Beispiel #1
0
        // get the wordVec
        // lowercase the word
        public virtual DoubleVec GetWordVec(string word)
        {
            string    inWord = GetKeyWord(word);
            DoubleVec @out   = wordVecMap_.GetValueOrNull(inWord);

            return(@out);
        }
        private static void TestOnSet(Word2Vec w2vIm, Word2Vec w2vOm)
        {
            string inText = "He was diagnosed early on set dementia 3 years ago.";

            TextObj         textObj    = new TextObj(inText);
            List <TokenObj> inTextList = textObj.GetTokenList();
            // remove space token from the list
            List <TokenObj> nonSpaceTokenList = TextObj.GetNonSpaceTokenObjList(inTextList);

            Console.WriteLine("==========================================");
            Console.WriteLine("-- inTextList: [" + inText + "]");
            int  tarPos           = 4;
            int  tarSize          = 2;   // "on set" has 2 tokens
            int  radius           = 2;
            bool word2VecSkipWord = true;
            bool debugFlag        = false;
            // 1 context with window radius
            DoubleVec    contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, w2vIm, radius, word2VecSkipWord, debugFlag);
            string       str1       = "onset";
            ContextScore s1         = new ContextScore(str1, contextVec, w2vOm);
            string       str2       = "on set";
            ContextScore s2         = new ContextScore(str2, contextVec, w2vOm);

            Console.WriteLine("- [" + str1 + "]: " + s1.ToString());
            Console.WriteLine("- [" + str2 + "]: " + s2.ToString());
        }
        private static void Test(string inTerm, DoubleVec contextVec, Word2Vec w2vIm, Word2Vec w2vOm)
        {
            double cwobScore  = GetScore(inTerm, contextVec, w2vOm);
            double cwobScore2 = GetScore2(inTerm, contextVec, w2vOm);
            double cosScore   = GetSimilarityScore(inTerm, contextVec, w2vIm);

            Console.WriteLine(inTerm + "|" + string.Format("{0,1:F8}", cwobScore) + "|" + string.Format("{0,1:F8}", cwobScore2) + "|" + string.Format("{0,1:F8}", cosScore));
        }
        // try two different method, inner dot or cos
        private static double GetCwobScore(DoubleVec vec1, DoubleVec vec2)
        {
            //double score = DoubleVecUtil.GetInnerDot(vec1, vec2);
            // normailzed score between -1.0 to 1.0
            double score = DoubleVecUtil.GetCosineSimilarity(vec1, vec2);

            return(score);
        }
        // private method
        private static double GetSimilarityScore(DoubleVec vec1, DoubleVec vec2)
        {
            double score = DoubleVecUtil.GetCosineSimilarity(vec1, vec2);

            // set the min. to 0.0
            score = (score <= 0.0 ? 0.0d : score);
            return(score);
        }
 // public constructor
 /// <summary>
 /// Public constructor for ContextScore
 /// </summary>
 /// <param name="inTerm"> target token or candidate (can be multiword) </param>
 /// <param name="contextVec"> wordVec of context from IM </param>
 /// <param name="word2Vec"> word2Vec matrix of OM </param>
 public ContextScore(string inTerm, DoubleVec contextVec, Word2Vec word2Vec)
 {
     term_ = inTerm;
     // Use Cosine Similarity between IM and OM
     score_ = Word2VecScore.GetScore(inTerm, contextVec, word2Vec);
     // TBD: use 2-3 gram
     //score_ = NgramScore.GetScore(word, ngram);
 }
        // use cosineSimilarity instead of inner dot
        private static double GetInnerDotScore(DoubleVec vec1, DoubleVec vec2)
        {
            double score = DoubleVecUtil.GetInnerDot(vec1, vec2);

            // set the min. to 0.0
            // it seems -score is useful for split ....
            //score = (score <= 0.0?0.0d:score);
            return(score);
        }
        // from ensemble paper, use the word vector (Input Matrix) for w2v
        // word2VecIm: word2Vec input matrix - syn0
        // Similarity score use word2Vec Im
        public static double GetSimilarityScore(string inTerm, DoubleVec contextVec, Word2Vec word2VecIm)
        {
            // 1. Get Avg. score for inTerm
            DoubleVec termVec = GetWordVecForTerm(inTerm, word2VecIm);
            // 2. Get Cosine similarity between contextVec and tarVec
            double score = GetSimilarityScore(termVec, contextVec);

            return(score);
        }
        // Use Avg. word2Vec Om for each word in the inTerm
        private static DoubleVec GetWordVecForTerm(string inTerm, Word2Vec w2vOm)
        {
            List <string> inWordList = TermUtil.ToWordList(inTerm);
            // avg. the wordVec if inTerm is a multiword
            DoubleVec outWordVec = GetAvgWordVecForList(inWordList, w2vOm);

            // TBD: take care of possesive
            return(outWordVec);
        }
Beispiel #10
0
        // context from all inTextList, no specify on window radius
        public static DoubleVec GetContextVec(int tarPos, int tarSize, List <TokenObj> nonSpaceTokenList, Word2Vec w2vIm, bool word2VecSkipWord, bool debugFlag)
        {
            // 1. get the context
            List <string> contextList = GetContext(tarPos, tarSize, nonSpaceTokenList, w2vIm, word2VecSkipWord, debugFlag);
            // 2. get the wordVec for the context
            DoubleVec contextVec = Word2VecScore.GetAvgWordVecForList(contextList, w2vIm);

            return(contextVec);
        }
        // public method
        // inTerm: candidate (can be multiword)
        // contextVec: wordVec of context
        // word2VecOm: word2Vec output matrix - syn1neg
        // Use CWOB model to predict the target word = H X OM
        public static double GetScore(string inTerm, DoubleVec contextVec, Word2Vec w2vOm)
        {
            // 1. Get Avg. Vec for term (candidate from prediction)
            DoubleVec termVec = GetWordVecForTerm(inTerm, w2vOm);
            // 2. got the inner dot between hidden layer (context) and OM
            // to predict the output matrix in CBOW
            double score = GetCwobScore(termVec, contextVec);

            return(score);
        }
 // private constructor
 public CSpellScore(string wordStr, string candStr, WordWcMap wordWcMap, DoubleVec contextVec, Word2Vec word2Vec, double wf1, double wf2, double wf3)
 {
     wordStr_ = wordStr;
     candStr_ = candStr;
     // calculate score
     oScore_ = new OrthographicScore(wordStr_, candStr_, wf1, wf2, wf3);
     fScore_ = new FrequencyScore(candStr_, wordWcMap);
     nScore_ = new NoisyChannelScore(wordStr_, candStr_, wordWcMap, wf1, wf2, wf3);
     cScore_ = new ContextScore(candStr_, contextVec, word2Vec);
 }
        // return the best ranked str from candidates using word2Vec score
        // inTokenList, includes space token, is not coreTerm.Lc
        // return null if no candidate is found to correct
        public static MergeObj GetTopRankMergeObj(HashSet <MergeObj> candidates, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, double rwMergeFactor, bool debugFlag)
        {
            // init the topRankMergeObj
            MergeObj topRankMergeObj = null;

            if (candidates.Count > 0)
            {
                // 1. find sorted score list for each candidates ...
                List <ContextScore> candScoreList = GetCandidateScoreList(candidates, nonSpaceTokenList, word2VecIm, word2VecOm, word2VecSkipWord, contextRadius, debugFlag);
                // 2. find the top ranked str
                // the 0 element has the highest score because it is sorted
                // only 1 candidate, use it for nonWord
                ContextScore topContextScore = null;
                if (candScoreList.Count > 0)
                {
                    topContextScore = candScoreList[0];
                }
                // 3. find the mergeObj from the topRankStr (if exist)
                if (topContextScore != null)
                {
                    // 3.1. convert mergeObj set to string set
                    // key: coreMergeWord, MergeObj
                    Dictionary <string, MergeObj> candStrMergeObjMap = new Dictionary <string, MergeObj>();
                    foreach (MergeObj mergeObj in candidates)
                    {
                        string mergeWord = mergeObj.GetCoreMergeWord();
                        candStrMergeObjMap[mergeWord] = mergeObj;
                    }
                    HashSet <string> andStrSet = new HashSet <string>(candStrMergeObjMap.Keys);
                    // 3.2 convert back from top rank str to MergeObj
                    // topRankStr should never be null because candidates is > 0
                    string topRankStr = topContextScore.GetTerm();
                    topRankMergeObj = candStrMergeObjMap.GetValueOrNull(topRankStr);
                    // 4. compare the top rank merge to the original string b4 merge
                    // 1. get the word2Vec score for the orgMergeTerm b4 merge
                    // 1.1 wordVec for context
                    int tarPos = topRankMergeObj.GetStartPos();
                    // tarSize is the total token No of the orgMergeWords
                    int       tarSize    = topRankMergeObj.GetEndPos() - topRankMergeObj.GetStartPos() + 1;
                    DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
                    // 1.2 wordVec for the original words before merge
                    string       orgMergeWord    = topRankMergeObj.GetOrgMergeWord();
                    ContextScore orgContextScore = new ContextScore(orgMergeWord, contextVec, word2VecOm);
                    // validate top merge candidate, set to null if false
                    if (IsTopCandValid(orgContextScore, topContextScore, rwMergeFactor, debugFlag) == false)
                    {
                        // set to null if score is not good enough for corection
                        topRankMergeObj = null;
                    }
                }
            }
            return(topRankMergeObj);
        }
        // private method
        // Test merge and Split
        private static void Test(string inText, int tarPos, int tarSize, int radius, string mergedWord, string splitWords, Word2Vec w2vIm, Word2Vec w2vOm)
        {
            // 0. process the inText
            TextObj         textObj    = new TextObj(inText);
            List <TokenObj> inTextList = textObj.GetTokenList();
            // remove space token from the list
            List <TokenObj> nonSpaceTokenList = TextObj.GetNonSpaceTokenObjList(inTextList);

            Console.WriteLine("==========================================");
            Console.WriteLine("-- inTextList: [" + inText + "]");
            bool word2VecSkipWord = true;
            bool debugFlag        = false;
            // 1.a context with window radius
            DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, w2vIm, radius, word2VecSkipWord, debugFlag);
            // 1.b context with all inText
            DoubleVec contextVecA = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, w2vIm, word2VecSkipWord, debugFlag);
            // 1.c get score1
            ContextScore score1  = new ContextScore(mergedWord, contextVec, w2vOm);
            ContextScore score1a = new ContextScore(mergedWord, contextVecA, w2vOm);

            Console.WriteLine(score1.ToString() + "|" + string.Format("{0,1:F8}", score1a.GetScore()));
            // 2. split words
            ContextScore score2  = new ContextScore(splitWords, contextVec, w2vOm);
            ContextScore score2a = new ContextScore(splitWords, contextVecA, w2vOm);

            Console.WriteLine(score2.ToString() + "|" + string.Format("{0,1:F8}", score2a.GetScore()));
            // 3. 3. 3. Use avg. score on single words
            // This method use different context for each single word
            List <string> splitWordList = TermUtil.ToWordList(splitWords);
            int           index         = 0;
            double        scoreSAvg     = 0.0d;  // radius
            double        scoreSAAvg    = 0.0d;  // all inText

            //debugFlag = false;
            foreach (string splitWord in splitWordList)
            {
                // window radius
                DoubleVec    contextVecS = Word2VecContext.GetContextVec(tarPos + index, 1, nonSpaceTokenList, w2vIm, radius, word2VecSkipWord, debugFlag);
                ContextScore scoreS      = new ContextScore(splitWord, contextVecS, w2vOm);
                //System.out.println("-- " + scoreS.ToString());
                scoreSAvg += scoreS.GetScore();
                // all text
                DoubleVec    contextVecSA = Word2VecContext.GetContextVec(tarPos + index, 1, nonSpaceTokenList, w2vIm, word2VecSkipWord, debugFlag);
                ContextScore scoreSA      = new ContextScore(splitWord, contextVecSA, w2vOm);
                //System.out.println("-- " + scoreSA.ToString());
                scoreSAAvg += scoreSA.GetScore();
                index++;
            }
            scoreSAvg  = scoreSAvg / index;            // window
            scoreSAAvg = scoreSAAvg / index;           // all text
            Console.WriteLine("Avg. Single Word|" + string.Format("{0,1:F8}", scoreSAvg) + "|" + string.Format("{0,1:F8}", scoreSAAvg));
        }
        private static void TestWin(string inTerm, DoubleVec contextVec, Word2Vec w2vIm, Word2Vec w2vOm)
        {
            double cwobScore  = GetScore(inTerm, contextVec, w2vOm);
            double cwobScore2 = GetScore2(inTerm, contextVec, w2vOm);
            double cosScore   = GetSimilarityScore(inTerm, contextVec, w2vIm);

            Console.Write(string.Format("{0,1:F4}", cwobScore) + "-" + string.Format("{0,1:F4}", cwobScore2) + "|");

            /*
             * System.out.println(inTerm + "|" + String.format("%1.8f", cwobScore)
             + "|" + String.format("%1.8f", cosScore));
             */
        }
Beispiel #16
0
        // return candidate set with cSpell score
        // wordStr is the srcTxt used to calculate the score between it and cand
        public static HashSet <CSpellScore> GetCandidateScoreSet(string wordStr, HashSet <string> candidates, WordWcMap wordWcMap, int tarPos, int tarSize, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, double wf1, double wf2, double wf3, bool debugFlag)
        {
            HashSet <CSpellScore> candScoreSet = new HashSet <CSpellScore>();

            foreach (string cand in candidates)
            {
                // find context for each candidates
                DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);

                CSpellScore cs = new CSpellScore(wordStr, cand, wordWcMap, contextVec, word2VecOm, wf1, wf2, wf3);
                candScoreSet.Add(cs);
            }
            return(candScoreSet);
        }
        private static bool CheckRealWord1To1Rules(ContextScore topContextScore, string inStr, int tarPos, int tarSize, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, double rw1To1Factor, bool debugFlag)
        {
            // return false if no topCand found
            if ((topContextScore == null) || (topContextScore.GetTerm().Equals(inStr)))
            {
                return(false);
            }
            // 1. get the word2Vec score for the org inStr b4 one-to-one
            // 1.1 wordVec for context
            DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
            // 1.2 wordVec for the original words before one-to-one
            ContextScore orgCs = new ContextScore(inStr, contextVec, word2VecOm);

            DebugPrint.Println("--- Real-Word One-To-One Context Score Detail: ---", debugFlag);
            DebugPrint.Println("- Score - orgTerm: " + orgCs.ToString(), debugFlag);
            DebugPrint.Println("- Score - top 1-to-1: " + topContextScore.ToString(), debugFlag);
            DebugPrint.Println("- rw1To1Factor: " + rw1To1Factor, debugFlag);
            // Score rules for one-to-one
            double orgScore = orgCs.GetScore();
            double topScore = topContextScore.GetScore();
            bool   flag     = false;

            // 2.1 no one-to-one correction if orgScore is 0.0d, no word2Vec information
            if (orgScore < 0.0d)
            {
                // 2.2a one-to-one if the org score is negative and top score is positive
                if (topScore > 0.0d)
                {
                    // another rule for word2Vec on real-word
                    if (((topScore - orgScore) > 0.085) && (orgScore > -0.085))                       // help from 0.6812 to 0.6877
                    {
                        flag = true;
                    }
                }
                // 2.2b one-to-one if the org score is negative and top score is better
                else if ((topScore < 0.0d) && (topScore > orgScore * rw1To1Factor))
                {
                    flag = true;
                }
            }
            else if (orgScore > 0.0d)
            {
                // 2.3a merge if the org score is positive and better 0.01*topScore
                if (topScore * rw1To1Factor > orgScore)
                {
                    flag = true;
                }
            }
            return(flag);
        }
Beispiel #18
0
        private void ReadWordVectors(string inFile, bool verboseFlag)
        {
            // init
            wordVecMap_ = new Dictionary <string, DoubleVec>();
            int    lineNo = 0;
            string line   = null;

            try {
                StreamReader @in = new StreamReader(new FileStream(inFile, FileMode.Open, FileAccess.Read), Encoding.UTF8);
                // read in line by line from a file
                while (!string.ReferenceEquals((line = @in.ReadLine()), null))
                {
                    var buf = line.Split(" ");
                    // first line is the stats wordNo|dimension
                    if (lineNo == 0)
                    {
                        wordNo_    = int.Parse(buf[0]);
                        dimension_ = int.Parse(buf[1]);
                        //dimension_ = line0.split(delimiter).length -1;
                    }
                    else                         // word|vector
                    {
                        string   word  = buf[0];
                        double[] array = new double[dimension_];
                        for (int i = 0; i < dimension_; i++)
                        {
                            array[i] = double.Parse(buf[i + 1]);
                        }
                        // update wordVec
                        wordVecMap_[word] = new DoubleVec(array);
                    }
                    lineNo++;
                }
                // close
                @in.Close();
            } catch (IOException e) {
                Console.Error.WriteLine("** [email protected]( ), problem of reading file (" + inFile);
                Console.Error.WriteLine("Line: " + lineNo + " - " + line);
                Console.Error.WriteLine("Exception: " + e.ToString());
            }
            // print out
            if (verboseFlag == true)
            {
                Console.WriteLine("====== Word2Vec.ReadWordVectors( ) ======");
                Console.WriteLine("- inFile: " + inFile);
                Console.WriteLine("- Word no: " + wordNo_);
                Console.WriteLine("- dimension: " + dimension_);
            }
        }
        // return candidate set with context score
        // word2Vec is the word|wordVec map to get the wordVec
        // Not sorted, because it is a set
        // tarPos: starting position of target token
        // tarSize: token size of target token (single word = 1)
        // contextRadius: windown radius
        public static HashSet <ContextScore> GetCandidateScoreSet(HashSet <string> candidates, int tarPos, int tarSize, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, bool debugFlag)
        {
            // 1. get the context and contextVec, using input matrix
            DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);

            // 2. get context score for all candidates
            HashSet <ContextScore> candScoreSet = new HashSet <ContextScore>();

            foreach (string cand in candidates)
            {
                // get ContextSocre for each candidates, use output matrix
                ContextScore cs = new ContextScore(cand, contextVec, word2VecOm);
                candScoreSet.Add(cs);
            }
            return(candScoreSet);
        }
        // return candidate set with context score
        // word2Vec is the word|wordVec map to get the wordVec
        // Not sorted, because it is a set
        // tarPos: starting position of target token
        // tarSize: token size of target token (single word = 1)
        public static HashSet <ContextScore> GetCandidateScoreSet(HashSet <MergeObj> candidates, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, bool debugFlag)
        {
            HashSet <ContextScore> candScoreSet = new HashSet <ContextScore>();

            // get context score for all candidates
            // go through all merge candidates, all have differetn context
            foreach (MergeObj mergeObj in candidates)
            {
                // 1. get the context and contextVec, using input matrix
                int       tarPos     = mergeObj.GetStartPos();
                int       tarSize    = mergeObj.GetEndPos() - mergeObj.GetStartPos() + 1;
                DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
                // 2. get ContextSocre for each merge, use output matrix
                string       mergeWord = mergeObj.GetCoreMergeWord();
                ContextScore cs        = new ContextScore(mergeWord, contextVec, word2VecOm);
                candScoreSet.Add(cs);
            }
            return(candScoreSet);
        }
        // this method is to be deleted because it has same result as GetScore()
        public static double GetScore2(string inTerm, DoubleVec contextVec, Word2Vec w2vOm)
        {
            List <string> inWordList = TermUtil.ToWordList(inTerm);
            double        score      = 0.0d;
            int           count      = 0;

            foreach (string word in inWordList)
            {
                DoubleVec wordVec = w2vOm.GetWordVec(word);
                if (wordVec != null)
                {
                    score += GetCwobScore(wordVec, contextVec);
                }
                count++;
            }
            // add score first, then calculate the avg.
            score = score / count;
            return(score);
        }
        // return the best ranked str from candidates using context score
        // this method is replaced by GetTopRankStr, which sorted by comparator
        public static string GetTopRankStrByScore(string inStr, HashSet <string> candidates, int tarPos, int tarSize, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, bool debugFlag)
        {
            // 1. get the context and contextVec
            DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
            string    topRankStr = inStr;
            double    maxScore   = 0.0d;

            foreach (string cand in candidates)
            {
                ContextScore cs    = new ContextScore(cand, contextVec, word2VecOm);
                double       score = cs.GetScore();
                // update only if the score is > 0.0d
                if (score > maxScore)
                {
                    topRankStr = cand;
                    maxScore   = score;
                }
            }
            return(topRankStr);
        }
Beispiel #23
0
        // return the best ranked str from candidates using word2Vec score
        // inTokenList, includes space token, is not coreTerm.Lc
        // return the orignal inStr if no candidate has score > 0.0d
        public static string GetTopRankStr(string inStr, HashSet <string> candidates, int tarPos, int tarSize, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, int shortSplitWordLength, int maxShortSplitWordNo, double rwSplitFactor, int maxCandNo, bool debugFlag)
        {
            // init
            string topRankStr = inStr;

            // Find the correction str
            if (candidates.Count > 0)
            {
                // 1. sorted score list for each candidates ...
                // This ranking can be improved if n-gram model (frequecny) is used
                List <ContextScore> candScoreList = RankByContext.GetCandidateScoreList(candidates, tarPos, tarSize, nonSpaceTokenList, word2VecIm, word2VecOm, word2VecSkipWord, contextRadius, debugFlag);
                // 1.1 get the top tank candidate
                ContextScore topContextScore = candScoreList[0];
                // 2. validate the top rank
                // 2.1 wordVec for context
                DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
                // 2.2 wordVec for the original words before split
                ContextScore orgContextScore = new ContextScore(inStr, contextVec, word2VecOm);
                // 2.3 compare the top rank split to the original string b4 split
                if (IsTopCandValid(inStr, orgContextScore, topContextScore, rwSplitFactor, debugFlag) == true)
                {
                    // no correction: if score is not good enough for corection
                    topRankStr = topContextScore.GetTerm();
                }
                // debug print
                if (debugFlag == true)
                {
                    // print focus token (original)
                    DebugPrint.PrintCScore(orgContextScore.ToString(), debugFlag);
                    // print candidates
                    ContextScoreComparator <ContextScore> csc = new ContextScoreComparator <ContextScore>();
                    var list = candScoreList.OrderBy(x => x, csc).Take(maxCandNo).Select(obj => obj.ToString()).ToList();
                    foreach (var item in list)
                    {
                        DebugPrint.PrintCScore(item, debugFlag);
                    }
                }
            }
            return(topRankStr);
        }
        // Average wordVec for a list of words
        public static DoubleVec GetAvgWordVecForList(IList <string> wordList, Word2Vec word2Vec)
        {
            // init the matrix to all zero
            int       dimension  = word2Vec.GetDimension();
            DoubleVec aveWordVec = new DoubleVec(dimension);
            int       count      = 0;

            foreach (string word in wordList)
            {
                DoubleVec wordVec = word2Vec.GetWordVec(word);
                if (wordVec != null)
                {
                    aveWordVec.Add(wordVec);
                }
                count++;
            }
            // calculate the avg.
            if (count != 0)
            {
                aveWordVec.Divide(count);
            }
            return(aveWordVec);
        }
        // return the best ranked str from candidates using context score
        // this method is replaced by GetTopRankStr, which sorted by comparator
        public static MergeObj GetTopRankMergeObjByScore(HashSet <MergeObj> candidates, List <TokenObj> nonSpaceTokenList, Word2Vec word2VecIm, Word2Vec word2VecOm, bool word2VecSkipWord, int contextRadius, bool debugFlag)
        {
            MergeObj topRankMergeObj = null;
            double   maxScore        = 0.0d;

            foreach (MergeObj mergeObj in candidates)
            {
                // 1. get the context and contextVec
                int       tarPos     = mergeObj.GetStartPos();
                int       tarSize    = mergeObj.GetEndPos() - mergeObj.GetStartPos() + 1;
                DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
                // 2. get ContextSocre for each merge, use output matrix
                string       mergeWord = mergeObj.GetCoreMergeWord();
                ContextScore cs        = new ContextScore(mergeWord, contextVec, word2VecOm);
                double       score     = cs.GetScore();
                // update only if the score is > 0.0d
                if (score > maxScore)
                {
                    topRankMergeObj = mergeObj;
                    maxScore        = score;
                }
            }
            return(topRankMergeObj);
        }
        private static void Tests(Word2Vec w2vIm, Word2Vec w2vOm)
        {
            string          inText     = "for the last 10 years    was dianosed\n early on set deminita 3 years ago";
            List <TokenObj> inTextList = TextObj.TextToTokenList(inText);
            // remove space token from the list
            List <TokenObj> nonSpaceTokenList = TextObj.GetNonSpaceTokenObjList(inTextList);
            List <string>   testStrList       = new List <string>();

            testStrList.Add("diagnosed");
            testStrList.Add("diagnose");
            testStrList.Add("dianosed");
            // init context
            int       tarPos           = 6;
            int       tarSize          = 1;
            int       radius           = 2;
            bool      word2VecSkipWord = true;
            bool      debugFlag        = false;
            DoubleVec contextVec       = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, w2vIm, radius, word2VecSkipWord, debugFlag);

            Console.WriteLine("===== Test diagnosed|diagnose|dianosed (window-2) =====");
            Console.WriteLine("inText: [" + inText + "]");
            Console.WriteLine("============================================");
            Console.WriteLine("Candidates|CBOW score|CBOW score 2|Similarity score");
            Console.WriteLine("============================================");
            foreach (string testStr in testStrList)
            {
                Test(testStr, contextVec, w2vIm, w2vOm);
            }
            Console.WriteLine("===== Test diagnosed|diagnose|dianosed (whole text) =====");
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, w2vIm, word2VecSkipWord, debugFlag);
            foreach (string testStr in testStrList)
            {
                Test(testStr, contextVec, w2vIm, w2vOm);
            }
            string          inText1     = "Not all doctors know about this syndrome.";
            List <TokenObj> inTextList1 = TextObj.TextToTokenList(inText1);
            // remove space token from the list
            List <TokenObj> nonSpaceTokenList1 = TextObj.GetNonSpaceTokenObjList(inTextList1);

            Console.WriteLine("===== Test know about|know|about (window) =====");
            List <string> testStrList1 = new List <string>();

            testStrList1.Add("know about");
            testStrList1.Add("know");
            testStrList1.Add("about");
            tarPos     = 3;
            tarSize    = 2;
            radius     = 2;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList1, w2vIm, radius, word2VecSkipWord, debugFlag);
            Test(testStrList1[0], contextVec, w2vIm, w2vOm);
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList1, w2vIm, word2VecSkipWord, debugFlag);
            Test(testStrList1[0], contextVec, w2vIm, w2vOm);
            tarPos     = 3;
            tarSize    = 1;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList1, w2vIm, radius, word2VecSkipWord, debugFlag);
            Test(testStrList1[1], contextVec, w2vIm, w2vOm);
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList1, w2vIm, word2VecSkipWord, debugFlag);
            Test(testStrList1[1], contextVec, w2vIm, w2vOm);
            tarPos     = 4;
            tarSize    = 1;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList1, w2vIm, radius, word2VecSkipWord, debugFlag);
            Test(testStrList1[2], contextVec, w2vIm, w2vOm);
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList1, w2vIm, word2VecSkipWord, debugFlag);
            Test(testStrList1[2], contextVec, w2vIm, w2vOm);

            string          inText2     = "for the last   10 years was diagnosed early on set dementia 3 years ago.";
            List <TokenObj> inTextList2 = TextObj.TextToTokenList(inText2);
            // remove space token from the list
            List <TokenObj> nonSpaceTokenList2 = TextObj.GetNonSpaceTokenObjList(inTextList2);
            List <string>   testStrList2       = new List <string>();

            testStrList2.Add("onset");
            testStrList2.Add("on set");
            Console.WriteLine("===== Test onset|on set (window-3) =====");
            Console.WriteLine("inText: [" + inText + "]");
            tarPos     = 8;
            tarSize    = 2;
            radius     = 3;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList2, w2vIm, radius, word2VecSkipWord, debugFlag);
            foreach (string testStr in testStrList2)
            {
                Test(testStr, contextVec, w2vIm, w2vOm);
            }
            tarPos     = 8;
            tarSize    = 1;
            radius     = 3;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList2, w2vIm, radius, word2VecSkipWord, debugFlag);
            Test("on", contextVec, w2vIm, w2vOm);
            tarPos     = 9;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList2, w2vIm, radius, word2VecSkipWord, debugFlag);
            Test("set", contextVec, w2vIm, w2vOm);
            Console.WriteLine("===== Test onset|on set (whole text) =====");
            radius     = nonSpaceTokenList2.Count;
            contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList2, w2vIm, word2VecSkipWord, debugFlag);
            foreach (string testStr in testStrList2)
            {
                Test(testStr, contextVec, w2vIm, w2vOm);
            }
            Console.WriteLine("===== Go through each tokens with diff radius 1-9) =====");
            Console.WriteLine("tarPos|tarWord|r=1|r=2|r=3|r=4|r=5|r=6|r=7|r=8|r=9");
            //String inText3 = "Broken bones can not sleep at night!";
            string          inText3     = "not xyxy all doctors know about this syndrome.";
            List <TokenObj> inTextList3 = TextObj.TextToTokenList(inText3);
            // remove space token from the list
            List <TokenObj> nonSpaceTokenList3 = TextObj.GetNonSpaceTokenObjList(inTextList3);

            tarPos  = 0;
            tarSize = 1;
            radius  = 0;
            foreach (TokenObj tokenObj in nonSpaceTokenList3)
            {
                // skip the space token
                string tokenStr = tokenObj.GetTokenStr();
                string inStr    = Word2VecContext.NormWordForWord2Vec(tokenStr);
                Console.Write(tarPos + "|" + tokenStr + "|");
                // print out all radius
                for (int r = 1; r < 10; r++)
                {
                    contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, inTextList2, w2vIm, r, word2VecSkipWord, debugFlag);
                    TestWin(inStr, contextVec, w2vIm, w2vOm);
                }
                Console.WriteLine("");
                tarPos++;
            }
        }
        // return the best ranked str from candidates using word2Vec score
        // inTokenList, includes space token, is not coreTerm.Lc
        // return the orignal inStr if no candidate has score > 0.0d
        public static string GetTopRankStr(string inStr, HashSet <string> candidates, CSpellApi cSpellApi, int tarPos, List <TokenObj> nonSpaceTokenList, bool debugFlag)
        {
            // init
            WordWcMap wordWcMap        = cSpellApi.GetWordWcMap();
            Word2Vec  word2VecIm       = cSpellApi.GetWord2VecIm();
            Word2Vec  word2VecOm       = cSpellApi.GetWord2VecOm();
            int       contextRadius    = cSpellApi.GetRw1To1ContextRadius();
            bool      word2VecSkipWord = cSpellApi.GetWord2VecSkipWord();
            int       maxCandNo        = cSpellApi.GetCanMaxCandNo();
            double    wf1        = cSpellApi.GetOrthoScoreEdDistFac();
            double    wf2        = cSpellApi.GetOrthoScorePhoneticFac();
            double    wf3        = cSpellApi.GetOrthoScoreOverlapFac();
            int       tarSize    = 1;    // only for one-to-one, no merge here
            string    topRankStr = inStr;
            // use cSpell top candidates
            int                topNo           = 1; // top sort
            string             inStrLc         = inStr.ToLower();
            List <CSpellScore> cSpellScoreList = RankByCSpellRealWord1To1.GetCandidateScoreList(inStrLc, candidates, wordWcMap, tarPos, tarSize, nonSpaceTokenList, word2VecIm, word2VecOm, word2VecSkipWord, contextRadius, wf1, wf2, wf3, debugFlag);

            // Find the correction str and correct
            if (cSpellScoreList.Count > 0)
            {
                // the rw top rank must be in both NC and orthographic
                CSpellScore  topScore        = cSpellScoreList[0];
                double       topFScore       = topScore.GetFScore().GetScore();         //frequency
                double       topTScore       = topScore.GetOScore().GetTokenScore();    // Token
                double       topPScore       = topScore.GetOScore().GetPhoneticScore(); //Phone
                double       topOScore       = topScore.GetOScore().GetOverlapScore();  //overlap
                ContextScore orgContextScore = null;
                // check the frequency
                // get the max score of frequency, eidt, phonetic, and overlap
                // the top rank must have all top score for above
                if ((topFScore == CSpellScore.GetMaxFScore(cSpellScoreList)) && (topTScore == CSpellScore.GetMaxEScore(cSpellScoreList)) && (topPScore == CSpellScore.GetMaxPScore(cSpellScoreList)) && (topOScore == CSpellScore.GetMaxOScore(cSpellScoreList)))
                {
                    ContextScore topContextScore = topScore.GetCScore();
                    // 1.1 wordVec for context
                    DoubleVec contextVec = Word2VecContext.GetContextVec(tarPos, tarSize, nonSpaceTokenList, word2VecIm, contextRadius, word2VecSkipWord, debugFlag);
                    // 1.2 wordVec for the original words before one-to-one
                    orgContextScore = new ContextScore(inStr, contextVec, word2VecOm);
                    FrequencyScore orgFScore = new FrequencyScore(inStr, wordWcMap);
                    // pass the orgContextScore
                    if (IsTopCandValid(inStr, orgContextScore, topScore, orgFScore, cSpellApi, debugFlag) == true)
                    {
                        // no correction: if score is not good enough for corection
                        topRankStr = topScore.GetCandStr();
                        // debug print for ananlysis
                        /// <summary>
                        ///*
                        /// System.out.println("======= cSpellScoreList.size(): "
                        ///    + cSpellScoreList.size() + " ========");
                        /// System.out.println(inStr
                        ///    + "," + String.format("%1.8f", orgFScore.GetScore())
                        ///    + "," + String.format("%1.8f", orgContextScore.GetScore()));
                        /// System.out.println(CSpellScore.GetScoreHeader());
                        /// for(CSpellScore cSpellScore: cSpellScoreList)
                        /// {
                        ///    System.out.println(cSpellScore.ToString(","));
                        /// }
                        /// **
                        /// </summary>
                    }
                }
                // debug print
                if (debugFlag == true)
                {
                    // print focus token (original)
                    if (orgContextScore != null)
                    {
                        DebugPrint.PrintScore(orgContextScore.ToString(), debugFlag);
                    }
                    else
                    {
                        DebugPrint.PrintScore("No score for focus (" + inStr + ")", debugFlag);
                    }
                    // print candidate
                    var list = cSpellScoreList.Take(maxCandNo).Select(obj => obj.ToString()).ToList();
                    foreach (var item in list)
                    {
                        DebugPrint.PrintScore(item, debugFlag);
                    }
                }
            }
            return(topRankStr);
        }