public void WriteLinearMachineTest()
{
MemoryStream destination = new MemoryStream();
var model = new LibSvmModel()
{
Bias = -1,
Classes = 2,
Dimension = 123,
Labels = new[] { +1, -1 },
Solver = LibSvmSolverType.L1RegularizedLogisticRegression,
Weights = a9a_weights
};
model.Save(destination);
destination.Seek(0, SeekOrigin.Begin);
TextReader textReader = new StreamReader(destination);
string[] actual = textReader.ReadToEnd()
.Split(new[] { Environment.NewLine }, StringSplitOptions.None);
string[] expected = Resources.L1R_LR_a9a
.Split(new[] { "\r\n" }, StringSplitOptions.None);
Assert.AreEqual(expected.Length, actual.Length);
for (int i = 0; i < 7; i++)
Assert.AreEqual(expected[i], actual[i]);
for (int i = 6; i < expected.Length; i++)
{
if (expected[i] == actual[i])
continue;
double a = Double.Parse(expected[i], CultureInfo.InvariantCulture);
double b = Double.Parse(actual[i], CultureInfo.InvariantCulture);
Assert.AreEqual(a, b);
}
Assert.AreEqual(expected[expected.Length - 1], String.Empty);
}
/// <summary>
/// Creates a <see cref="LibSvmModel"/> from an existing <see cref="SupportVectorMachine{Linear}"/>.
/// </summary>
///
/// <param name="svm">The vector machine from which a libSVM model definition should be created.</param>
///
/// <returns>
/// A <see cref="LibSvmModel"/> class representing a support vector machine in LibSVM format.
/// </returns>
///
public static LibSvmModel FromMachine(SupportVectorMachine <Linear> svm)
{
var model = new LibSvmModel()
{
Solver = LibSvmSolverType.Unknown,
Bias = -1,
NumberOfClasses = svm.NumberOfClasses,
NumberOfInputs = svm.NumberOfInputs + 1,
Labels = new[] { 1, -1 }
};
if (svm.SupportVectors.Length == 1)
{
model.Weights = svm.Threshold.Concatenate(svm.SupportVectors[0]);
}
else
{
model.Weights = svm.Weights;
model.Vectors = svm.SupportVectors.InsertColumn(svm.Threshold, index: 0);
}
return(model);
}
/// <summary>
/// Loads a model specified using LibSVM's model format from a stream.
/// </summary>
///
/// <param name="stream">The stream from where the model should be loaded.</param>
///
/// <returns>The <see cref="LibSvmModel"/> stored on <paramref name="stream"/>.</returns>
///
public static LibSvmModel Load(Stream stream)
{
LibSvmModel model = new LibSvmModel();
using (StreamReader reader = new StreamReader(stream))
{
while (!reader.EndOfStream)
{
string line = reader.ReadLine();
string[] words = line.Split(' ');
if (words[0] == "solver_type")
{
string key = words[1].ToUpperInvariant();
if (!model.map.TryGetValue(key, out model.type))
{
model.type = LibSvmSolverType.Unknown;
}
}
else if (words[0] == "nr_class")
{
model.Classes = Int32.Parse(words[1]);
}
else if (words[0] == "nr_feature")
{
model.Dimension = Int32.Parse(words[1]);
}
else if (words[0] == "bias")
{
model.Bias = Double.Parse(words[1], CultureInfo.InvariantCulture);
}
else if (words[0] == "w")
{
break;
}
else if (words[0] == "label")
{
model.Labels = new int[words.Length - 1];
for (int i = 1; i < words.Length; i++)
{
model.Labels[i - 1] = Int32.Parse(words[i]);
}
}
else
{
System.Diagnostics.Trace.WriteLine("Unknown field: " + words[0]);
}
}
List <double> weights = new List <double>();
while (!reader.EndOfStream)
{
string line = reader.ReadLine();
weights.Add(Double.Parse(line, CultureInfo.InvariantCulture));
}
model.Weights = weights.ToArray();
}
return(model);
}
/// <summary>
/// Loads a model specified using LibSVM's model format from a stream.
/// </summary>
///
/// <param name="stream">The stream from where the model should be loaded.</param>
///
/// <returns>The <see cref="LibSvmModel"/> stored on <paramref name="stream"/>.</returns>
///
public static LibSvmModel Load(Stream stream)
{
LibSvmModel model = new LibSvmModel();
using (StreamReader reader = new StreamReader(stream))
{
while (!reader.EndOfStream)
{
string line = reader.ReadLine();
string[] words = line.Split(' ');
if (words[0] == "solver_type")
{
string key = words[1].ToUpperInvariant();
if (!model.map.TryGetValue(key, out model.type))
model.type = LibSvmSolverType.Unknown;
}
else if (words[0] == "nr_class")
model.Classes = Int32.Parse(words[1]);
else if (words[0] == "nr_feature")
model.Dimension = Int32.Parse(words[1]);
else if (words[0] == "bias")
model.Bias = Double.Parse(words[1], CultureInfo.InvariantCulture);
else if (words[0] == "w")
break;
else if (words[0] == "label")
{
model.Labels = new int[words.Length - 1];
for (int i = 1; i < words.Length; i++)
model.Labels[i - 1] = Int32.Parse(words[i]);
}
else
{
System.Diagnostics.Trace.WriteLine("Unknown field: " + words[0]);
}
}
List<double> weights = new List<double>();
while (!reader.EndOfStream)
{
string line = reader.ReadLine();
weights.Add(Double.Parse(line, CultureInfo.InvariantCulture));
}
model.Weights = weights.ToArray();
}
return model;
}
