public void TestRegression()
{
var dataTable = new DataTableBuilder();
dataTable.AddColumn(ColumnType.Float, "value");
dataTable.AddColumn(ColumnType.Float, "result", true);
// simple linear relationship: result is twice value
dataTable.AddRow(new object[] { 1f, 2f });
dataTable.AddRow(new object[] { 2f, 4f });
dataTable.AddRow(new object[] { 4f, 8f });
dataTable.AddRow(new object[] { 8f, 16f });
var index = dataTable.Build();
var classifier = index.CreateLinearRegressionTrainer(_lap);
//var theta = classifier.Solve();
//var predictor = theta.CreatePredictor(_lap);
//var prediction = predictor.Predict(3f);
//Assert.IsTrue(Math.Round(prediction) == 6f);
var theta = classifier.GradientDescent(20, 0.01f);
var predictor = theta.CreatePredictor(_lap);
var prediction = predictor.Predict(3f);
Assert.IsTrue(Math.Round(prediction) == 6f);
var prediction3 = predictor.Predict(new[] {
new float[] { 10f },
new float[] { 3f }
});
Assert.IsTrue(Math.Round(prediction3[1]) == 6f);
}
/// <summary>
/// Creates a new data table with the vectorised weighted classification outputs linked with each row's classification
/// </summary>
/// <param name="output">The output stream to write the table to (optional)</param>
public IDataTable GetTable(Stream output = null)
{
if (_trainingData.Any())
{
var fieldCount = _trainingData.First().Item1.Length;
var builder = new DataTableBuilder();
for (var i = 0; i < fieldCount; i++)
{
builder.AddColumn(ColumnType.Float, "v" + i);
}
builder.AddColumn(ColumnType.String, "target", true);
foreach (var item in _trainingData)
{
var data = new object[fieldCount + 1];
for (var i = 0; i < fieldCount; i++)
{
data[i] = item.Item1[i];
}
data[fieldCount] = item.Item2;
builder.AddRow(data);
}
return(builder.Build(output));
}
return(null);
}
/// <summary>
/// Converts the sparse vectors to a data table
/// </summary>
/// <param name="stream">Optional stream to write the data table to</param>
public IDataTable ConvertToTable(Stream stream = null)
{
var max = GetMaximumIndex();
var dataTable = new DataTableBuilder();
for (var i = 0; i < max; i++)
{
dataTable.AddColumn(ColumnType.Float, "term " + i.ToString());
}
dataTable.AddColumn(ColumnType.String, "classification", true);
foreach (var item in Classification)
{
var data = new object[max + 1];
for (var i = 0; i < max; i++)
{
data[i] = 0f;
}
foreach (var index in item.Data)
{
data[index.Index] = index.Weight;
}
data[max] = item.Name;
dataTable.AddRow(data);
}
return(dataTable.Build(stream));
}
public void AddRow(IEnumerable <object> row)
{
_dataTable.AddRow(row);
++_rowCount;
if ((_rowCount % DataTable.BLOCK_SIZE) == 0)
{
_Write();
}
}
public IRow AddRow(IRow row)
{
var ret = _dataTableBuilder.AddRow(row);
++_rowCount;
if (_rowCount % _dataTableBuilder.BlockSize == 0)
{
_Write();
}
return(ret);
}
public IRow AddRow(IRow row)
{
var ret = _dataTableBuilder.AddRow(row);
++_rowCount;
if ((_rowCount % DataTable.BLOCK_SIZE) == 0)
{
_Write();
}
return(ret);
}
public void TestLogisticRegression()
{
var dataTable = new DataTableBuilder();
dataTable.AddColumn(ColumnType.Float, "hours");
dataTable.AddColumn(ColumnType.Boolean, "pass", true);
// sample data from: https://en.wikipedia.org/wiki/Logistic_regression
dataTable.AddRow(new object[] { 0.5f, false });
dataTable.AddRow(new object[] { 0.75f, false });
dataTable.AddRow(new object[] { 1f, false });
dataTable.AddRow(new object[] { 1.25f, false });
dataTable.AddRow(new object[] { 1.5f, false });
dataTable.AddRow(new object[] { 1.75f, false });
dataTable.AddRow(new object[] { 1.75f, true });
dataTable.AddRow(new object[] { 2f, false });
dataTable.AddRow(new object[] { 2.25f, true });
dataTable.AddRow(new object[] { 2.5f, false });
dataTable.AddRow(new object[] { 2.75f, true });
dataTable.AddRow(new object[] { 3f, false });
dataTable.AddRow(new object[] { 3.25f, true });
dataTable.AddRow(new object[] { 3.5f, false });
dataTable.AddRow(new object[] { 4f, true });
dataTable.AddRow(new object[] { 4.25f, true });
dataTable.AddRow(new object[] { 4.5f, true });
dataTable.AddRow(new object[] { 4.75f, true });
dataTable.AddRow(new object[] { 5f, true });
dataTable.AddRow(new object[] { 5.5f, true });
var index = dataTable.Build();
var trainer = index.CreateLogisticRegressionTrainer(_lap);
var theta = trainer.GradientDescent(1000, 0.1f);
var predictor = theta.CreatePredictor(_lap);
var probability1 = predictor.Predict(2f);
Assert.IsTrue(probability1 < 0.5f);
var probability2 = predictor.Predict(4f);
Assert.IsTrue(probability2 >= 0.5f);
var probability3 = predictor.Predict(new[] {
new float[] { 1f },
new float[] { 2f },
new float[] { 3f },
new float[] { 4f },
new float[] { 5f },
});
Assert.IsTrue(probability3[0] <= 0.5f);
Assert.IsTrue(probability3[1] <= 0.5f);
Assert.IsTrue(probability3[2] >= 0.5f);
Assert.IsTrue(probability3[3] >= 0.5f);
Assert.IsTrue(probability3[4] >= 0.5f);
}
public void AddRow()
{
_builder.AddRow();
}
