Skip to content

Latest commit

 

History

History
104 lines (79 loc) · 4.89 KB

CausalGraphs.md

File metadata and controls

104 lines (79 loc) · 4.89 KB
Raw
Symbols and Functions > WolframModel and WolframModelEvolutionObject > Properties >

Causal Graphs

An event A causes an event B if there exists an expression (set element) created by A and destroyed by ** B**. If we then consider all such relationships between events, we create a "CausalGraph". In a causal graph, vertices correspond to events, and edges correspond to the set elements (aka spatial edges).

To make it even more explicit, we have another property, "ExpressionsEventsGraph". In this graph, there are two types of vertices corresponding to events and expressions, and edges correspond to a given expression being an input or an output of a given event.

For example, if we consider our simple arithmetic model {a_, b_} :> a + b starting from {3, 8, 8, 8, 2, 10, 0, 9, 7} we get an expressions-events graph which quite clearly describes what's going on:

In[] := WolframModel[<|"PatternRules" -> {a_, b_} :> a + b|>,
  {3, 8, 8, 8, 2, 10, 0, 9, 7}, Infinity]["ExpressionsEventsGraph",
 VertexLabels -> Placed[Automatic, After]]

The causal graph is very similar, it just has the expression-vertices contracted:

In[] := WolframModel[<|"PatternRules" -> {a_, b_} :> a + b|>,
 {3, 8, 8, 8, 2, 10, 0, 9, 7}, Infinity, "CausalGraph"]

Here is an example for a hypergraph model (admittedly considerably harder to understand). Multiedges correspond to situations where multiple set elements were both created and destroyed by the same pair of events:

In[] := WolframModel[{{1, 2, 3}, {4, 5, 6}, {1, 4}} ->
  {{3, 7, 8}, {9, 2, 10}, {11, 12, 5}, {13, 14, 6}, {7, 12}, {11,
    9}, {13, 10}, {14, 8}},
 {{1, 1, 1}, {1, 1, 1}, {1, 1}, {1, 1}, {1, 1}}, 20, "CausalGraph"]

"LayeredCausalGraph" generates the same graph but layers events generation-by-generation. For example, in our arithmetic causal graph, note how it's arranged differently from an example above:

In[] := WolframModel[<|"PatternRules" -> {a_, b_} :> a + b|>,
 {3, 8, 8, 8, 2, 10, 0, 9, 7}, Infinity, "LayeredCausalGraph"]

Note how slices through the expressions-events graph correspond to states returned by "StatesList". Pay attention to intersections of the slices with edges as well, as they correspond to unused expressions from previous generations that remain in the state:

In[] := With[{evolution =
   WolframModel[<|"PatternRules" -> {a_, b_} :> a + b|>,
    {3, 8, 8, 8, 2, 10, 0, 9, 7}, Infinity]},
 evolution["ExpressionsEventsGraph",
  VertexLabels -> Placed[Automatic, {After, Above}],
  Epilog -> {Red, Dotted,
    Table[Line[{{-10, k}, {10, k}}], {k, 0, 9, 2}]}]]

In[] := WolframModel[<|"PatternRules" -> {a_, b_} :> a + b|>,
 {3, 8, 8, 8, 2, 10, 0, 9, 7}, Infinity, "StatesList"]
Out[] = {{3, 8, 8, 8, 2, 10, 0, 9, 7}, {7, 11, 16, 12, 9}, {9, 18, 28}, {28,
  27}, {55}}

"ExpressionsEventsGraph" is particularly useful for multiway systems, as it allows one to immediately see multiway branching. For example, here the expression-vertex {2} has the out-degree of 2, which indicates it was used in two conflicting events, which indicates multiway branching:

In[] := WolframModel[{{1}, {1, 2}} -> {{2}}, {{1}, {1, 2}, {2, 3}, {2, 4}},
  Infinity,
  "EventSelectionFunction" -> None]["ExpressionsEventsGraph",
 VertexLabels -> Placed[Automatic, After]]

"CausalGraph", "LayeredCausalGraph" and "ExpressionsEventsGraph" properties all accept Graph options, as was demonstrated above with VertexLabels. Some options have special behavior for the Automatic value, i.e., VertexLabels -> Automatic in "ExpressionsEventsGraph" displays the contents of expressions, which are not the vertex names in that graph (as there can be multiple expressions with the same contents). VertexLabels -> "Index", on the other hand, displays the vertex indices of both expressions and events in the graph:

In[] := WolframModel[{{{x, y}, {x, z}} -> {{x, z}, {x, w}, {y, w}, {z, w}}},
  {{0, 0}, {0, 0}}, 2]["ExpressionsEventsGraph", VertexLabels -> "Index"]