public string ClassifyArticle(PreprocessedArticle article)
{
KnnArticle knnArticle = new KnnArticle {
Label = article.Label, FeaturesVector = FeaturesVectorService.GetFeaturesVector(article)
};
List <(string, double)> neighbors = new List <(string, double)>();
foreach (KnnArticle classifiedArticle in ClassifiedArticles)
{
neighbors.Add((classifiedArticle.PredictedLabel, Metric.CalculateDistance(classifiedArticle.FeaturesVector, knnArticle.FeaturesVector)));
}
var orderedNeighbors = neighbors.OrderBy(t => t.Item2).Take(K).ToList();
var distinctNeighborsCount = orderedNeighbors.GroupBy(t => t.Item1).Select(g => (g.Key, g.Count())).OrderByDescending(e => e.Item2).ToList();
var neighborsWithSameCount =
distinctNeighborsCount.Where(t => t.Item2 == distinctNeighborsCount[0].Item2).ToList();
if (neighborsWithSameCount.Count != 1)
{
var orderedNeighborsWithSameCount =
orderedNeighbors.Where(c => neighborsWithSameCount.Select(d => d.Item1).Contains(c.Item1));
var distinctNeighborsSum = orderedNeighborsWithSameCount.GroupBy(t => t.Item1).Select(g => (g.Key, g.Sum(t => t.Item2))).OrderBy(e => e.Item2).ToList();
knnArticle.PredictedLabel = distinctNeighborsSum[0].Item1;
ClassifiedArticles.Add(knnArticle);
return(distinctNeighborsSum[0].Item1);
}
knnArticle.PredictedLabel = distinctNeighborsCount[0].Item1;
ClassifiedArticles.Add(knnArticle);
return(distinctNeighborsCount[0].Item1);
}
public Dictionary <string, Dictionary <string, int> > CalculateConfusionMatrix(List <string> tags)
{
Dictionary <string, Dictionary <string, int> > matrix = new Dictionary <string, Dictionary <string, int> >();
foreach (var tag in tags)
{
Dictionary <string, int> row = new Dictionary <string, int>();
foreach (var tag1 in tags)
{
row.Add(tag1, 0);
}
matrix.Add(tag, row);
}
foreach (KnnArticle classifiedArticle in ClassifiedArticles.Skip(coldStartNumber))
{
matrix[classifiedArticle.Label][classifiedArticle.PredictedLabel]++;
}
return(matrix);
}
