Native C# multithreaded version of LIBSVM
Original code is taken from one of LIBSVM ports, fixed some small bugs + added multithreading for better performance
Simple training example
SvmTrainClient train = new SvmTrainClient(arff);
var model = await train.Train(header, token);
Full training example
var arff = ArffDataSet.CreateSimple("subjectivity");
arff.Header.RegisterNominalClass(labels);
foreach (var definition in documentSet.Document.Where(item => item.Labels.Length > 0))
{
var label = definition.Labels.Last();
var review = dataHolder.AddReview();
review.Class.Value = label;
foreach (var record in definition.WordsTable)
{
review.AddRecord(record.Key).Value = record.Value;
}
}
arff.CompactHeader(3);
await arff.CompactReviews(3);
arff.CompactClass(3);
if (arff.Reviews.Length < 10)
{
throw new LearningException("Not enough documents to learn patterns");
}
arff.Normalize(header.Normalization);
SvmTrainClient train = new SvmTrainClient(arff);
var model = await train.Train(header, token);
Grid search to find best parameters:
var training = new TrainingModel();
var taskFactory = new TaskFactory(new LimitedConcurrencyLevelTaskScheduler(2));
var parameters = new GridSearchParameters(5, new double[] { 1, 2, 3, 4 }, new double[] { 1, 2, 3, 4 }, new Parameter());
var instance = new GridParameterSelection(taskFactory, training, parameters);
var dataSet = ArffDataSet.CreateSimple("Test");
dataSet.Header.RegisterNominalClass("One", "Two", "Three");
dataSet.UseTotal = true;
var one = dataSet.AddDocument();
one.Class.Value = "One";
one.AddRecord("Good");
problem = dataSet.GetProblem();
var result = await instance.Find(problem, CancellationToken.None);