public static void RunNonBagging(string trainingSetPath, string testingSetPath, int maxTreeDepth)
{
Trace.TraceInformation("Starting Bias-Variance for NonBagging");
Trace.TraceInformation("TrainingSetPath: {0}", trainingSetPath);
Trace.TraceInformation("TestingSetPath: {0}", testingSetPath);
Trace.TraceInformation("Max Tree Depth: {0}", maxTreeDepth);
// Train
Instances trainingInstances = Diabetes.LoadData(trainingSetPath, Mode.Train);
Bagging.Bagging bagging = new Bagging.Bagging();
bagging.Train(trainingInstances, 0, null, maxTreeDepth);
// Predict
Instances testingInstances = Diabetes.LoadData(testingSetPath, Mode.Test);
List <int> predictions = bagging.TestNonBagging(testingInstances);
// Calculate Bias and Variance
// Get the class for each test example
int targetAttribute = testingInstances.numAttributes() - 1;
List <int> classes = new List <int>();
for (int i = 0; i < testingInstances.numInstances(); i++)
{
classes.Add((int)testingInstances.instance(i).value(targetAttribute));
}
// Get the prediction for every test example, for every set of instanes
List <List <int> > allPredictions = new List <List <int> >();
for (int j = 0; j < 1; j++)
{
for (int i = 0; i < testingInstances.numInstances(); i++)
{
if (j == 0)
{
allPredictions.Add(new List <int>());
}
allPredictions[i].Add(predictions[i]);
}
}
BiasVariance.biasvar(classes, allPredictions, testingInstances.numInstances(), 1);
}
public static void RunBagging(string trainingSetPath, string testingSetPath, int numberOfModels, int bootstrapSamples, int?randomSeed, int maxTreeDepth)
{
Trace.TraceInformation("Starting Bias-Variance for Bagging");
Trace.TraceInformation("TrainingSetPath: {0}", trainingSetPath);
Trace.TraceInformation("TestingSetPath: {0}", testingSetPath);
Trace.TraceInformation("Models: {0}", numberOfModels);
Trace.TraceInformation("Bootstrap Samples: {0}", bootstrapSamples);
Trace.TraceInformation("Random Seed: {0}", randomSeed.ToString());
Trace.TraceInformation("Max Tree Depth: {0}", maxTreeDepth);
Instances trainingInstances = Diabetes.LoadData(trainingSetPath, Mode.Train);
Instances testingInstances = Diabetes.LoadData(testingSetPath, Mode.Test);
// Stores the predictions for every training set
List <List <int> > allPredictions = new List <List <int> >();
for (int i = 0; i < bootstrapSamples; i++)
{
Bagging.Bagging bagging = new Bagging.Bagging();
bagging.Train(trainingInstances, numberOfModels, randomSeed, maxTreeDepth);
List <int> predictions = bagging.Test(testingInstances);
for (int k = 0; k < testingInstances.numInstances(); k++)
{
if (i == 0)
{
allPredictions.Add(new List <int>());
}
allPredictions[k].Add(predictions[k]);
}
}
// Calculate Bias and Variance
// Get the class for each test example
int targetAttribute = testingInstances.numAttributes() - 1;
List <int> classes = new List <int>();
for (int i = 0; i < testingInstances.numInstances(); i++)
{
classes.Add((int)testingInstances.instance(i).value(targetAttribute));
}
BiasVariance.biasvar(classes, allPredictions, testingInstances.numInstances(), bootstrapSamples);
}
public static void Main(string[] args)
{
Trace.Listeners.Add(new ConsoleTraceListener());
string trainingFile = @"C:\coding\ml\data\diabetes\diabetes_libsvmformat_train.txt";
string testingFile = @"C:\coding\ml\data\diabetes\diabetes_libsvmformat_test.txt";
int bootstrapSamples = 10;
int kernel = 3;
int? randomSeed = null;
// Read in the training data
List <string> trainingData = new List <string>();
using (StreamReader sr = File.OpenText(trainingFile))
{
string s = String.Empty;
while ((s = sr.ReadLine()) != null)
{
if (!String.IsNullOrWhiteSpace(s))
{
trainingData.Add(s);
}
}
}
// Read in the testing data
List <int> testingTargetValues = new List <int>();
using (StreamReader sr = File.OpenText(testingFile))
{
string s = String.Empty;
while ((s = sr.ReadLine()) != null)
{
if (!String.IsNullOrWhiteSpace(s))
{
int value = Int32.Parse(s.Split(' ')[0]);
// Map -1 to 0
testingTargetValues.Add(value == -1 ? 0 : value);
}
}
}
// Stores the predictions for every sampled training set
List <List <int> > allPredictions = new List <List <int> >();
for (int i = 0; i < bootstrapSamples; i++)
{
List <int> predictions = Program.RunSvm(trainingData, testingFile, kernel, randomSeed);
for (int k = 0; k < testingTargetValues.Count; k++)
{
if (i == 0)
{
allPredictions.Add(new List <int>());
}
// Map -1 to 0
allPredictions[k].Add(predictions[k] == -1 ? 0 : predictions[k]);
}
}
// Calculate bias and variance
BiasVariance.biasvar(testingTargetValues, allPredictions, testingTargetValues.Count, bootstrapSamples);
Trace.TraceInformation("");
if (System.Diagnostics.Debugger.IsAttached)
{
Console.WriteLine("Done. Press enter to continue.");
Console.Read();
}
}
