public static double rho(RegressionSet simlexInstance, List <SimpleEntry> calculatedResults, int idx)
{
double avg1 = 0;
double avg2 = 0;
double avg12 = 0;
foreach (Entry simLexEntry in simlexInstance.Entries)
{
//W kolejnej linii pobierana jest wartość wygenerowanego podobieństwa pomiędzy słowami z aktualnie przetwarzanego Entry z SimLex'a
SimpleEntry calculatedEntry = calculatedResults.Where(x => x.w1 == simLexEntry.word1 && x.w2 == simLexEntry.word2).First();
avg1 += simLexEntry.simLex;
avg2 += calculatedEntry.result;
avg12 += simLexEntry.simLex * calculatedEntry.result;
}
avg1 /= simlexInstance.Entries.Count;
avg2 /= simlexInstance.Entries.Count;
avg12 /= simlexInstance.Entries.Count;
double cov = avg12 - (avg1 * avg2);
double dev1 = 0;
double dev2 = 0;
foreach (Entry simLexEntry in simlexInstance.Entries)
{
SimpleEntry calculatedEntry = calculatedResults.Where(x => x.w1 == simLexEntry.word1 && x.w2 == simLexEntry.word2).First();
dev1 += (simLexEntry.simLex - avg1) * (simLexEntry.simLex - avg1);
dev2 += (calculatedEntry.result - avg2) * (calculatedEntry.result - avg2);
}
dev1 /= (simlexInstance.Entries.Count - 1);
dev2 /= (simlexInstance.Entries.Count - 1);
dev1 = Math.Sqrt(dev1);
dev2 = Math.Sqrt(dev2);
return(cov / (dev1 * dev2));
}
public List <SimpleEntry> generateEntriesForComparison(RegressionSet input)
{
List <SimpleEntry> entries = new List <SimpleEntry>();
int i = 0;
foreach (Entry entry in input.Entries)
{
System.Console.WriteLine(i++);
//Console.WriteLine(i++);
var t = compare(entry.word1, entry.word2);
double result = 0.5;
try
{
t.Wait();
result = t.GetAwaiter().GetResult();
}
catch
{
}
entries.Add(new SimpleEntry(entry.word1, entry.word2, result));
}
return(entries);
}
static void Main(string[] args)
{
Model m = new Model("D://glove_vec//paragram_filtered//paragram_filtered_vecs.txt");
m = m.cos_retrofit("D://glove_vec//ppdb_synonyms.txt", "D://glove_vec//wordnet_antonyms.txt", 20);
//m.saveModel("D://glove_vec//glove_mrksic//glove_retro_vecs.txt");
Model.loadCentroids("D://glove_vec//glove_mrksic//centroids_retro10");
RegressionSet sl = new RegressionSet("D://SimLex-999//SimLex-999.txt");
Model.sl = sl;
ClassificationSet esl = new ClassificationSet("D://glove_vec//esl", ClassificationSet.SetType.ESL);
ClassificationSet toefl = new ClassificationSet("D://glove_vec//toefl", ClassificationSet.SetType.TOEFL);
//var dict = HelperFunctions.generateDictionary(new List<TestSet> { sl999, esl, toefl });
Model.compStyle = Vector.comparisonStyle.Cosine;
Classifier cEsl = new Classifier(esl, m);
Classifier cToefl = new Classifier(toefl, m);
var r0 = m.generateEntriesForComparison(sl);
double acc = cEsl.accuracy();
double acc2 = cToefl.accuracy();
double rho = HelperFunctions.rho(sl, r0, 0);
//double rho = 0;
//double acc = 0;
//double acc2 = 0;
System.Console.WriteLine("ESL: {0}\nTOEFL: {1}\nSimLex999: {2}", acc, acc2, rho);
Console.ReadKey();
/*
*
* Model.loadCentroids("D://glove_vec/centroids5");
* Model.sl = sl;
*
* Model.compStyle = Vector.comparisonStyle.Cosine;
* var r0 = m.generateEntriesForComparison(sl);
* var result0 = HelperFunctions.rho(sl, r0, 0);
* Console.WriteLine(result0);
* StreamWriter sw = new StreamWriter("D://results.txt");
* Model.compStyle = Vector.comparisonStyle.CosineHR;
*/
/*for (int i = 0; i < 100 ; i++)
* {
* //Model.Beta1 = i/10.0;
* if (i % 50 == 0) Console.WriteLine(i);
* var r = m.generateEntriesForComparison(sl);
* var result = HelperFunctions.rho(sl, r, 0);
* Model.msqe /= 999.0;
* sw.WriteLine("{0}\t{1:F6}\tBeta1:{2}\tBeta2:{3}\tBeta3:{4}\tMSQE: {5}",i/100.0,result,Model.Beta1,Model.Beta2,Model.Beta3, Model.msqe);
* Model.msqe = 0;
* sl.shuffle();
* Model.resetBatch();
* }
* sw.Close();*/
/*
* int remainder = 1029 + 420;
* Stopwatch s = new Stopwatch();
* //Model centroids = new Model(m.Vectors.Length, m.Vectors[0].V_size);
* //remainder = m.Vectors.Length;
* Model centroids = new Model(1029 + 420, m.Vectors[0].V_size);
* s.Start();
* foreach (Vector v in m.Vectors)
* {
* if (sl.Entries.Select(x => x.word1).Contains(v.Label) ||
* sl.Entries.Select(x => x.word2).Contains(v.Label) ||
* esl.Questions.Select(x => x.Answer).Contains(v.Label) ||
* esl.Questions.Select(x => x.Possibilities[0]).Contains(v.Label) ||
* esl.Questions.Select(x => x.Possibilities[1]).Contains(v.Label) ||
* esl.Questions.Select(x => x.Possibilities[2]).Contains(v.Label) ||
* esl.Questions.Select(x => x.Possibilities[3]).Contains(v.Label) ||
* toefl.Questions.Select(x => x.Answer).Contains(v.Label) ||
* toefl.Questions.Select(x => x.Possibilities[0]).Contains(v.Label) ||
* toefl.Questions.Select(x => x.Possibilities[1]).Contains(v.Label) ||
* toefl.Questions.Select(x => x.Possibilities[2]).Contains(v.Label) ||
* toefl.Questions.Select(x => x.Possibilities[3]).Contains(v.Label)
*
* )
* {
*
* var t = m.findNMostSimilar(v, 10);
* var c = m.calculateCentroid(v.Label, t);
* Console.WriteLine("{0}/{1} : {2:F5}%\t {3:D2}:{4:D2}:{5:D2}", remainder, 1029+420, 100 * (1 - (remainder / (1029.0+420.0))), s.Elapsed.Hours, s.Elapsed.Minutes, s.Elapsed.Seconds);
* remainder--;
*
* centroids.Vectors[remainder] = c;
* }
* }
*
* centroids.saveModel("D://glove_vec//centroids_original5");
*/
//var entries = m.generateEntriesForComparison(sl);
//double res = HelperFunctions.rho(sl, entries, 0);
//System.Console.WriteLine("{0}", res);
//Model m = new Model("../../../data/nmrksic/counter-fitting.git/trunk/word_vectors/");
}
