private IEnumerable <SuspiciousNode> SelectSuspiciousElementsWithMaskWithSmallGrowing(
IReadOnlyList <List <SuspiciousNode> > candidateNodesLists, BigInteger xor, BigInteger mask)
{
for (int i = 0; i < candidateNodesLists.Count; i++)
{
var vector = BigInteger.Zero;
var candidates = candidateNodesLists[i];
while (true)
{
var minDiffCount = int.MaxValue;
SuspiciousNode newNode = null;
foreach (var candidate in candidates)
{
var newVector = (vector | (candidate.Vector ^ xor)) & mask;
var diff = newVector ^ vector;
var diffCount = LearningExperimentUtil.CountBits(diff);
if (diffCount > 0 && minDiffCount > diffCount)
{
minDiffCount = diffCount;
vector = newVector;
newNode = candidate;
}
}
if (newNode != null)
{
yield return(newNode);
}
else
{
break;
}
}
}
}
private SuspiciousNode SelectMostDifferentElement(
IEnumerable <BigInteger> existings, IEnumerable <SuspiciousNode> candidates,
BigInteger mask)
{
if (!existings.Any())
{
return(candidates.FirstOrDefault());
}
var maxDiff = 0;
SuspiciousNode ret = null;
foreach (var candidate in candidates)
{
var vector = candidate.Vector & mask;
var diff = existings.Min(f => LearningExperimentUtil.CountBits((f & mask) ^ vector));
if (maxDiff < diff)
{
maxDiff = diff;
ret = candidate;
}
}
return(ret);
}
public ClassificationResult Classify(
int count, Classifier classifier, GroupCache groupCache,
EncodingResult encodingResult,
RevealedVectorSet trainingSet = null)
{
trainingSet = trainingSet ?? new RevealedVectorSet();
var correctlyAccepted = 0;
var correctlyRejected = 0;
var wronglyAccepted = 0;
var wronglyRejected = 0;
var correctlyRejectedInRejecting = 0;
var wronglyRejectedInRejecting = 0;
var rejectAccept = new List <List <SuspiciousNode> >();
var rejectReject = new List <List <SuspiciousNode> >();
var acceptAccept = new List <List <SuspiciousNode> >();
var acceptReject = new List <List <SuspiciousNode> >();
for (int i = 0; i < classifier.GroupCount; i++)
{
rejectAccept.Add(new List <SuspiciousNode>());
rejectReject.Add(new List <SuspiciousNode>());
acceptAccept.Add(new List <SuspiciousNode>());
acceptReject.Add(new List <SuspiciousNode>());
}
var wrongNodeCount = 0;
var wronglyRejectedFeatures = new List <BigInteger>();
var wronglyAcceptedFeatures = new List <BigInteger>();
foreach (var vectorAndGroupPath in encodingResult.IdealVectorSet.Accepted)
{
var vector = vectorAndGroupPath.Key;
var groupPath = vectorAndGroupPath.Value;
var groupIndex = groupCache.GetGroupIndex(vector);
var rejected = classifier.IsRejected(vector, groupIndex);
var accepted = classifier.IsAccepted(vector, groupIndex);
if (!trainingSet.Accepted.ContainsKey(vector) &&
!trainingSet.Rejected.ContainsKey(vector))
{
var target = new SuspiciousNode {
Vector = vector,
GroupKey = groupPath,
Used = false,
};
if (accepted)
{
if (!rejected)
{
acceptAccept[groupIndex].Add(target);
}
else
{
acceptReject[groupIndex].Add(target);
}
}
else
{
if (!rejected)
{
rejectAccept[groupIndex].Add(target);
}
else
{
rejectReject[groupIndex].Add(target);
}
}
}
if (!accepted)
{
wronglyRejected++;
wrongNodeCount += encodingResult.Vector2Count[vector];
wronglyRejectedFeatures.Add(vector);
}
else if (!rejected)
{
correctlyAccepted++;
}
else
{
wronglyRejectedInRejecting++;
wrongNodeCount += encodingResult.Vector2Count[vector];
wronglyRejectedFeatures.Add(vector);
}
}
foreach (var vectorAndGroupPath in encodingResult.IdealVectorSet.Rejected)
{
var vector = vectorAndGroupPath.Key;
var groupPath = vectorAndGroupPath.Value;
var groupIndex = groupCache.GetGroupIndex(vector);
var rejected = classifier.IsRejected(vector, groupIndex);
var accepted = classifier.IsAccepted(vector, groupIndex);
if (!trainingSet.Accepted.ContainsKey(vector) &&
!trainingSet.Rejected.ContainsKey(vector))
{
var target = new SuspiciousNode {
Vector = vector,
GroupKey = groupPath,
Used = false,
};
if (accepted)
{
if (!rejected)
{
acceptAccept[groupIndex].Add(target);
}
else
{
acceptReject[groupIndex].Add(target);
}
}
else
{
if (!rejected)
{
rejectAccept[groupIndex].Add(target);
}
else
{
rejectReject[groupIndex].Add(target);
}
}
}
if (!accepted)
{
correctlyRejected++;
}
else if (!rejected)
{
wronglyAccepted++;
wrongNodeCount += encodingResult.Vector2Count[vector];
wronglyAcceptedFeatures.Add(vector);
}
else
{
correctlyRejectedInRejecting++;
}
}
Console.WriteLine("done");
Console.WriteLine(
"WA: " + wronglyAccepted + ", WR: " + wronglyRejected + "/"
+ wronglyRejectedInRejecting + ", CA: "
+ correctlyAccepted + ", CR: " + correctlyRejected + "/"
+ correctlyRejectedInRejecting);
Console.WriteLine(
"TR: "
+ trainingSet.Count + ", AF: "
+ String.Join(", ", classifier.CountAcceptingFeatures())
+ ", RF: "
+ String.Join(", ", classifier.CountRejectingFeatures()));
var wrongFeatureCount = wronglyAccepted + wronglyRejected
+ wronglyRejectedInRejecting;
var selector = new SuspiciousNodeSelector(classifier.AcceptingFeatureBitMask,
classifier.RejectingFeatureBitMask);
var suspiciousNodes = selector.SelectSuspiciousNodes(count, classifier, acceptAccept,
acceptReject, rejectAccept, rejectReject);
return(new ClassificationResult(suspiciousNodes, wronglyAcceptedFeatures,
wronglyRejectedFeatures,
wrongFeatureCount, wrongNodeCount, encodingResult));
}
