public void CanHandleUtf8()
{
using var fastText = new FastTextWrapper(loggerFactory: _loggerFactory);
string outPath = Path.Combine(_tempDir, "rus");
fastText.Supervised("data.rus.txt", outPath, new SupervisedArgs());
var labels = fastText.GetLabels();
labels.Length.Should().Be(2);
labels.Should().Contain(new[] { "__label__оператор", "__label__выход" });
var pred = fastText.PredictSingle("Позови оператора");
pred.Probability.Should().BeGreaterThan(0);
pred.Label.Should().Be("__label__оператор");
var sourceWords = File.ReadAllText("data.rus.txt")
.Split(new[] { " ", "\r\n", "\n" }, StringSplitOptions.RemoveEmptyEntries)
.Where(x => !x.StartsWith("__label__"))
.Distinct().ToArray();
var nn = fastText.GetNearestNeighbours("оператор", 2);
nn.Length.Should().Be(2);
sourceWords.Should().Contain(nn.Select(x => x.Label));
foreach (var prediction in nn)
{
prediction.Probability.Should().BeGreaterThan(0);
}
}
public void SkipgramAndCBowLearnDifferentRepresentations()
{
using var sg = new FastTextWrapper(loggerFactory: _loggerFactory);
string outSG = Path.Combine(_tempDir, "cooking");
sg.Unsupervised(UnsupervisedModel.SkipGram, "cooking.train.nolabels.txt", outSG);
using var cbow = new FastTextWrapper(loggerFactory: _loggerFactory);
string outCbow = Path.Combine(_tempDir, "cooking");
cbow.Unsupervised(UnsupervisedModel.CBow, "cooking.train.nolabels.txt", outCbow);
var nnSg = sg.GetNearestNeighbours("pot", 10);
var nnCbow = cbow.GetNearestNeighbours("pot", 10);
var nnSup = _fixture.FastText.GetNearestNeighbours("pot", 10);
void CheckPair(Prediction[] first, Prediction[] second)
{
int samePredictions = 0;
foreach (var prediction in first)
{
if (second.Any(x => x.Label == prediction.Label))
{
samePredictions++;
}
}
// We want less than a half of same predictions.
samePredictions.Should().BeLessThan(first.Length / 2);
}
CheckPair(nnSg, nnCbow);
CheckPair(nnSg, nnSup);
CheckPair(nnCbow, nnSup);
}
