private void buttonRun_Click(object sender, EventArgs e)
{
try
{
var documentDataList = documentDataBusinessLogic.GetDocumentDataForMultipleXmlFiles(filepathsToUseForDocumentData);
var lists = SplitListIntoTwoSeparateLists(documentDataList, 70);
var listForTraining = lists.Item1;
var listForValidation = lists.Item2;
var datasetRepresentationTraining = documentDataList.ToDatasetRepresentation();
datasetRepresentationTraining = datasetRepresentationTraining.ReconstructByEliminatingWordsBelowAndAboveThresholds(5, 95);
var stopwatch = new Stopwatch();
stopwatch.Start();
var features = featureSelector.GetMostImportantWords(datasetRepresentationTraining);
stopwatch.Stop();
var featuresJson = JsonConvert.SerializeObject(features);
File.WriteAllText("features.json", featuresJson);
datasetRepresentationTraining = datasetRepresentationTraining.ReconstructByKeepingOnlyTheseWords(features);
var datasetJson = JsonConvert.SerializeObject(datasetRepresentationTraining);
var datasetArff = datasetRepresentationTraining.ToArffFileFormat();
File.WriteAllText("dataset.json", datasetJson);
File.WriteAllText("dataset.arff", datasetArff);
topicPredictor.Train(datasetRepresentationTraining);
double total = listForValidation.Count;
var successfullyPredicted = 0;
foreach (var documentData in listForValidation)
{
var predictedTopic = topicPredictor.PredictTopic(documentData);
if (documentData.Topics.Contains(predictedTopic))
{
successfullyPredicted++;
}
}
var accuracy = successfullyPredicted / total * 100;
//documentDataDisplayUserControl.DisplayDocumentData(documentData);
SetStatusLabel("Done", Color.GreenYellow);
}
catch (Exception exception)
{
SetStatusLabel("Error", Color.Red);
//textBoxResult.Text = exception.ToString();
}
}
public List <ClassEvaluationResult> EvaluateTopicPredictor(ITopicPredictor topicPredictor, List <DocumentData> documentDataList)
{
ArgumentValidator.ValidateObject(topicPredictor);
ArgumentValidator.ValidateNotEmptyList(documentDataList);
var evaluationResults = new Dictionary <string, ClassEvaluationResult>();
double total = documentDataList.Count;
var successfullyPredicted = 0;
var distinctTopics = documentDataList.ToDatasetRepresentation().GetAllDistinctTopics();
foreach (var topic in distinctTopics)
{
evaluationResults.Add(topic, new ClassEvaluationResult());
}
var predicted = new int[documentDataList.Count(x => x.Topics.Count > 0)];
var expected = new int[documentDataList.Count(x => x.Topics.Count > 0)];
foreach (var documentData in documentDataList.Where(x => x.Topics.Count > 0))
{
var predictedTopic = topicPredictor.PredictTopic(documentData);
var expectedTopic = documentData.Topics[0];
evaluationResults.TryGetValue(predictedTopic, out var resultForPredictedTopic);
evaluationResults.TryGetValue(expectedTopic, out var resultForExpectedTopic);
if (resultForPredictedTopic == null && resultForExpectedTopic == null)
{
continue;
}
if (predictedTopic != expectedTopic)
{
if (resultForPredictedTopic != null)
{
resultForPredictedTopic.FalsePositives++;
}
if (resultForExpectedTopic != null)
{
resultForExpectedTopic.FalseNegatives++;
}
}
else
{
if (resultForPredictedTopic != null)
{
resultForPredictedTopic.TruePositives++;
}
}
var remainingTopics = distinctTopics
.Where(x => x != predictedTopic && x != expectedTopic)
.ToList();
foreach (var remainingTopic in remainingTopics)
{
if (evaluationResults.TryGetValue(remainingTopic, out var resultForRemainingTopic))
{
resultForRemainingTopic.TrueNegatives++;
}
}
}
return(evaluationResults.Values.ToList());
}