the OWL2 Entities tab, select the Classes tab and observe the superclass Thing at

the top of the Class hierarchy pane v. Click on the superclass Thing, and select the Add subclass button vi. Enter the first subclass, VehicleType, and select OK vii. Select the VehicleType class and then the Add sibling class button viii. Enter the next subclass, Terrain, and select OK ix. Repeat the procedure for the following sibling classes:

• Contact

• Report

• Region

• ImagingDetection

When you finish, the Class hierarchy pane should look like the following (Figure 3)

x. Figure 3 – Class Hierarchy

xi. On the Toolbar, click the Save net button (4th from left) xii. In the Save dialogue highlight VehicleIDPO.ubf in the dialogue, change to

VehicleIDPO_a.ubf, and select Save xiii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK xiv. Observe the File saved dialogue and click OK

Note: We will be saving the ontology and probabilistic ontology after each major step in the process. It is recommended that these be saved using sequential letters to prevent total loss of work in case of a catastrophic software glitch.


3. Create Object Properties of Classes i. Select the Object Properties tab and then click the topObjectProperty in the

Object property hierarchy pane ii. Select the Add sub property button, enter hasVehicleType, and OK iii. Select the hasVehicleType data property and then the Add sibling property

button iv. Enter the next data property, hasTerrainType, and OK v. Repeat the procedure for the following object properties:

• hasReportedEntity

• hasLocation

• hasImagingDetection

When you finish, the Object property hierarchy pane should resemble the following (Figure 4)

Figure 4 – Object Property Hierarchy

vi. On the Toolbar, click the Save net button (4th from left) vii. In the Save dialogue highlight VehicleIDPO_a.ubf in the dialogue, change to

VehicleIDPO_b.ubf, and select Save


viii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the dropdown menu and click OK

ix. Observe the File saved dialogue and click OK

4. Assign Domain and Range to Each Object Property i. On the Object Properties tab, select hasImagingDetection and select Domains in

the Description pane ii. On the Class hierarchy tab click Thing, highlight Report, and click OK iii. On the Object Properties tab, select Ranges in the Description pane iv. On the Class hierarchy tab click Thing, highlight ImagingDetection, and click OK v. Repeat this process for each of the Object properties in Table 1 vi. On the Toolbar, click the Save net button (4th from left) vii. In the Save dialogue highlight VehicleIDPO_b.ubf in the dialogue, change to

VehicleIDPO_c.ubf, and select Save viii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK ix. Observe the File saved dialogue and click OK

Table 1 – Object Property Domain and Range

Property Domain Range hasImagingDetection Report ImagingDetection hasLocation Contact Region hasReportedEntity Report Contact hasTerrainType Region Terrain hasVehicleType Contact VehicleType

5. Create Data Properties of Classes with Domain and Range

i. Select the Data Properties tab and then click the topDataProperty in the Object property hierarchy pane

ii. Select the Add sub property button, enter hasClearWeather, and OK iii. On the Data Properties tab, select hasClearWeather and select Domains in the

Description pane iv. On the Class hierarchy tab click Thing, highlight Region, and click OK v. Now select Ranges in the Description pane vi. On the Built in datatypes tab, highlight Boolean, and select OK vii. On the Toolbar, click the Save net button (4th from left) viii. In the Save dialogue highlight VehicleIDPO_c.ubf in the dialogue, change to

VehicleIDPO_d.ubf, and select Save ix. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK x. Observe the File saved dialogue and click OK


6. Assign Individuals to Classes i. On the Individuals tab expand the superclass Thing in the Class hierarchy pane ii. Select ImagingDetection and observe the empty Members list: in the Members list

pane iii. Select the Add individual button, type Image_NoVehicle, and OK iv. Repeat the process for the other individuals as specified in Table 2 v. From the File menu, select Save and OK

Table 2 – Individuals

Class Individual(s) ImagingDetection Image_NoVehicle

Image_Tracked Image_Wheeled

Report RepImg_1 RepImg_2 RepImg_3

Contact Track_1 Track_2 Track_3

Region Region_1 Region_2

Terrain Terr_OffRoad Terr_OnRoad Terr_VeryRough

VehicleType Obj_NonVehicle Obj_Tracked Obj_Wheeled

7. Assign Evidence (Assertions) to Individuals

i. On the Individuals tab expand the superclass Thing in the Class hierarchy pane and click Report

ii. In the Members list pane, click RepImg_1 iii. In the Property assertions pane, select Object property assertions iv. Click the topObjectProperty to expand and select hasReportedEntity v. Scroll through the list of Individuals in the right column, select Track_1, and OK vi. Repeat the process for the other assertions as specified in Table 3. vii. Click Region in the Class hierarchy pane viii. In the Members list pane, click Region_1 ix. In the Property assertions pane, select Data property assertions x. Click the topDataProperty to expand and select hasClearWeather xi. Select boolean from the type dropdown menu, type true in the Value pane, and select

OK. xii. On the Toolbar, click the Save net button (4th from left) xiii. In the Save dialogue highlight VehicleIDPO_d.ubf in the dialogue, change to

VehicleIDPO_e.ubf, and select Save


xiv. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the dropdown menu and click OK

xv. Observe the File saved dialogue and click OK

Table 3 – Evidence (Assertions)

Class Individual(s) Property Assertion Report RepImg_1 hasReportedEntity Track_1

hasImagingDetection Image_Tracked RepImg_2 hasReportedEntity Track_2

hasImagingDetection Image_Wheeled RepImg_3 hasReportedEntity Track_3

hasImagingDetection Image_Wheeled Contact Track_1 hasLocation Region_1

Track_2 hasLocation Region_2 Track_3 hasLocation Region_2

Region Region_1 hasTerrainType Terr_OnRoad Region_2

8. Evaluate Logic with the Protégé Reasoner i. From the Reasoner menu, select Synchronize reasoner

§ Observe Reasoner progress bar ii. Select Classes tab and Class hierarchy (inferred) sub-‐tab

§ Click Thing and observe the subclasses § There should be no RED values

iii. Close the MEBN by selecting the X in the upper right corner

Congratulations! You have successfully created the Vehicle ID Ontology.

Probabilistic Ontology Development:

9. Create Initial MEBN Fragments (MFrags) i. On the Toolbar, click the Open net button (3rd from left) to open the MEBN ii. In the Open dialogue, navigate to the location of your saved VehicleIDPO_e.ubf file,

highlight the file, and select Open iii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK iv. In the MTheory tab, observe 3 panes: the MTheory tree pane at the top left, the Description

pane at the bottom left, and the MFrag workspace pane on the right. If you see only two panes, drag the lower right-‐hand corner of the window to enlarge the window until three panes are visible. Click the Insert MFrag button (Flag with an ‘F’)

v. Highlight DMFrag1 in the text box at the center of the workspace (Figure 5), type VehicleType_MF, and RETURN. Observe the MFrag name change in the text box, the top of the MFrag window, and the MTheory tree


Figure 5 – MFrag Creation

vi. Create additional MFrags. Be sure to select RETURN for each after entering the name of the MFrag.

§ ImagingReport_MF § Reference_MF

vii. Now change the name of the MTheory. To do this, click the Edit the MTheory button just above the Insert MFrag button. Highlight MEBN in the text box at the center of the workspace, type VehicleIDPO, and RETURN. Observe the MTheory name change in the text box and the MTheory tree

viii. When you have finished, the MTheory Tree should appear similar to Figure 6

Figure 6 – MTheory Tree

ix. On the Toolbar, click the Save net button (4th from left) x. In the Save dialogue highlight VehicleIDPO_e.ubf in the dialogue, change to

VehicleIDPO_f.ubf, and select Save


xi. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the dropdown menu and click OK

xii. Observe the File saved dialogue and click OK xiii. Close UnBBayes to update the owl files

10. Populate the initial set of MFrags

i. Start UnBBayes and from the Toolbar, select the Open net button, highlight VehicleIDPO_f.ubf, and click Open

ii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the dropdown menu and click OK

iii. In the Property2Node tab, select the Show MTheory tree button, and double-‐click VehicleType_MF MFrag

iv. Select the Show OWL Properties button, and highlight the hasVehicleType property v. Drag the hasVehicleType property into the workspace to create a resident node vi. Drag the hasTerrainType property into the workspace to create a resident node vii. Right click on the resident nodes and Resize to fit text viii. Create a resident node(s) for each of the other MFrags by dragging and dropping the

appropriate property as shown in Table 4. In each case, perform the following steps on the Property2Node tab:

a. Select the Show MTheory tree button b. Double-‐click the MFrag of interest (e.g. ImagingReport_MF) c. Click the Show OWL Properties tab d. Drag the appropriate OWL Property into the workspace (e.g. hasClearWeather) e. Right click on the resident node and Resize to fit text

Table 4 – Resident Nodes in MFrags

MFrag OWL Property / Resident Node VehicleType_MF hasTerrainType

hasVehicleType Reference_MF hasReportedEntity

hasLocation ImagingReport_MF hasClearWeather

hasImagingDetection

ix. On the Toolbar, click the Save net button (4th from left) x. In the Save dialogue highlight VehicleIDPO_f.ubf in the dialogue, change to

VehicleIDPO_g.ubf, and select Save e xi. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from

the dropdown menu and click OK xii. Observe the File saved dialogue and click OK xiii. Close UnBBayes to update the owl files


11. Assign States to resident nodes i. Start UnBBayes and from the Toolbar, select the Open net button, highlight

VehicleIDPO_g.ubf, and click Open ii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK iii. Select the Show Mtheory tree button of the MTheory tab iv. Double-‐click the VehicleType_MF MFrag in the MTheory Tree pane and click the

hasVehicleType() resident node v. Select the Insert category states button and the Add a pre-defined state

button vi. Select Obj_NonVehicle from the Existent States pane and the plus sign (+) vii. Observe the state (Obj_NonVehicle) in the Resident Node pane viii. Select the Obj_Wheeled and Obj_Tracked states in a similar fashion and observe their

presence in the Resident Node pane ix. Assign category states to other appropriate nodes according to the first five rows of Table 5

by completing the following steps: a. Select the Show MTheory tree button b. Double-‐click the appropriate MFrag in the MTheory Tree pane and click the

resident node in the workspace c. Select the Insert category states button and Add a pre-‐defined state button d. Select the state from the Existent States pane and the plus sign(+) e. Observe the Resident Node pane and repeat for each state of the MFrag

Table 5 – MFrag States

MFrag Resident Node Entries Type VehicleType_MF hasVehicleType Obj_NonVehicle/Obj_Wheeled/

Obj_Tracked Category

hasTerrainType Terr_VeryRough/Terr_OffRoad/ Terr_OnRoad

Category

ImagingReport_MF hasImagingDetection Image_NoVehicle/Image_Wheeled/Image_Tracked

Category

hasClearWeather true / false / absurd Boolean Reference_MF hasReportedEntity Contact Object

hasLocation Region Object

x. Assign Boolean states to hasClearWeather node in ImagingReport_MF as shown in

in Table 5 by completing the following steps a. Select the ShowMTheory tree button b. Double-‐click ImagingReport_MF MFrag in the MTheory Tree pane and click the

hasClearWeather resident node in the workspace c. Select the Insert boolean entity states button on the Resident Node

pane d. Click the plus sign(+)


xi. Assign Object entity states to hasLocation and hasReportedEntity nodes in Reference_MF MFrag by completing the following steps

a. Select the ShowMTheory tree button b. Double-‐click Reference_MF MFrag in the MTheory Tree pane and click the

hasReportedEntity resident node in the workspace c. Select the Insert object entity states button on the Resident Node

pane d. Select Contact from the dropdown menu and the plus sign(+) e. Click the hasLocation resident node in the workspace f. Select the Insert object entity states button on the Resident Node

pane g. Select Region from the dropdown menu and the plus sign(+)

xii. On the Toolbar, click the Save net button (4th from left) xiii. In the Save dialogue highlight VehicleIDPO_g.ubf in the dialogue, change to

VehicleIDPO_h.ubf, and select Save xiv. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from

the dropdown menu and click OK xv. Observe the File saved dialogue and click OK xvi. Close UnBBayes to update the owl files

12. Assign Ordinary Variables to the MFrags

i. Start UnBBayes and from the Toolbar, select the Open net button, highlight VehicleIDPO_h.ubf, and click Open


iii. Select the MTheory tab and observe the newly created resident nodes (Figure 7)

Figure 7 – Resident Nodes of MFrags


iv. Double-‐click on the VehicleType_MF MFrag and observe the hasVehicleType() node in the workspace

v. Select the Insert Ordinary Variable button and click in the workspace above the hasVehicleType() node

vi. Highlight the variable name above the workspace, type the variable name ctc, and RETURN (note the lower-‐case variable name)

vii. From the dropdown menu above the workspace, select Contact viii. Right-‐click the new variable and Resize to fit text ix. Click on the hasVehicleType resident node and observe the Resident Node pane appear

to the left of the workspace x. Select the Edit Arguments button, () xi. Double-‐click the ctc(Contact) variable under the VehicleType_MF MFrag and observe

the variable name ctc has appeared in the Name: field and in the resident node in the workspace

xii. Right-‐click the resident node and Resize to fit text xiii. Create Ordinary Variables for the MFrags according to Table 6. In each case, perform the

following steps: a. Select the Show MTheory tree button from the MTheory tab b. Double-‐click the appropriate MFrag (e.g. VehicleType_MF) c. Select the Insert Ordinary Variable button and click in the workspace above

the resident node d. Highlight the variable name above the workspace, type the appropriate variable

name, and RETURN (remember to use lower-‐case variable names) e. From the dropdown menu above the workspace, select the appropriate class of

variable f. Right-‐click the ordinary variable and Resize to fit text g. Click on the appropriate resident node and observe the Resident Node pane appear

to the left of the workspace h. Select the Edit Arguments button, () i. Double-‐click the appropriate variable under the appropriate MFrag and observe the

variable name has appeared in the Name: field and in the resident node in the workspace

j. Right-‐click the resident node and Resize to fit text

Table 6 – Ordinary Variables

MFrag Variable Name Variable Type Resident Node VehicleType_MF ctc Contact hasVehicleType

rgn Region hasTerrainType Reference_MF ctc Contact hasLocation

rpt Report hasReportedEntity ImagingReport_MF rpt Report hasImagingDetection

rgn Region hasClearWeather ctc Contact


xiv. On the Toolbar, click the Save net button (4th from left) xv. In the Save dialogue highlight VehicleIDPO_h.ubf in the dialogue, change to

VehicleIDPO_i.ubf, and select Save xvi. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from

the dropdown menu and click OK xvii. Observe the File saved dialogue and click OK xviii. Close UnBBayes to update the owl files

13. Insert Input Nodes and Arcs for the MFrags

i. Start UnBBayes and from the Toolbar, select the Open net button, highlight VehicleIDPO_i.ubf, and click Open


iii. Select the MTheory tab and observe the resident nodes iv. Double-‐click the ImagingReport_MF MFrag and observe it appear in the workspace v. Select the Insert Input Node button and click in the workspace above the

hasImagingDetection resident node to place the input node vi. Select the Reset button to use the selection tool vii. Click on the newly created input node viii. Double click on the hasVehicleType node in the Resident List ix. Select ctc from the Contact_label dropdown list x. Right click on the hasVehicleType(ctc) input node and Resize to fit text xi. Select the Insert edge button (Arrow) and click/drag an edge from the input node to

the resident node xii. Click/drag an edge from the hasClearWeather resident node to the

hasImagingDetection resident node xiii. Click in the workspace

The ImagingDetection_MF MFrag should appear as in Figure 8

Figure 8 – ImagingDetection_MF MFrag with Input Node


xiv. Double-‐click the VehicleType_MF MFrag and observe it appear in the workspace xv. Select the Insert edge button (Arrow) and click/drag an edge from the

hasTerrainType(rgn) node to the hasVehicleType(ctc) node xvi. On the Toolbar, click the Save net button (4th from left) xvii. the Save dialogue highlight VehicleIDPO_i.ubf in the dialogue, change to

VehicleIDPO_j.ubf, and select Save xviii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from

the dropdown menu and click OK xix. Observe the File saved dialogue and click OK xx. Close UnBBayes to update the owl files

13. Populate the Local Distribution Tables (LDT)

i. Start UnBBayes and from the Toolbar, select the Open net button, highlight VehicleIDPO_j.ubf, and click Open


iii. Select the Show MTheory tree button of the MTheory tab iv. Double-‐click VehicleType_MF and observe it appear in the workspace v. Click the hasTerrainType(rgn) resident node in the workspace vi. Select the Edit table button in the Resident node pane and enter the following in the

text workspace:

The dialogue should now appear as shown in Figure 9

Figure 9 – Local Distribution Table for hasTerrainType(rgn) Resident node

[ Terr_OnRoad = .41, Terr_OffRoad = .31, Terr_VeryRough = .28 ]


Note: the buttons at the top of the LPD text workspace and the states and arguments in the States and Arguments pane can be used to expedite creation of LPDs. For example, the distribution shown in Figure 9 can be entered as follows. First, click on the default button at the top of the window. Then, fill in numbers for the word “formula” in each line of the expression that appears. Note also that typing state or variable names in LPD formulas is unnecessary. Double-‐clicking on a state name in the States window will cause that state name to be entered into the LPD text window. Double-‐clicking on an ordinary variable name in the Arguments window will cause that ordinary variable name to be entered into the LPD text window.

vii. Select the Compile button and observe the Success dialogue viii. Select the Save button and observe the Success dialogue ix. Select the Exit button x. Repeat the process for each of the other LDTs, using Figures 10-‐12 below. For each LDT,

complete the following steps: a. Select the Show MTheory tree button of the MTheory tab b. Double-‐click the appropriate MFrag and in the MTheory Tree pane and the

appropriate resident node in the workspace c. Select the Edit table button in the Resident node pane and duplicate the code

in the input pane d. Select the Compile button and observe the Success dialogue e. Select the Save button and observe the Success dialogue f. Select the Exit button

Figure 10 – LDT for hasVehicleType(ctc) Resident node

Figure 11 – LDT hasClearWeather(rgn) Resident node

if any rgn have ( hasTerrainType = Terr_OnRoad ) [ Obj_Tracked = .13, Obj_Wheeled = .22, Obj_NonVehicle = .65 ] else if any rgn have ( hasTerrainType = Terr_OffRoad ) [ Obj_Tracked = .10, Obj_Wheeled = .03, Obj_NonVehicle = .87 ] else [ Obj_Tracked = .07, Obj_Wheeled = 0, Obj_NonVehicle = .93 ]

[ true = .75, false = .25 ]



Figure 12 – LDT for hasImagingDet(rpt) Resident node

xi. On the Toolbar, click the Save net button (4th from left)

if any ctc have ( hasVehicleType = Obj_Tracked ) [ if any rgn have ( hasClearWeather = true ) [ Image_Tracked = .80, Image_Wheeled = .15, Image_NoVehicle = .05 ] else if any rgn have ( hasClearWeather = false ) [ Image_Tracked = .60, Image_Wheeled = .30, Image_NoVehicle = .10 ] else [ Image_Tracked = .15, Image_Wheeled = .15, Image_NoVehicle = .70 ] ] else if any ctc have ( hasVehicleType = Obj_Wheeled ) [ if any rgn have ( hasClearWeather = true ) [ Image_Tracked = .10, Image_Wheeled = .80, Image_NoVehicle = .10 ] else if any rgn have ( hasClearWeather = false ) [ Image_Tracked = .20, Image_Wheeled = .60, Image_NoVehicle = .20 ] else [ Image_Tracked = .15, Image_Wheeled = .15, Image_NoVehicle = .70 ] ] else if any ctc have ( hasVehicleType = Obj_NonVehicle ) [ if any rgn have ( hasClearWeather = true ) [ Image_Tracked = .05, Image_Wheeled = .05, Image_NoVehicle = .90 ] else [ Image_Tracked = .15, Image_Wheeled = .15, Image_NoVehicle = .70 ] ] else [ Image_Tracked = .15, Image_Wheeled = .15, Image_NoVehicle = .70 ]


xii. In the Save dialogue highlight VehicleIDPO_j.ubf in the dialogue, change to VehicleIDPO_k.ubf, and select Save

xiii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the dropdown menu and click OK

xiv. Observe the File saved dialogue and click OK xv. Close UnBBayes to update the owl files

14. Add Context Nodes

i. Start UnBBayes and from the Toolbar, select the Open net button, highlight VehicleIDPO_k.ubf, and click Open


iii. Double-‐click the VehicleType_MF MFrag in the MTheory Tree of the Mtheory Tab iv. Click the Insert Context Node button and drop a context node onto the workspace v. Click the Reset button vi. Click on the new context node and observe the Context Node pane to the left of the

workspace vii. Click formula in the Context Node pane and then the Equal sign (=) button viii. Click op_1 and then the Add ordinary variable button ix. Double-‐click rgn(Region) and observe the variable rgn in the new context node x. Now click op_2 and then the Add node button xi. Double-‐click the hasLocation resident node xii. Now double-‐click hasLocation in the formula editor of the Context Node pane xiii. Select ctc from the dropdown menu of the Contact_label pane, below xiv. Right click on the context node and Resize to fit text xv. Add context nodes to each of the MFrags in Table 9 using the following steps:

a. Select the Show MTheory tree button from the MTheory tab b. Double-‐click the appropriate MFrag (e.g. VehicleType_MF) c. Select the Insert Context Node button and click in the workspace below the

ordinary variables d. Click on the new context node, then click formula and the Equal sign (=) in

the Context Node pane e. Click op_1 and then the Add ordinary variable button f. Double-‐click the appropriate variable g. Click op_2 and then the Add node button h. Double-‐click the appropriate resident node i. Double-‐click the resident node in the formula editor and select the appropriate

variable from the dropdown menu j. Right click the new context node and Resize to fit text


Table 9 – Context Nodes

MFrag Ordinary Variable

Resident Node Resident Node Variable

VehicleType_MF rgn hasLocation ctc ImagingReport_MF rgn hasLocation ctc

ctc hasReportedEntity rpt

xvi. Observe the full MTheory by clicking on the See MTheory. The MFrags and nodes can be moved around on the page to improve the appearance of the result, as shown in Figure 13.

Figure 13 – Vehicle ID Probabilistic Ontology

xvii. On the Toolbar, click the Save net button (4th from left) xviii. In the Save dialogue highlight VehicleIDPO_k.ubf in the dialogue, change to

VehicleIDPO_l.ubf, and select Save xix. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from

the dropdown menu and click OK xx. Observe the File saved dialogue and click OK xxi. Close UnBBayes to update the owl files


15. Test the MEBN Theory with a simple network i. Start UnBBayes and from the Toolbar, select the Open net button, highlight

VehicleIDPO_l.ubf, and click Open ii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK iii. On the MTheory tab, select the Open MEBN Options button (8th from left) iv. On the Knowledge Base tab of the MEBN Options dialogue, check the Protégé

reasoners (for PR-OWL 2) button and Confirm v. Click the Load Knowledge Base button (2nd from left) and observe Loading

knowledge base… in the lower left corner of the window vi. When the Message dialogue appears, click OK vii. Select the Execute Query button (1st on left), highlight hasVehicleType and

Select viii. In the Query pane, select Track_1(Contact) from the Contact Label dropdown

menu and click Execute ix. Observe Generating SSBN… in the lower left corner of the window and wait for the

reasoner to finish x. Click the Propagate evidences button (3rd from left) and observe the created SSBN

shown in Figure 14

Figure 14 – Track_1 Simple Test Case

xi. Note how this SSBN matches the original Netica BN with similar evidence (Figure 15).


Figure 15 – Netica Model of Track_1 Simple Test Case

16. Test the MEBN Theory with an Advanced Case i. Start UnBBayes and from the Toolbar, select the Open net button, highlight

VehicleIDPO_l.ubf, and click Open ii. In the I/O Extension conflict dialogue, select UnBBayes File with PR-OWL 2.0 from the

dropdown menu and click OK iii. On the MTheory tab, select the Open MEBN Options button (8th from left) iv. On the Knowledge Base tab of the MEBN Options dialogue, check the Protégé

reasoners (for PR-OWL 2) button and Confirm v. Click the Load Knowledge Base button (2nd from left) and observe Loading

knowledge base… in the lower left corner of the window vi. Click OK in the Message dialogue vii. Select the Execute Query button (1st on left), highlight hasVehicleType and

Select viii. In the Query pane, select Track_3(Contact) from the Contact Label dropdown

menu and click Execute ix. Observe Generating SSBN… in the lower left corner of the window and wait for the

reasoner to finish x. Click the Propagate evidences button (3rd from left) and observe the created SSBN.

Drag nodes to match the SSBN shown in Figure 19


Figure 16 – Track_3 Advanced Test Case

xi. Note how this SSBN matches the original Netica BN with similar evidence (Figure 17).

Figure 17 – Netica Model of Track_3 Advanced Test Case

Congratulations! You have completed the Vehicle Identification Tutorial!

VehicleIDPO Tutorial kbl 120304 v7 - George Mason Universitymason.gmu.edu/~klaskey/GraphicalModels/VehicleIDPO Tutorial_kbl_120304... · Vehicle’IDPO’Tutorial’ ’ 4’ Spring’2012’

Documents