2.3 Rule-Based Graph

Attributes can depend on each other. These dependencies build a hierarchy, which is depicted by the Rule-Based Graph.

Figure R-12: ‘Tender’ depends on ‘Reference’ and ‘Finance’, ‘Finance’ depends on ‘Warranty’ and ‘Price’.

The Rule-Based Reasoning produces new case features based on rules and the case features of the input attributes.

2.3.1 View

Each node represents an attribute from the «Attributes» pane. The decision attribute stands on the top of the graph. The factor attributes are connected from below to the nodes of their dependent attributes. This edge represents that the attribute depends on its factors. Input attributes are the leafs of the graph, they have no factors.

Even if more decision attributes exist, only one graph is shown, which is for the decision attribute (R-2.1.2.9) representing the conclusion of the knowledge base. The other ones are placed without their factor nodes below the graph.

2.3.1.1 Customize

Choose «Customize» from the menu «View».

Figure R-13: Customize «Rule-Based Graph» pane

This dialog controls the appearance of the «Rule-Based Graph» pane. Following switches are supported:

• Horizontal – Graph is mirrored that the root node is on the left of the window, and the factors are on the right of it.
• Condensed – Nodes with more than two leafs can take a wide area of the screen. Check this checkbox to arrange such leafs into two rows to reduce total width of the graph (in horizontal view it will arrange the leafs into two columns to reduce the hight).
• Additional nodes – Select it, whether all other attributes, only the other decision attributes, or no other attributes to be sown below the graph.

Optionally some additional information can be placed in the nodes of the attributes. Following «Node Legends» are supported:

• None – No node legends.
• Values – Values of the attribute are listed in the nodes.
• Cases – Selected cases features are listed in the nodes.
• Cases with Names – Selected cases names and features are listed in the nodes.

Use the list box on the right to select cases to show for the last two legend mode. Additionally the appearance of node legends can be limited by the following ways:

• For a Level Only – Check this checkbox to show node legends only at the current node and at the nodes on the same level. Select a node on another level and choose «Show legends on this level» from the context menu to change the current level to show legends on.
• For Input Attributes Only – Check this flag to show node legends only on leaf nodes.

If both of these checkboxes are set, legends are shown only on leaf nodes with the current level (‘and’ relation).

2.3.1.2 Other Decision Attributes

A knowledge base can have sometimes (actually without attribute dependencies always) more than one decision attributes. To see the graphs of the decision attributes other than the current one, ensure to display them as Additional nodes, than set it to Decision Attribute (R-2.1.2.9).

2.3.1.3 Hide Above

To see only a certain branch of a huge Rule-Based Graph select the node to show it as root and choose «Hide Above» from the context menu. The parent and its factor nodes are hidden. Only a short edge indicates that the node is not the real root.

To see the full graph again, choose «Hide Above» once more, or set the original root node shown below the graph to the Decision Attribute (R-2.1.2.9).

2.3.1.3 Hide Below

To hide factor nodes of a node choose «Hide Below» from the context menu. The factor branches are hidden only some short edges indicate that the node is not a leaf.

To see the factors again, choose «Hide Below» once more.

2.3.2 Edit

2.3.2.1 Connect

To connect a factor node to another attribute node, drag the node to connect from below the graph to the node to connect to. A dynamic edge notifies what connection is about to establish by dropping the node.

Advanced: The rule domain of the dependent attribute obtains an additional dimension by connecting the new factor. Rule domain is expanded by inserting full ranges of the new factor, thus the rule set remains independent of it.

2.3.2.2 Arrange

To change the order of the factors drag one of the factor nodes holding down the «Shift» key. The factor will neither be pruned nor connected to another node. The order of factor nodes defines the order of factors on the «Rules» pane.

2.3.2.3 Prune

To disconnect a factor node, first select it, then choose «Prune» from the «Edit» or context menu. It is removed from its dependent node and appears below the graph.

Advanced: The rule domain of the dependent attribute loses a dimension, but no rules are deleted. So the rule set becomes less distinctive, thus multiply covered domains will arise.

2.3.3 Examine

2.3.3.1 Explain

Set node legends (R-2.3.1.1) for «Cases with Names» and select a dependent node. Choose «Explain» from the context menu and select a case to explain its reasoned feature from the submenu. The rule, by which the case feature is resulted, is selected on the «Rules» pane.

2.3.4 Manage

2.3.4.1 Deductive Reasoning

Set node legends (R-2.3.1.1) for «Cases with Names» and choose «Deductive Reasoning» from the menu «Knowledge Management». The case features are overwritten by reasoned values.

Advanced: The behaviour of special case features during the reasoning are detailed in Rules – Rule List & Rule Processing (R-2.4.1.2).

2.3.4.2 Accept Advice

See Cases – Accept Advice (R-2.2.4.2).

Cases – Deductive Reasoning (R-2.2.4.1) and Cases – Accept Advice (R-2.2.4.2) are applicable with the difference that the result can be shown in the node legends.