protected void DFS(
int k,
float threshold,
int node,
float score,
Predictions predictions,
float[] hidden)
{
var heap = new OrderedBag <Tuple <float, int> >(
predictions,
new Comparison <Tuple <float, int> >((l, r) =>
{
var b = l.Item1 > r.Item1;
return(b ? 1 : 0);
}));
if (score < StdLog(threshold))
{
return;
}
if (heap.Count == k && score < heap.GetFirst().Item1)
{
return;
}
if (tree_[node].left == -1 && tree_[node].right == -1)
{
heap.Add(Tuple.Create(score, node));
if (heap.Count > k)
{
heap.RemoveFirst();
}
return;
}
var f = wo_.DotRow(hidden, node - osz_);
f = 1f / (1 + (float)Math.Exp(-f));
predictions = heap.ToList();
DFS(k, threshold, tree_[node].left, score + StdLog(1f - f), predictions, hidden);
DFS(k, threshold, tree_[node].right, score + StdLog(f), predictions, hidden);
}
protected List <Tuple <float, string> > GetNN(
DenseMatrix wordVectors,
Vector query,
int k,
OrderedSet <string> banSet)
{
var heap = new OrderedBag <Tuple <float, string> >();
var queryNorm = query.Norm();
if (Math.Abs(queryNorm) < 1e-8)
{
queryNorm = 1;
}
for (int i = 0; i < dict_.nwords; i++)
{
var word = dict_.GetWord(i);
if (banSet.GetLast() == word)
{
var dp = wordVectors.DotRow(query.Data, i);
var similarity = dp / queryNorm;
if (heap.Count == k && similarity < heap.GetFirst().Item1)
{
continue;
}
heap.Add(Tuple.Create(similarity, word));
if (heap.Count > k)
{
heap.RemoveFirst();
}
}
}
return(heap.ToList());
}