private static void ClearDatabase(EnronContext context, bool clearEmailAddresses, bool clearEmailEntities, bool clearEmailDestinations)
{
Console.WriteLine("Preparing the clearing query...");
if (clearEmailAddresses)
{
context.EmailAddresses.RemoveRange(context.EmailAddresses);
}
if (clearEmailEntities)
{
context.Emails.RemoveRange(context.Emails);
}
if (clearEmailDestinations)
{
context.DestinationEmails.RemoveRange(context.DestinationEmails);
}
Console.WriteLine("Removing elements from database...");
SaveChanges(context);
Console.WriteLine("Database cleared");
}
public static Graph <string, double, double> GetGraph(EnronContext context, IQueryable <EmailObject> emails)
{
var graph = new Graph <string, double, double>(directed: true);
var enronAddresses = new HashSet <string>();
var directedEmails = new Dictionary <(string, string), List <EmailObject> >();
// for each email that day
foreach (var triplet in
emails
.Join(
context.DestinationEmails,
eo => eo.Id,
de => de.EmailId,
(eo, de) => new { eo, de }
)
.Join(
context.EmailAddresses,
(pair) => pair.de.EmailAddressId,
ea => ea.Id,
(pair, ea) => new { pair.eo, ea } // email object and one of the senders
)
.Where(pair => pair.ea.BelongsToEnron)
.Join(
context.EmailAddresses,
(pair) => pair.eo.FromId,
ea => ea.Id,
(pair, ea) => new { pair.eo, pair.ea, eaFrom = ea }
)
.Where(triplet => triplet.eaFrom.BelongsToEnron)
)
{
var emailTo = triplet.ea;
var email = triplet.eo;
var fromAddress = triplet.eaFrom;
enronAddresses.Add(fromAddress.Address);
enronAddresses.Add(emailTo.Address);
var key = (fromAddress.Address, emailTo.Address);
if (directedEmails.ContainsKey(key))
{
// the same email could have been cloned to a different directory
if (directedEmails[key].FirstOrDefault(em => em.SendDate == email.SendDate) == default)
{
directedEmails[key].Add(email);
}
}
else
{
directedEmails.Add(key, new List <EmailObject>()
{
email
});
}
}
foreach (var enronAddress in enronAddresses)
{
graph.AddVertex(enronAddress);
}
foreach (var directedEmailKVP in directedEmails)
{
var edge = directedEmailKVP.Key;
if (!graph.Directed && graph.ContainsEdge(edge))
{
graph[edge] = Math.Min(graph[edge], directedEmailKVP.Value.Count);
}
else
{
graph.AddEdge(edge, directedEmailKVP.Value.Count);
}
}
return(graph);
}
static void Main(string[] args)
{
using var context = new EnronContext();
Console.WriteLine($"{context.Emails.Count()} emails in total.");
// var resultsFilePath = "./results.txt";
var topResultsFilePath = "./top_results.txt";
var topTopResultsFilePath = "./top_top_results.txt";
var matchingFeatureSelectors = new List <(string, Func <VertexPartialMatchingNode <string, double, double>, double>)>()
{
("Upper bound", matching => matching.BestUpperBound),
("Lower bound", matching => matching.BestLowerBound)
};
var edgeCostTypes = new List <(string, CostType, List <double>, List <double>)>()
{
// ("Unit cost Wojciechowski .5", CostType.UnitCost, new List<double>(){.5}, new List<double>(){.5}),
// ("Unit cost Wojciechowski (1.) .5", CostType.UnitCost, new List<double>(){1}, new List<double>(){.5}),
// ("Unit cost Wojciechowski .5 Riesen Bunke 1.", CostType.UnitCost, new List<double>(){.5, 1}, new List<double>(){1, .5}),
// ("Unit cost Riesen Bunke 1.", CostType.UnitCost, new List<double>(){1}, new List<double>(){1}),
("Absolute value Wojciechowski .5", CostType.AbsoluteValue, new List <double>()
{
.5
}, new List <double>()
{
.5
}),
// ("Absolute value Wojciechowski (1.) .5", CostType.AbsoluteValue, new List<double>(){1}, new List<double>(){.5}),
// ("Absolute value Wojciechowski .5 Riesen Bunke 1.", CostType.AbsoluteValue, new List<double>(){.5, 1}, new List<double>(){1, .5}),
("Absolute value Riesen Bunke 1.", CostType.AbsoluteValue, new List <double>()
{
1
}, new List <double>()
{
1
}),
// ("Absolute value bounded Wojciechowski .5", CostType.AbsoluteValueBounded, new List<double>(){.5}, new List<double>(){.5}),
// ("Absolute value bounded Wojciechowski (1.) .5", CostType.AbsoluteValueBounded, new List<double>(){1}, new List<double>(){.5}),
// ("Absolute value bounded Wojciechowski .5 Riesen Bunke 1.", CostType.AbsoluteValueBounded, new List<double>(){.5, 1}, new List<double>(){1, .5}),
// ("Absolute value bounded Riesen Bunke 1.", CostType.AbsoluteValueBounded, new List<double>(){1}, new List<double>(){1}),
};
var results = new Dictionary <(int vertexUpperBound, int k, string distanceScorerName, string edgeCostType, string matchingFeatureSelectorName), (double testAccuracy, double validationAccuracy)>();
var topResults = new Dictionary <(int vertexUpperBound, int k, string distanceScorerName, string edgeCostType, string matchingFeatureSelectorName), (double testAccuracy, double validationAccuracy)>();
for (int iteration = 0; iteration < 10; iteration++)
{
for (int vertexUpperBound = 3; vertexUpperBound < 8; vertexUpperBound++)
{
var dataset = GenerateDataSet(
context,
trainingProportion: 8,
validatingProportion: 1,
testingProportion: 1,
vertexUpperBound: vertexUpperBound,
randomSeed: iteration
);
foreach (var(matchingFeatureSelectorName, matchingFeatureSelector) in matchingFeatureSelectors)
{
foreach (var(edgeCostTypeName, edgeCostType, aCollection, bCollection) in edgeCostTypes)
{
var matchingParameters = GraphMatchingParameters <string, double, double> .DoubleCostComposer(
CostType.UnitCost,
edgeCostType
);
matchingParameters.aCollection = aCollection;
matchingParameters.bCollection = bCollection;
// determine closest neighbours for each test and validation graph
var testMatchingClassPairsList = new List <(List <(VertexPartialMatchingNode <string, double, double>, bool)>, bool)>();
var validationMatchingClassPairsList = new List <(List <(VertexPartialMatchingNode <string, double, double>, bool)>, bool)>();
foreach (var(pair, i) in dataset.testSet.Zip(Enumerable.Range(0, int.MaxValue)))
{
var pairToAdd = (
KNNClassifier.KNNClassifier.FindClosest(pair.Item1, dataset.trainingSet, matchingParameters, matchingFeatureSelector),
pair.Item2
);
testMatchingClassPairsList.Add(pairToAdd);
// System.Console.WriteLine($"Test set: computed distance: {i * 100d / dataset.testSet.Count:f2}%.");
}
foreach (var(pair, i) in dataset.validationSet.Zip(Enumerable.Range(0, int.MaxValue)))
{
var pairToAdd = (
KNNClassifier.KNNClassifier.FindClosest(pair.Item1, dataset.trainingSet, matchingParameters, matchingFeatureSelector),
pair.Item2
);
validationMatchingClassPairsList.Add(pairToAdd);
// System.Console.WriteLine($"Validation set: computed distance: {i * 100d / dataset.validationSet.Count:f2}%.");
}
Func <int, Func <int, VertexPartialMatchingNode <string, double, double>, double>, List <(List <(VertexPartialMatchingNode <string, double, double>, bool)>, bool)>, double> getAccuracy = (k, distanceScorer, matchingClassPairsList) =>
{
var truePositives = 0;
var falsePositives = 0;
var trueNegatives = 0;
var falseNegatives = 0;
foreach (var(matchingClassPairs, testGraphLabel) in matchingClassPairsList)
{
var classificationResult = KNNClassifier.KNNClassifier.Classify <string, double, double, bool>(
matchingClassPairs,
distanceScorer,
k: k
);
var result = (expected : testGraphLabel, received : classificationResult.graphClass);
if (result.expected)
{
if (result.received)
{
truePositives += 1;
}
else
{
falseNegatives += 1;
}
}
else
{
if (result.received)
{
falsePositives += 1;
}
else
{
trueNegatives += 1;
}
}
}
return((truePositives + trueNegatives) * 1d / matchingClassPairsList.Count);
};
var ks = new int[]
{
1,
2,
3,
4,
5,
6,
7,
8,
9,
10,
-1
};
var distanceScorers = new (string, Func <int, VertexPartialMatchingNode <string, double, double>, double>)[]
