public List <WordPair> GenerateAllNaivePairs(BidirectionalGraph <Word, Edge <Word> > graph)
{
/*BidirectionalGraph<Word, Edge<Word>> graph = new BidirectionalGraph<Word, Edge<Word>>(false);
#region Prepare graph
* foreach (var item in g.Vertices)
* graph.AddVertex(item);
#endregion*/
List <WordPair> pairs = new List <WordPair>();
//List<KeyValuePair<Word, Word>> pairs = new List<KeyValuePair<Word, Word>>();
//List<KeyValuePair<List<KeyValuePair<Word, Word>>, float>> pairsProb = new List<KeyValuePair<List<KeyValuePair<Word, Word>>, float>>();
var uWords = graph.Vertices.Where(t => t.Language == Language.Uyghur);
var kWords = graph.Vertices.Where(t => t.Language == Language.Kazak);
var cWords = graph.Vertices.Where(t => t.Language == Language.Chinese);
foreach (var uWord in uWords)
{
float connectedUC = (float)graph.InDegree(uWord);
foreach (var kWord in kWords)
{
float connectedCK = (float)graph.InDegree(kWord);
WordPair pair = new WordPair(uWord, kWord);
pair.Prob = (connectedUC + connectedCK) / (2 * cWords.Count());
pairs.Add(pair);
}
}
/*var output = pairs.Select(t => string.Format("{0},{1}", t.Key, t.Value));
* System.IO.File.WriteAllLines(@"buffer\NaiveCombination.txt", output);
* System.Media.SoundPlayer simpleSound = new System.Media.SoundPlayer(@"c:\Windows\Media\Ring03.wav");
* simpleSound.Play();
* Debug.WriteLine("Generate All Naive pairs is done");
*/
return(pairs);
}
private WordPair createNewGreenEdges(Word uWord, Word kWord, BidirectionalGraph <Word, Edge <Word> > graph)
{
Cache1.Clear();
List <SPath> paths = new List <SPath>();
foreach (var item in graph.OutEdges(uWord))
{
SLink linkCU = new SLink(item.Target, uWord);
linkCU.Exists = graph.ContainsEdge(uWord, item.Target);
SLink linkCK = new SLink(item.Target, kWord);
linkCK.Exists = graph.ContainsEdge(item.Target, kWord);
SPath path = new SPath(linkCU, linkCK);
paths.Add(path);
Cache1.Add(item.Target.ID, true);
}
foreach (var item in graph.InEdges(kWord))
{
if (Cache1.ContainsKey(item.Source.ID))
{
continue;
}
SLink linkCK = new SLink(item.Source, kWord);
linkCK.Exists = graph.ContainsEdge(item.Source, kWord);
SLink linkCU = new SLink(item.Source, uWord);
linkCU.Exists = graph.ContainsEdge(uWord, item.Source);
SPath path = new SPath(linkCU, linkCK);
paths.Add(path);
}
//calculate probability
float pUK = 0;
float pKU = 0;
float probUK = 0;
float probKU = 0;
foreach (var item in paths)
{
if (!item.LinkCU.Exists || !item.LinkCK.Exists) //containing non-existance link
{
continue;
}
float PrCU = 0.0f;
float PrKC = 0.0f;
float PrCK = 0.0f;
float PrUC = 0.0f;
foreach (var downEdgeCU in graph.OutEdges(item.LinkCU.WordNonPivot)) //Loop through down-path from nonpivot1 to pivot
{
PrCU += 1.0f / LinkWeightCache[new SLink(downEdgeCU.Target, downEdgeCU.Source)]; //P(C|U) = P(C&U)/P(U)
}
foreach (var downEdgeCK in graph.OutEdges(item.LinkCK.WordPivot)) //Loop through down-path from pivot to nonpivot2
{
PrKC += 1.0f / LinkWeightCache[new SLink(downEdgeCK.Source, downEdgeCK.Target)]; //P(K|C) = P(K&C)/P(C)
}
foreach (var upEdgeCK in graph.InEdges(item.LinkCK.WordNonPivot)) //Loop through up-path from nonpivot2 to pivot
{
PrCK += 1.0f / LinkWeightCache[new SLink(upEdgeCK.Source, upEdgeCK.Target)]; //P(C|K) = P(C&K)/P(K)
}
foreach (var upEdgeCU in graph.InEdges(item.LinkCU.WordPivot)) //Loop through up-path from pivot to nonpivot1
{
PrUC += 1.0f / LinkWeightCache[new SLink(upEdgeCU.Target, upEdgeCU.Source)]; //P(U|C) = P(U&C)/P(C)
}
PrCU = 1.0f / PrCU;
PrKC = 1.0f / PrKC;
PrCK = 1.0f / PrCK;
PrUC = 1.0f / PrUC;
pUK += PrUC * PrCK;
pKU += PrKC * PrCU;
}
probUK = pUK * pKU;
WordPair pair = new WordPair(uWord, kWord);
pair.Paths = paths;
pair.Prob = (float)probUK;
//set link weights
foreach (var item in pair.Paths)
{
//CU
if (!item.LinkCU.Exists)
{
float value = 0;
if (LinkWeightCache.TryGetValue(item.LinkCU, out value))
{
if (pair.Prob > value)
{
item.LinkCU.Pr = LinkWeightCache[item.LinkCU] = pair.Prob;
}
else
{
item.LinkCU.Pr = value;
}
}
else
{
item.LinkCU.Pr = pair.Prob;
LinkWeightCache.Add(item.LinkCU, pair.Prob);
}
}
//CK
if (!item.LinkCK.Exists)
{
float value = 0;
if (LinkWeightCache.TryGetValue(item.LinkCK, out value))
{
if (pair.Prob > value)
{
LinkWeightCache[item.LinkCK] = item.LinkCK.Pr = pair.Prob;
}
else
{
item.LinkCK.Pr = value;
}
}
else
{
item.LinkCK.Pr = pair.Prob;
LinkWeightCache.Add(item.LinkCK, pair.Prob);
}
}
}
return(pair);
}
private List <WordPair> GeneratePossiblePairs(BidirectionalGraph <Word, Edge <Word> > g)
{
LinkCache = new Dictionary <int, SLink>();
LinkWeightCache = new Dictionary <SLink, float>();
graph = new BidirectionalGraph <Word, Edge <Word> >(false);
semiCompleteGraph = new BidirectionalGraph <Word, Edge <Word> >(false);
completeGraph = new BidirectionalGraph <Word, Edge <Word> >(false);
#region Prepare graph
foreach (var item in g.Vertices)
{
graph.AddVertex(item);
semiCompleteGraph.AddVertex(item);
completeGraph.AddVertex(item);
}
foreach (var item in g.Edges)
{
if (item.Source.Language == Console.Language.Chinese && item.Target.Language == Console.Language.Uyghur)
{
graph.AddEdge(new Edge <Word>(item.Target, item.Source));
semiCompleteGraph.AddEdge(new Edge <Word>(item.Target, item.Source));
completeGraph.AddEdge(new Edge <Word>(item.Target, item.Source));
}
else
{
graph.AddEdge(new Edge <Word>(item.Source, item.Target));
semiCompleteGraph.AddEdge(new Edge <Word>(item.Source, item.Target));
completeGraph.AddEdge(new Edge <Word>(item.Source, item.Target));
}
}
#endregion
List <WordPair> ooPairs = new List <WordPair>();
Dictionary <WordPair, bool> ooPairsDict = new Dictionary <WordPair, bool>();
var uWords = graph.Vertices.Where(t => t.Language == Language.Uyghur);
var kWords = graph.Vertices.Where(t => t.Language == Language.Kazak);
var cWords = graph.Vertices.Where(t => t.Language == Language.Chinese);
//int uWordCount = uWords.Count();
//int kWordCount = kWords.Count();
//int cWordCount = cWords.Count();
int u = 0, k = 0;
/*foreach (var uWord in uWords)
* {
* foreach (var edge1 in graph.OutEdges(uWord))
* {
* foreach (var edge2 in graph.OutEdges(edge1.Target))
* {
* Word kWord = edge2.Target;
* WordPair ooPair = new WordPair(uWord, kWord);
* ooPair = createNewEdges(uWord, kWord, graph);//, uWordCount, kWordCount);
* if (ooPairsDict.ContainsKey(ooPair))
* continue;
* else
* {
* ooPairsDict.Add(ooPair, true);
* ooPairs.Add(ooPair);
* }
* }
* }
*
* }*/
currentCycle = 1;
while (currentCycle <= symmetryCycle)
{
Debug.WriteLine(currentCycle);
if (currentCycle > 1)
{
//semiCompleteGraph.Clear();
semiCompleteGraph = completeGraph;
}
ooPairsDict.Clear();
ooPairs.Clear();
foreach (var uWord in uWords)
{
Word cWord;
Word kWord;
foreach (var edge1 in graph.OutEdges(uWord))
{
cWord = edge1.Target;
foreach (var edge2 in graph.OutEdges(cWord))
{
kWord = edge2.Target;
WordPair pair = new WordPair(uWord, kWord);
pair = createNewEdges(uWord, kWord, semiCompleteGraph, completeGraph);
if (ooPairsDict.ContainsKey(pair))
{
continue;
}
else
{
ooPairsDict.Add(pair, true);
ooPairs.Add(pair);
}
}
}
}
currentCycle++;
}
/*//2nd cycle to add new blue edge and generate more pairs
* //Add new pair candidate from the semiCompleteGraph
* if (false)//languageOption == 2 && symmetryCycle > 1)
* {
* ooPairsDict.Clear();
* ooPairs.Clear();
*
* foreach (var uWord in uWords)
* {
* Word cWord;
* Word kWord;
*
* foreach (var edge1 in semiCompleteGraph.OutEdges(uWord))
* {
* cWord = edge1.Target;
* foreach (var edge2 in semiCompleteGraph.OutEdges(cWord))
* {
* kWord = edge2.Target;
*
* WordPair pair = new WordPair(null, null);
* pair = createNewBlueEdges(uWord, kWord, semiCompleteGraph);
* if (ooPairsDict.ContainsKey(pair))
* continue;
* else
* {
* ooPairsDict.Add(pair, true);
* ooPairs.Add(pair);
* }
* }
* }
* }
* }
*
* //3rd cycle to add new green edge and generate more pairs
* //Add new pair candidate from the semiCompleteGraph
* if (false)//languageOption == 2 && symmetryCycle > 2)
* {
* ooPairsDict.Clear();
* ooPairs.Clear();
*
* foreach (var uWord in uWords)
* {
* Word cWord;
* Word kWord;
*
* foreach (var edge1 in completeGraph.OutEdges(uWord))
* {
* cWord = edge1.Target;
* foreach (var edge2 in completeGraph.OutEdges(cWord))
* {
* kWord = edge2.Target;
*
* WordPair pair = new WordPair(null, null);
* pair = createNewGreenEdges(uWord, kWord, completeGraph);
* if (ooPairsDict.ContainsKey(pair))
* continue;
* else
* {
* ooPairsDict.Add(pair, true);
* ooPairs.Add(pair);
* }
* }
* }
* }
* }*/
float maxWeight = 0;
return(ooPairs);
}
private WordPair createNewEdges(Word uWord, Word kWord, BidirectionalGraph <Word, Edge <Word> > semiCompleteGraph, BidirectionalGraph <Word, Edge <Word> > completeGraph)//, int uCount, int kCount)
{
Cache1.Clear();
List <SPath> paths = new List <SPath>();
WordPair pair = new WordPair(uWord, kWord);
pair.Polysemy = 0f;
foreach (var item in semiCompleteGraph.OutEdges(uWord)) // Using the updated graph with new edges
{
int inDegreePivot = (int)semiCompleteGraph.InDegree(item.Target);
int outDegreePivot = (int)semiCompleteGraph.OutDegree(item.Target);
int totalSenseEdge = (Math.Max(inDegreePivot, outDegreePivot) - 1) * (inDegreePivot + outDegreePivot);
pair.Polysemy += totalSenseEdge;
//Word pivot_sense = item.Target;
//for (int sense = 1; sense <= totalSense; sense++)
//{
//pivot_sense.Value = pivot_sense.Value + "_sense" + sense;
SLink linkCU = new SLink(item.Target, uWord);
linkCU.Exists = true;// semiCompleteGraph.ContainsEdge(uWord, pivot_sense);
SLink linkCK = new SLink(item.Target, kWord);
linkCK.Exists = semiCompleteGraph.ContainsEdge(item.Target, kWord);
SPath path = new SPath(linkCU, linkCK);
paths.Add(path);
Cache1.Add(item.Target.ID, true);
//}
}
foreach (var item in semiCompleteGraph.InEdges(kWord)) // Using the updated graph with new edges
{
if (Cache1.ContainsKey(item.Source.ID))
{
continue;
}
int inDegreePivot = (int)semiCompleteGraph.InDegree(item.Source);
int outDegreePivot = (int)semiCompleteGraph.OutDegree(item.Source);
int totalSenseEdge = (Math.Max(inDegreePivot, outDegreePivot) - 1) * (inDegreePivot + outDegreePivot);
pair.Polysemy += totalSenseEdge;
SLink linkCK = new SLink(item.Source, kWord);
linkCK.Exists = true;// semiCompleteGraph.ContainsEdge(item.Source, kWord);
SLink linkCU = new SLink(item.Source, uWord);
linkCU.Exists = semiCompleteGraph.ContainsEdge(uWord, item.Source);
SPath path = new SPath(linkCU, linkCK);
paths.Add(path);
}
//calculate probability
//float couverage = Math.Min(uCount, kCount) / (float)Math.Max(uCount, kCount);
float pUK = 0;
float pKU = 0;
float probUK = 0;
float probKU = 0;
//bool hasPolysemy = false;
foreach (var item in paths)
{
//if (!item.LinkCU.Exists || !item.LinkCK.Exists) //containning non-existance link
if (!item.LinkCU.Exists)
{
if (languageOption == 2)
{
completeGraph.AddEdge(new Edge <Word>(item.LinkCU.WordNonPivot, item.LinkCU.WordPivot));
}
continue;
}
if (!item.LinkCK.Exists)
{
if (languageOption == 2)
{
completeGraph.AddEdge(new Edge <Word>(item.LinkCK.WordPivot, item.LinkCK.WordNonPivot));
}
continue;
}
//if ((float)graph.InDegree(item.LinkCU.WordPivot) > 1 || (float)graph.OutDegree(item.LinkCK.WordPivot) > 1)
// hasPolysemy = true;
if (currentCycle == 1)
{
float PrCU = 1.0f / (float)semiCompleteGraph.OutDegree(item.LinkCU.WordNonPivot); //P(C|U) = P(C&U)/P(U)
float PrKC = 1.0f / (float)semiCompleteGraph.OutDegree(item.LinkCK.WordPivot); //P(K|C) = P(K&C)/P(C)
float PrCK = 1.0f / (float)semiCompleteGraph.InDegree(item.LinkCK.WordNonPivot); //P(C|K) = P(C&K)/P(K)
float PrUC = 1.0f / (float)semiCompleteGraph.InDegree(item.LinkCU.WordPivot); //P(U|C) = P(U&C)/P(C)
pKU += PrCU * PrKC;
pUK += PrCK * PrUC;
}
else
{
float PrCU = 0.0f;
float PrKC = 0.0f;
float PrCK = 0.0f;
float PrUC = 0.0f;
foreach (var downEdgeCU in semiCompleteGraph.OutEdges(item.LinkCU.WordNonPivot)) //Loop through down-path from nonpivot1 to pivot
{
PrCU += 1.0f / LinkWeightCache[new SLink(downEdgeCU.Target, downEdgeCU.Source)]; //P(C|U) = P(C&U)/P(U)
}
foreach (var downEdgeCK in semiCompleteGraph.OutEdges(item.LinkCK.WordPivot)) //Loop through down-path from pivot to nonpivot2
{
PrKC += 1.0f / LinkWeightCache[new SLink(downEdgeCK.Source, downEdgeCK.Target)]; //P(K|C) = P(K&C)/P(C)
}
foreach (var upEdgeCK in semiCompleteGraph.InEdges(item.LinkCK.WordNonPivot)) //Loop through up-path from nonpivot2 to pivot
{
PrCK += 1.0f / LinkWeightCache[new SLink(upEdgeCK.Target, upEdgeCK.Source)]; //P(C|K) = P(C&K)/P(K)
}
foreach (var upEdgeCU in semiCompleteGraph.InEdges(item.LinkCU.WordPivot)) //Loop through up-path from pivot to nonpivot1
{
PrUC += 1.0f / LinkWeightCache[new SLink(upEdgeCU.Source, upEdgeCU.Target)]; //P(U|C) = P(U&C)/P(C)
}
PrCU = 1.0f / PrCU;
PrKC = 1.0f / PrKC;
PrCK = 1.0f / PrCK;
PrUC = 1.0f / PrUC;
pUK += PrUC * PrCK;
pKU += PrKC * PrCU;
}
}
probUK = pUK * pKU;
//WordPair pair = new WordPair(uWord, kWord);
//pair.HasMissingEdge = hasPolysemy;
pair.Paths = paths;
pair.Prob = (float)probUK;
pair.Polysemy *= (1 - pair.Prob);
//set link weights
foreach (var item in pair.Paths)
{
//CU
//float polysemyCost = 1 / ((float)graph.InDegree(item.LinkCU.WordPivot) * (float)graph.OutDegree(item.LinkCK.WordPivot));
if (item.LinkCU.Exists)
{
item.LinkCU.Pr = 1f; //polysemyCost;
if (!LinkWeightCache.ContainsKey(item.LinkCU))
{
LinkWeightCache.Add(item.LinkCU, item.LinkCU.Pr);
}
}
else
{
//pair.HasMissingCUEdge = true;
float value = 0;
if (LinkWeightCache.TryGetValue(item.LinkCU, out value))
{
if (pair.Prob > value)
{
item.LinkCU.Pr = LinkWeightCache[item.LinkCU] = pair.Prob; //polysemyCost *
}
else
{
item.LinkCU.Pr = value;
}
}
else
{
item.LinkCU.Pr = pair.Prob; //polysemyCost *
LinkWeightCache.Add(item.LinkCU, pair.Prob);
}
}
//CK
if (item.LinkCK.Exists) //false)//
{
item.LinkCK.Pr = 1f; //polysemyCost
if (!LinkWeightCache.ContainsKey(item.LinkCK))
{
LinkWeightCache.Add(item.LinkCK, item.LinkCK.Pr);
}
}
else
{
float value = 0;
if (LinkWeightCache.TryGetValue(item.LinkCK, out value))
{
if (pair.Prob > value)
{
LinkWeightCache[item.LinkCK] = item.LinkCK.Pr = pair.Prob; //polysemyCost *
}
else
{
item.LinkCK.Pr = value;
}
}
else
{
item.LinkCK.Pr = pair.Prob; //polysemyCost *
LinkWeightCache.Add(item.LinkCK, pair.Prob);
}
}
}
return(pair);
}