-
Notifications
You must be signed in to change notification settings - Fork 1
Graphia_Transforms
Calculates the shortest path between every node and assigns the longest path length found for that node. This is a measure of a node's position within the overall graph structure.
MCL Cluster
Finds discrete groups (clusters) of nodes based on a flow simulation model.
File:Before MCL.png|Unclustered graph File:After MCL.png|Colours represent clustered nodes
Centrality measurement for each node. This can be viewed as measure of a node's relative importance in the graph.
In some situations an attribute may contain some useful information, in amongst other data. The Attribute Synthesis transform allows a new attribute to be created based on the values containing in an existing one. Values are extracted from attributes by using regular expression captures.
File:Attribute Synthesis Before.png|The original attribute contains a lot of information, but for our purposes, we're only interested in the part before the first comma... File:Attribute Synthesis After.png|...so we can apply an Attribute Synthesis transform that extracts this section, and creates a new attribute named Major Food
Create an attribute whose value is the result of the specified condition. The attribute is always given the value True or False. These transforms are primarily useful for creating an attribute that can subsequently be used for visualising conditional information. In the following example, the yellow edges are those that have a correlation coefficient above a given threshold, whereas the red nodes are those whose names include a particular bit of text.
File:Boolean Attribute Before.png File:Boolean Attribute After.png
Create a new attribute by combining two other attributes.
File:Combine Attributes.png|Group (sub-class) and MCL Cluster have been combined to create a third attribute named Group Cluster
Remove nodes, edges or components based on a condition. These transforms are the most fundamental way to transform a graph. Note that the filters are expressed in two ways: Remove and Keep. These are functionally equivalent, but with the sense of the conditional reversed. The reason to present both is that sometimes it is easier to think about a filtering operation in terms of what is removed, whereas for other operations it makes more sense to think about what remains.
File:Edge Filter Before.png|A graph with all of its edges present... File:Edge Filter After2.png|...then removed using a conditional filter based on edge weight
Remove edges which match the specified condition while simultaneously merging the pairs of nodes that they previously joined. Often it is superfluous to display two or more nodes that share a similar characteristic; in this case we can use edge contraction to merge said nodes.
File:Edge Contraction Before.png|The orange edge will be contracted... File:Edge Contraction After.png|...leaving nodes A and B merged
Contract edges whose nodes share the same attribute value, such as cluster name.
File:Contract By Attribute Before.png|Before File:Contract By Attribute After.png|After
Reduce the complexity of the graph by removing a percentage of each node's edges, down to a specified minimum. The remaining edges are a representative sample of the original graph, but the result is potentially less expensive to display. This transform is useful in the case where k-NN cannot be used because the graph's edges do not have a rankable attribute such as a correlation coefficient.
Reduce the number of edges in the graph using the k-Nearest Neighbours algorithm. This is an approach where a node's edges are ranked in terms of strength, on the basis of some attribute's value. Thereafter all but the top k edges are removed. This has the effect of maintaining the strongest relationships in the graph, and thus the structure, while removing a large fraction of the edges. This can help in both visually revealing otherwise hidden structure, and by reducing the computational requirements for displaying the graph.
File:K-NN Before.png|Before File:K-NN After.png|After
Reduce the number of edges in the graph using a variation of the k-Nearest Neighbours algorithm, but instead of choosing the top k edges, choose a percentage of the highest ranking edges instead.
Nodes that are only connected by a single edge are removed, iteratively. This has the effect of removing all the branches (sometimes called spurs) from the graph, leaving only cycles.
File:Remove Branches Before.png|Before File:Remove Branches After.png|After
Like the Remove Branches transform, this removes singly connected nodes, but only for a specified number of iterations. This means the branches are pruned, but not removed completely.
Remove edges whose nodes have different attribute values. This has the effect of dividing a graph into components, where the attribute values of the specified attribute are equal.
File:Separate By Attribute Before.png|Before File:Separate By Attribute After.png|After
Every graph component has a sub-graph which is also a tree, meaning the graph has no cycles. This sub-tree is created by starting at an arbitrary node, and visiting each other node until all the nodes have been observed. This process "spans" the entire graph and thus the new graph is known as a spanning tree. The graph can be visited by observing a child node and their siblings first (known as Breadth First), or by visiting a child and the child's children in turn (known as Depth First). For a highly connected graph, the former will tend to create "bushy" structures, whereas the latter will result in "stringy" structures. This transform finds a spanning tree for each component.
File:Spanning Tree Before.png|Before File:Spanning Tree After.png|After