/// <summary>
/// Algo 1: Find subgraph frequency (mappings help in memory)
/// </summary>
/// <param name="inputGraph"></param>
/// <param name="qGraph">The query graph to be searched for. If not available, we use expansion trees (MODA). Otherwise, we use Grochow's (Algo 2)</param>
/// <param name="subgraphSize"></param>
/// <param name="thresholdValue">Frequency value, above which we can comsider the subgraph a "frequent subgraph"</param>
/// <returns></returns>
public static Dictionary <QueryGraph, ICollection <Mapping> > Algorithm1(UndirectedGraph <int> inputGraph, QueryGraph qGraph, int subgraphSize = -1, int thresholdValue = 0)
{
// The enumeration module (Algo 3) needs the mappings generated from the previous run(s)
Dictionary <QueryGraph, ICollection <Mapping> > allMappings;
int numIterations = -1;
if (inputGraph.VertexCount < 121)
{
numIterations = inputGraph.VertexCount;
}
if (qGraph == null) // Use MODA's expansion tree
{
#region Use MODA's expansion tree
var treatedNodes = new HashSet <QueryGraph>();
allMappings = new Dictionary <QueryGraph, ICollection <Mapping> >(_builder.NumberOfQueryGraphs);
do
{
qGraph = GetNextNode()?.QueryGraph;
if (qGraph == null)
{
break;
}
ICollection <Mapping> mappings;
if (qGraph.IsTree(subgraphSize))
{
if (UseModifiedGrochow)
{
// Modified Mapping module - MODA and Grockow & Kellis
mappings = Algorithm2_Modified(qGraph, inputGraph, numIterations, false);
}
else
{
// Mapping module - MODA and Grockow & Kellis.
var inputGraphClone = inputGraph.Clone();
mappings = Algorithm2(qGraph, inputGraphClone, numIterations, false);
inputGraphClone.Clear();
inputGraphClone = null;
}
}
else
{
// Enumeration moodule - MODA
// This is part of Algo 3; but performance tweaks makes it more useful to get it here
var parentQueryGraph = GetParent(qGraph, _builder.ExpansionTree);
if (parentQueryGraph.IsTree(subgraphSize))
{
treatedNodes.Add(parentQueryGraph);
}
string file;
mappings = Algorithm3(allMappings, inputGraph, qGraph, _builder.ExpansionTree, parentQueryGraph, out file);
}
if (mappings != null && mappings.Count > thresholdValue)
{
qGraph.IsFrequentSubgraph = true;
}
// Save mappings. Do we need to save to disk? Maybe not!
allMappings.Add(qGraph, mappings);
// Do not call mappings.Clear()
mappings = null;
// Check for complete-ness; if complete, break
if (qGraph.IsComplete(subgraphSize))
{
qGraph = null;
break;
}
qGraph = null;
}while (true);
if (treatedNodes.Count > 0)
{
foreach (var mapping in allMappings)
{
if (mapping.Key.IsTree(subgraphSize) && !treatedNodes.Contains(mapping.Key))
{
mapping.Key.RemoveNonApplicableMappings(mapping.Value, inputGraph);
}
}
treatedNodes.Clear();
}
treatedNodes = null;
#endregion
}
else
{
ICollection <Mapping> mappings;
if (UseModifiedGrochow)
{
// Modified Mapping module - MODA and Grockow & Kellis
mappings = Algorithm2_Modified(qGraph, inputGraph, numIterations, true);
// mappings = ModaAlgorithm2Parallelized.Algorithm2_Modified(qGraph, inputGraph, numIterations);
}
else
{
mappings = Algorithm2(qGraph, inputGraph, numIterations, true);
}
qGraph.RemoveNonApplicableMappings(mappings, inputGraph);
allMappings = new Dictionary <QueryGraph, ICollection <Mapping> >(1)
{
{ qGraph, mappings }
};
// Do not call mappings.Clear()
mappings = null;
}
return(allMappings);
}
/// <summary>
/// Algo 1: Find subgraph frequency (mappings found are saved to disk to be retrieved later during Algo 3).
/// The value of the dictionary returned is in the form: $"{mappings.Count}#{qGraph.Label}.ser"
/// </summary>
/// <param name="inputGraph"></param>
/// <param name="qGraph">The query graph to be searched for. If not available, we use expansion trees (MODA). Otherwise, we use Grochow's (Algo 2)</param>
/// <param name="subgraphSize"></param>
/// <param name="thresholdValue">Frequency value, above which we can comsider the subgraph a "frequent subgraph"</param>
/// <returns></returns>
public static Dictionary <QueryGraph, string> Algorithm1_C(UndirectedGraph <int> inputGraph, QueryGraph qGraph, int subgraphSize, int thresholdValue)
{
// The enumeration module (Algo 3) needs the mappings generated from the previous run(s)
Dictionary <QueryGraph, string> allMappings;
int numIterations = -1;
if (inputGraph.VertexCount < 121)
{
numIterations = inputGraph.VertexCount;
}
if (qGraph == null) // Use MODA's expansion tree
{
#region Use MODA's expansion tree
var treatedNodes = new HashSet <QueryGraph>();
allMappings = new Dictionary <QueryGraph, string>(_builder.NumberOfQueryGraphs);
do
{
qGraph = GetNextNode()?.QueryGraph;
if (qGraph == null)
{
break;
}
ICollection <Mapping> mappings;
if (qGraph.EdgeCount == (subgraphSize - 1)) // i.e. if qGraph is a tree
{
if (UseModifiedGrochow)
{
// Modified Mapping module - MODA and Grockow & Kellis
mappings = Algorithm2_Modified(qGraph, inputGraph, numIterations, false);
}
else
{
var inputGraphClone = inputGraph.Clone();
mappings = Algorithm2(qGraph, inputGraphClone, numIterations, false);
inputGraphClone.Clear();
inputGraphClone = null;
}
// Because we're saving to file, we're better off doing this now
qGraph.RemoveNonApplicableMappings(mappings, inputGraph, false);
treatedNodes.Add(qGraph);
}
else
{
// Enumeration moodule - MODA
// This is part of Algo 3; but performance tweaks makes it more useful to get it here
var parentQueryGraph = GetParent(qGraph, _builder.ExpansionTree);
if (parentQueryGraph.EdgeCount == (subgraphSize - 1))
{
treatedNodes.Add(parentQueryGraph);
}
string _filename;
if (allMappings.TryGetValue(parentQueryGraph, out _filename))
{
string newFileName; // for parentQueryGraph
mappings = Algorithm3(null, inputGraph, qGraph, _builder.ExpansionTree, parentQueryGraph, out newFileName, _filename);
if (!string.IsNullOrWhiteSpace(newFileName))
{
// We change the _filename value in the dictionary since this means some of the mappings from parent fit the child
allMappings[parentQueryGraph] = newFileName;
}
}
else
{
mappings = new Mapping[0];
}
}
if (mappings.Count > thresholdValue)
{
qGraph.IsFrequentSubgraph = true;
}
// Save mappings.
var fileName = qGraph.WriteMappingsToFile(mappings);
if (mappings.Count > 0)
{
mappings.Clear();
}
allMappings.Add(qGraph, fileName);
// Check for complete-ness; if complete, break
if (qGraph.IsComplete(subgraphSize))
{
qGraph = null;
break;
}
qGraph = null;
}while (true);
#endregion
}
else
{
ICollection <Mapping> mappings;
if (UseModifiedGrochow)
{
// Modified Mapping module - MODA and Grockow & Kellis
mappings = Algorithm2_Modified(qGraph, inputGraph, numIterations, true);
}
else
{
mappings = Algorithm2(qGraph, inputGraph, numIterations, true);
}
qGraph.RemoveNonApplicableMappings(mappings, inputGraph);
var fileName = $"{mappings.Count}#{qGraph.Identifier}.ser";
System.IO.File.WriteAllText(fileName, Extensions.CompressString(Newtonsoft.Json.JsonConvert.SerializeObject(mappings)));
if (mappings.Count > 0)
{
mappings.Clear();
}
allMappings = new Dictionary <QueryGraph, string>(1)
{
{ qGraph, fileName }
};
}
return(allMappings);
}
/// <summary>
/// Mapping module; aka FindSubgraphInstances in Grochow & Kellis
/// </summary>
/// <param name="queryGraph">H</param>
/// <param name="inputGraphClone">G</param>
/// <param name="numberOfSamples">To be decided. If not set, we use the <paramref name="inputGraphClone"/> size / 3</param>
internal static ICollection <Mapping> Algorithm2(QueryGraph queryGraph, UndirectedGraph <int> inputGraphClone, int numberOfSamples, bool getInducedMappingsOnly)
{
if (numberOfSamples <= 0)
{
numberOfSamples = inputGraphClone.VertexCount / 3;
}
// Do we need this clone? Can't we just remove the node directly from the graph?
// We do need it.
var theMappings = new Dictionary <int[], List <Mapping> >(MappingNodesComparer);
var inputGraphDegSeq = inputGraphClone.GetNodesSortedByDegree(numberOfSamples);
var queryGraphVertices = queryGraph.Vertices.ToArray();
var queryGraphEdges = queryGraph.Edges.ToArray();
var subgraphSize = queryGraphVertices.Length;
var threadName = System.Threading.Thread.CurrentThread.ManagedThreadId;
Console.WriteLine("Thread {0}:\tCallingu Algo 2:\n", threadName);
for (int i = 0; i < inputGraphDegSeq.Count; i++)
{
var g = inputGraphDegSeq[i];
for (int j = 0; j < subgraphSize; j++)
{
var h = queryGraphVertices[j];
if (Utils.CanSupport(queryGraph, h, inputGraphClone, g))
{
#region Can Support
//Remember: f(h) = g, so h is Domain and g is Range
var f = new Dictionary <int, int>(1);
f[h] = g;
var mappings = Utils.IsomorphicExtension(f, queryGraph, queryGraphEdges, inputGraphClone, getInducedMappingsOnly);
f.Clear();
f = null;
if (mappings.Count > 0)
{
foreach (var item in mappings)
{
if (item.Value.Count > 1)
{
queryGraph.RemoveNonApplicableMappings(item.Value, inputGraphClone, getInducedMappingsOnly);
}
//Recall: f(h) = g
List <Mapping> maps;
if (theMappings.TryGetValue(item.Key, out maps))
{
maps.AddRange(item.Value);
}
else
{
theMappings[item.Key] = item.Value;
}
}
mappings.Clear();
}
mappings = null;
#endregion
}
}
//Remove g
inputGraphClone.RemoveVertex(g);
if (inputGraphClone.EdgeCount == 0)
{
break;
}
}
Array.Clear(queryGraphEdges, 0, queryGraphEdges.Length);
queryGraphEdges = null;
Array.Clear(queryGraphVertices, 0, subgraphSize);
queryGraphVertices = null;
inputGraphDegSeq.Clear();
inputGraphDegSeq = null;
var toReturn = GetSet(theMappings);
theMappings = null;
Console.WriteLine("Thread {0}:\tAlgorithm 2: All tasks completed. Number of mappings found: {1}.", threadName, toReturn.Count);
return(toReturn);
}