static void Main(string[] args)
{
Trace.Listeners.Add(new ConsoleTraceListener());
Trace.TraceInformation("");
string trainingSetPathConfig = ConfigurationManager.AppSettings["TrainingSetPath"];
string testingSetPathConfig = ConfigurationManager.AppSettings["TestingSetPath"];
string numberOfModelsConfig = ConfigurationManager.AppSettings["NumberOfModels"];
string randomSeedConfig = ConfigurationManager.AppSettings["RandomSeed"];
int? randomSeed = null;
if (String.IsNullOrWhiteSpace(trainingSetPathConfig))
{
trainingSetPathConfig = "TrainingRatings.txt";
Trace.TraceInformation("TrainingSetPath in config file is not set. Default to {0} in current directory.", trainingSetPathConfig);
}
if (String.IsNullOrWhiteSpace(testingSetPathConfig))
{
testingSetPathConfig = "TestingRatings.txt";
Trace.TraceInformation("TestingSetPath in config file is not set. Default to {0} in current directory.", testingSetPathConfig);
}
if (String.IsNullOrWhiteSpace(numberOfModelsConfig))
{
numberOfModelsConfig = "1";
Trace.TraceInformation("NumberOfModels in config file is not set. Default to {0}.", numberOfModelsConfig);
}
if (String.IsNullOrWhiteSpace(randomSeedConfig))
{
randomSeed = null;
Trace.TraceInformation("RandomSeed in config file is not set. Default to null.");
}
else
{
randomSeed = Int32.Parse(randomSeedConfig);
}
Diabetes.Run(trainingSetPathConfig, testingSetPathConfig, Int32.Parse(numberOfModelsConfig), randomSeed);
Trace.TraceInformation("");
if (System.Diagnostics.Debugger.IsAttached)
{
Console.WriteLine("Done. Press enter to continue.");
Console.Read();
}
}
private static List <double[]> CalculateBucketBoundaries(List <double[]> data)
{
// Calculate the bucket boundaries for each attribute except the target attribute
List <double[]> allBucketBoundaries = new List <double[]>();
for (int i = 0; i < Diabetes.NumAttributes - 1; i++)
{
double[] attributeData = data.Select(x => x[i]).ToArray();
double[] bucketBoundaries = Diabetes.BucketBoundaries(Diabetes.NumBuckets, attributeData);
allBucketBoundaries.Add(bucketBoundaries);
}
return(allBucketBoundaries);
}
public static void Run(string trainingSetPath, string testingSetPath, int numberOfModels, int?randomSeed)
{
Trace.TraceInformation("Starting bagging");
Trace.TraceInformation("TrainingSetPath: {0}", trainingSetPath);
Trace.TraceInformation("TestingSetPath: {0}", testingSetPath);
Trace.TraceInformation("Models: {0}", numberOfModels);
Instances trainingInstances = Diabetes.LoadData(trainingSetPath, Mode.Train);
Bagging bagging = new Bagging();
bagging.Train(trainingInstances, numberOfModels, randomSeed);
Instances testingInstances = Diabetes.LoadData(testingSetPath, Mode.Test);
bagging.TestNonBagging(testingInstances);
bagging.Test(testingInstances);
}
public static void GenerateNaiveBayesInputFiles(string trainingInputFile, string testingInputFile, string trainingOutputFile, string testingOutputFile)
{
// We don't want to change the naive bayes implementation for spam detection. So, let target attribute 0 = "ham" and 1 = "spam"
// Training Data
Instances trainingInstances = Diabetes.LoadData(trainingInputFile, Mode.Train, Category.Buckets);
int targetAttributeIndex = Diabetes.NumAttributes - 1;
StringBuilder trainingOutput = new StringBuilder();
for (int i = 0; i < trainingInstances.numInstances(); i++)
{
trainingOutput.AppendFormat("{0} {1}", i, trainingInstances.instance(i).value(targetAttributeIndex) == 0 ? "ham" : "spam");
for (int j = 0; j < trainingInstances.numAttributes() - 1; j++)
{
trainingOutput.AppendFormat(" {0}_{1} {2}", j, trainingInstances.instance(i).value(j), 1);
//trainingOutput.AppendFormat(" {0} {1}", j, trainingInstances.instance(i).value(j));
}
trainingOutput.AppendLine();
}
File.WriteAllText(trainingOutputFile, trainingOutput.ToString());
// Testing Data
Instances testingInstances = Diabetes.LoadData(testingInputFile, Mode.Test, Category.Buckets);
StringBuilder testingOutput = new StringBuilder();
for (int i = 0; i < testingInstances.numInstances(); i++)
{
testingOutput.AppendFormat("{0} {1}", i, testingInstances.instance(i).value(targetAttributeIndex) == 0 ? "ham" : "spam");
for (int j = 0; j < testingInstances.numAttributes() - 1; j++)
{
testingOutput.AppendFormat(" {0}_{1} {2}", j, testingInstances.instance(i).value(j), 1);
//testingOutput.AppendFormat(" {0} {1}", j, testingInstances.instance(i).value(j));
}
testingOutput.AppendLine();
}
File.WriteAllText(testingOutputFile, testingOutput.ToString());
}
public static Instances LoadData(string filePath, Mode mode, Category category = Category.Buckets)
{
Trace.TraceInformation("Loading {0} data from {1}", Enum.GetName(mode.GetType(), mode), filePath);
List <double[]> data = new List <double[]>();
// Read in each row
using (StreamReader sr = File.OpenText(filePath))
{
string s = String.Empty;
while ((s = sr.ReadLine()) != null)
{
string[] parts = s.Split(Diabetes.Delimiter);
double[] row = new double[Diabetes.NumAttributes];
for (int i = 0; i < parts.Length; i++)
{
row[i] = Double.Parse(parts[i]);
}
data.Add(row);
}
}
// Set the attribute values, add to the Instances object
Instances instances;
if (category == Category.Buckets)
{
instances = Diabetes.DefineBucketAttributes(Diabetes.NumBuckets);
// Calculate the bucket boundaries for each attribute except the target attribute
if (mode == Mode.Train)
{
Diabetes.AllBucketBoundaries = Diabetes.CalculateBucketBoundaries(data);
}
// Put each value into a quartile/bucket
for (int i = 0; i < data.Count; i++)
{
Instance instance = new Instance(Diabetes.NumAttributes);
instance.setDataset(instances);
for (int j = 0; j < Diabetes.NumAttributes - 1; j++)
{
// Figure out which quartile/bucket the value should get dropped into
double[] bucketBoundaries = Diabetes.AllBucketBoundaries[j];
for (int k = 0; k < bucketBoundaries.Length; k++)
{
double value = data[i][j];
if (k == 0 && value <= bucketBoundaries[k])
{
// bucket k
instance.setValue(j, k.ToString());
break;
}
else if (k == bucketBoundaries.Length - 1)
{
// bucket k+1
instance.setValue(j, (k + 1).ToString());
break;
}
else if (value > bucketBoundaries[k] && value <= bucketBoundaries[k + 1])
{
// bucket k+1
instance.setValue(j, (k + 1).ToString());
break;
}
}
}
Diabetes.ValueForDiabetes(instance, Diabetes.NumAttributes - 1, data[i][Diabetes.NumAttributes - 1]);
instances.add(instance);
}
}
else
{
instances = Diabetes.DefineCategoricalAttributes();
foreach (double[] row in data)
{
Instance instance = new Instance(Diabetes.NumAttributes);
instance.setDataset(instances);
Diabetes.ValueForNumberOfTimesPregnant(instance, 0, row[0]);
Diabetes.ValueForPlasmaGlucoseConcentrationt(instance, 1, row[1]);
Diabetes.ValueForDiastolicBloodPressure(instance, 2, row[2]);
Diabetes.ValueForTricepsSkinFoldThickness(instance, 3, row[3]);
Diabetes.ValueForTwoHourSerumInsulin(instance, 4, row[4]);
Diabetes.ValueForBmi(instance, 5, row[5]);
Diabetes.ValueForDiabetesPedigreeFunction(instance, 6, row[6]);
Diabetes.ValueForAge(instance, 7, row[7]);
Diabetes.ValueForDiabetes(instance, 8, row[8]);
instances.add(instance);
}
}
return(instances);
}
