public void SetUp()
{
loggerFactory = new NullLoggerFactory();
vectorSource = new EmbeddingVectorSource(WordModel.Load(Path.Combine(TestContext.CurrentContext.TestDirectory, @"Data\model.bin")));
documentReconstructor = new DocumentReconstructor();
documents = JsonConvert.DeserializeObject <Document[]>(File.ReadAllText(Path.Combine(TestContext.CurrentContext.TestDirectory, "Data", "docs.json")));
instance = CreateModelStorage();
}
public async Task LogicTest()
{
var negative = documents.Take(2).Concat(documents.Skip(30)).Select(item => new ProcessingTextBlock(item.Sentences.ToArray())).ToArray();
var positive = documents.Skip(2).Take(29).Select(item => new ProcessingTextBlock(item.Sentences.ToArray())).ToArray();
instance.Add(DataType.Positive, positive);
instance.Add(DataType.Negative, negative);
var pageDetector = await instance.Train(CancellationToken.None).ConfigureAwait(false);
var result = pageDetector.Predict(new DocumentBlock(documents).Pages).Select(item => item ? 1 : 0).ToArray();
var expected = new int[documents.Length];
for (int i = 2; i <= 30; i++)
{
expected[i] = 1;
}
var cm = new GeneralConfusionMatrix(2, expected, result);
Assert.GreaterOrEqual(cm.PerClassMatrices[0].FScore, 0.8);
Assert.GreaterOrEqual(cm.PerClassMatrices[1].FScore, 0.9);
instance.Save("Result");
SvmModelStorageFactory storageFactory = new SvmModelStorageFactory(
loggerFactory,
vectorSource,
documentReconstructor,
new StorageConfig {
Location = TestContext.CurrentContext.TestDirectory
});
instance = (SvmModelStorage)storageFactory.Construct("Result");
result = pageDetector.Predict(new DocumentBlock(documents).Pages).Select(item => item ? 1 : 0).ToArray();
cm = new GeneralConfusionMatrix(2, expected: expected, predicted: result);
Assert.GreaterOrEqual(cm.PerClassMatrices[0].FScore, 0.8);
Assert.GreaterOrEqual(cm.PerClassMatrices[1].FScore, 0.9);
}
