/// <summary>
/// Enumeration module. NB: If either of <paramref name="allMappings"/> or <paramref name="fileName"/> is null, the other will not be.
/// </summary>
/// <param name="allMappings"></param>
/// <param name="inputGraph">G</param>
/// <param name="queryGraph">H</param>
/// <param name="expansionTree">T_k</param>
/// <param name="parentQueryGraph"></param>
/// <param name="fileName"></param>
/// <param name="parentGraphMappings">NB: This param is still used even outside this method is call. So, be careful how you set/clear its values.</param>
private static IList <Mapping> Algorithm3(Dictionary <QueryGraph, ICollection <Mapping> > allMappings, UndirectedGraph <int> inputGraph, QueryGraph queryGraph,
AdjacencyGraph <ExpansionTreeNode> expansionTree,
QueryGraph parentQueryGraph, out string newFileName, string fileName = null)
{
newFileName = null;
ICollection <Mapping> parentGraphMappings;
if (string.IsNullOrWhiteSpace(fileName))
{
if (!allMappings.TryGetValue(parentQueryGraph, out parentGraphMappings))
{
return(new Mapping[0]);
}
}
else
{
parentGraphMappings = parentQueryGraph.ReadMappingsFromFile(fileName);
}
if (parentGraphMappings.Count == 0)
{
return(new Mapping[0]);
}
var subgraphSize = queryGraph.VertexCount;
var parentQueryGraphEdges = new HashSet <Edge <int> >();
foreach (var edge in parentQueryGraph.Edges)
{
parentQueryGraphEdges.Add(edge);
}
var newEdge = GetEdgeDifference(queryGraph, parentQueryGraph, parentQueryGraphEdges);
parentQueryGraphEdges.Clear();
parentQueryGraphEdges = null;
// if it's NOT a valid edge
if (newEdge.Source == Utils.DefaultEdgeNodeVal)
{
return(new Mapping[0]);
}
var list = new List <Mapping>();
int oldCount = parentGraphMappings.Count, id = 0, queryGraphEdgeCount = queryGraph.EdgeCount;
var queryGraphEdges = queryGraph.Edges.ToArray();
var groupByGNodes = parentGraphMappings.GroupBy(x => x.Function.Values.ToArray(), MappingNodesComparer); //.ToDictionary(x => x.Key, x => x.ToArray(), MappingNodesComparer);
foreach (var set in groupByGNodes)
{
// function.value (= set of G nodes) are all same here. So build the subgraph here and pass it dowm
var subgraph = Utils.GetSubgraph(inputGraph, set.Key);
foreach (var item in set)
{
item.Id = id++;
// Remember, f(h) = g
// if (f(u), f(v)) ϵ G and meets the conditions, add to list
if (item.SubGraphEdgeCount == queryGraphEdgeCount)
{
var isMapping = Utils.IsMappingCorrect2(item.Function, subgraph, queryGraphEdges, true);
if (isMapping.IsCorrectMapping)
{
list.Add(item);
}
isMapping = null;
}
else if (item.SubGraphEdgeCount > queryGraphEdgeCount)
{
var newEdgeImage = item.GetImage(inputGraph, newEdge);
// if it's a valid edge...
if (newEdgeImage.Source != Utils.DefaultEdgeNodeVal &&
inputGraph.ContainsEdge(newEdgeImage.Source, newEdgeImage.Target))
{
list.Add(item);
}
}
}
subgraph = null;
}
Array.Clear(queryGraphEdges, 0, queryGraphEdges.Length);
queryGraphEdges = null;
var threadName = System.Threading.Thread.CurrentThread.ManagedThreadId;
// Remove mappings from the parent qGraph that are found in this qGraph
// This is because we're only interested in induced subgraphs
var theRest = parentGraphMappings.Except(list).ToList();
parentQueryGraph.RemoveNonApplicableMappings(theRest, inputGraph);
parentGraphMappings.Clear();
foreach (var item in theRest)
{
parentGraphMappings.Add(item);
}
theRest.Clear();
theRest = null;
// Now, remove duplicates
queryGraph.RemoveNonApplicableMappings(list, inputGraph);
if (!string.IsNullOrWhiteSpace(fileName) && oldCount > parentGraphMappings.Count)
{
// This means that some of the mappings from parent fit the current query graph
newFileName = parentQueryGraph.WriteMappingsToFile(parentGraphMappings);
try
{
System.IO.File.Delete(fileName);
}
catch { } // we can afford to let this fail
}
Console.WriteLine("Thread {0}:\tAlgorithm 3: All tasks completed. Number of mappings found: {1}.\n", threadName, list.Count);
return(list);
}
/// <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);
}