libsvm ported to C#
Install-Package cs-libsvmThe sample codes below shows how to do binary classification using SVC:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
namespace LibSvmSharp
{
class SVCDemo
{
private static Random random = new Random();
public static void SolveBinaryClassificationProblem()
{
SVC svc = new SVC();
List<KeyValuePair<double[], bool>> data = new List<KeyValuePair<double[], bool>>();
for(int i=0; i < 100; ++i)
{
double[] x = new double[2];
x[0] = i / 50.0;
x[1] = (i+1) / 100.0;
bool y = Math.Sin((x[0] + x[1]) * Math.PI) > 0;
data.Add(new KeyValuePair<double[], bool>(x, y));
}
var tuple = TrainTestSplit(data);
var train_data = tuple.Item1;
var test_data = tuple.Item2;
var metric = svc.Fit(train_data, test_data);
Console.WriteLine("Train Acc: {0}, Test Acc: {1}", metric.TrainAccuracy, metric.TestAccuracy);
foreach(KeyValuePair<double[], bool> entry in test_data.GetRange(0, 10)) {
double[] x = entry.Key;
bool y = entry.Value;
double output = svc.Predict(x);
bool prediction = output > 0;
Console.WriteLine("Predicted: {0} Actual: {1}", prediction, y);
}
}
private static Tuple<List<T>, List<T>> TrainTestSplit<T>(List<T> a, double test_size=0.2)
{
Shuffle(a);
int splitIndex = (int)(a.Count * test_size);
List<T> test_data = new List<T>();
List<T> train_data = new List<T>();
for(int i=0; i < a.Count; ++i)
{
if(i < splitIndex)
{
test_data.Add(a[i]);
} else
{
train_data.Add(a[i]);
}
}
return new Tuple<List<T>, List<T>>(train_data, test_data);
}
private static void Shuffle<T>(List<T> a)
{
int i = 0;
while(i < a.Count)
{
int j = random.Next(i + 1);
Swap(a, i++, j);
}
}
private static void Swap<T>(List<T> a, int i, int j)
{
T temp = a[i];
a[i] = a[j];
a[j] = temp;
}
}
}The sample codes below shows how to do regression using SVR:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
namespace LibSvmSharp
{
class SVRDemo
{
private static Random random = new Random();
public static void SolveRegressionProblem()
{
SVR svr = new SVR();
List<KeyValuePair<double[], double>> data = new List<KeyValuePair<double[], double>>();
for(int i=0; i < 100; ++i)
{
double[] x = new double[2];
x[0] = i / 50.0;
x[1] = (i+1) / 100.0;
double y = Math.Sin((x[0] + x[1]) * Math.PI);
data.Add(new KeyValuePair<double[], double>(x, y));
}
var tuple = TrainTestSplit(data);
var train_data = tuple.Item1;
var test_data = tuple.Item2;
var metric = svr.Fit(train_data, test_data);
Console.WriteLine("Train MSE: {0}, Test MSE: {1}", metric.TrainMSE, metric.TestMSE);
foreach(KeyValuePair<double[], double> entry in test_data.GetRange(0, 10)) {
double[] x = entry.Key;
double y = entry.Value;
double output = svr.Predict(x);
Console.WriteLine("Predicted: {0} Actual: {1}", output, y);
}
}
private static Tuple<List<T>, List<T>> TrainTestSplit<T>(List<T> a, double test_size=0.2)
{
Shuffle(a);
int splitIndex = (int)(a.Count * test_size);
List<T> test_data = new List<T>();
List<T> train_data = new List<T>();
for(int i=0; i < a.Count; ++i)
{
if(i < splitIndex)
{
test_data.Add(a[i]);
} else
{
train_data.Add(a[i]);
}
}
return new Tuple<List<T>, List<T>>(train_data, test_data);
}
private static void Shuffle<T>(List<T> a)
{
int i = 0;
while(i < a.Count)
{
int j = random.Next(i + 1);
Swap(a, i++, j);
}
}
private static void Swap<T>(List<T> a, int i, int j)
{
T temp = a[i];
a[i] = a[j];
a[j] = temp;
}
}
}The sample codes below shows how to do one-class SVM for outlier detection:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
namespace LibSvmSharp
{
class OneClassSVMDemo
{
private static Random _random = new Random();
public static void SolveOutlierDetectionProblem()
{
List<KeyValuePair<double[], double>> trainingBatch = new List<KeyValuePair<double[], double>>();
// add some normal data
for (int i = 0; i < 100; ++i)
{
trainingBatch.Add(new KeyValuePair<double[], double>(new double[] { randn() * 0.3 + 2, randn() * 0.3 + 2 }, -1.0));
trainingBatch.Add(new KeyValuePair<double[], double>(new double[] { randn() * 0.3 - 2, randn() * 0.3 - 2 }, -1.0));
}
List<KeyValuePair<double[], double>> crossValidationBatch = new List<KeyValuePair<double[], double>>();
// add some validation data
for (int i = 0; i < 20; ++i)
{
crossValidationBatch.Add(new KeyValuePair<double[], double>(new double[] { randn() * 0.3 + 2, randn() * 0.3 + 2 }, -1.0));
crossValidationBatch.Add(new KeyValuePair<double[], double>(new double[] { randn() * 0.3 - 2, randn() * 0.3 - 2 }, -1.0));
}
List<KeyValuePair<double[], double>> outliers = new List<KeyValuePair<double[], double>>();
// add some outliers data
for (int i = 0; i < 20; ++i)
{
outliers.Add(new KeyValuePair<double[], double>(new double[] { rand(-4, 4), rand(-4, 4) }, 1.0));
outliers.Add(new KeyValuePair<double[], double>(new double[] { rand(-4, 4), rand(-4, 4) }, 1.0));
}
OneClassSVM algorithm = new OneClassSVM();
algorithm.set_gamma(0.1);
algorithm.set_nu(0.1);
var batches = from p in trainingBatch
select p.Key;
algorithm.Fit(batches.ToList());
for (int i = 0; i < crossValidationBatch.Count; ++i)
{
double predicted = algorithm.IsOutlier(crossValidationBatch[i].Key) ? 1.0 : -1.0;
Console.WriteLine("predicted: " + predicted + "\texpected: " + crossValidationBatch[i].Value);
}
for (int i = 0; i < outliers.Count; ++i)
{
double predicted = algorithm.IsOutlier(crossValidationBatch[i].Key) ? 1.0 : -1.0;
Console.WriteLine("predicted: " + predicted + "\texpected: " + outliers[i].Value);
}
}
public static double randn()
{
return _random.NextDouble();
}
public static double rand(double lower, double upper)
{
return randn() * (upper - lower) + lower;
}
private List<double[]> GetInputVectors(List<KeyValuePair<double[], double>> samples)
{
List<double[]> inputVectors = new List<double[]>();
foreach (KeyValuePair<double[], double> point in samples)
{
inputVectors.Add(point.Key);
}
return inputVectors;
}
}
}