//classifies the dataset with ID = id
public ActionResult Index(int id)
{
//remove any data from the trees and nodes tables
foreach(var t in db.Trees.ToList())
{
db.Trees.Remove(t);
}
foreach(var n in db.Nodes.ToList())
{
db.Nodes.Remove(n);
}
db.SaveChanges();
Models.DataSet dataset = db.DataSets.Single(i => i.ID == id);
List<Instance> instances = db.Instances.Where(i => i.DataSetID == dataset.ID).ToList();
List<Tree> decisionTrees = new List<Tree>();
List<double> successRates = new List<double>();
List<string> correctlyClassifiedInstances = new List<string>();
Tree decisionTree;
//stringwriter for exporting to CSV
results = new StringWriter();
results.WriteLine("\"Test No.\",\"Predicted Class\",\"Actual Class\"");
DataTable dataTable = new DataTable();
dataTable.Columns.Add(new DataColumn("ID"));
//performs the following sequence 10 times, as specified
for (int j = 0; j < 10; j++)
{
decisionTree = new Tree { Name = dataset.Name + " Tree" };
db.Trees.Add(decisionTree);
db.SaveChanges();
decisionTrees.Add(decisionTree);
//generates testing and training lists from the datasets instances: 1/3 testing and 2/3 training
int noOfTestingDataInstances = instances.Count / 3;
List<Instance> testingInstances = new List<Instance>();
List<Instance> trainingInstances = new List<Instance>();
for (int i = 0; i < noOfTestingDataInstances; i++)
{
var x = rnd.Next(instances.Count);
testingInstances.Add(instances.ElementAt(rnd.Next(instances.Count)));
}
trainingInstances = instances.Where(i => testingInstances.Contains(i) == false).ToList();
//generates a decision tree based on the training instances
attributesUsedSoFar = new List<string>();
CreateDecisionTree(trainingInstances, new Node { EdgeValue = null }, decisionTree.ID, String.Empty, null);
//classifies each test instance based on the current decision tree, and computes a list of success rates
double successes = 0;
double failures = 0;
foreach (var i in testingInstances)
{
string result = classifyInstance(i, decisionTree);
results.WriteLine("\"{0}\",\"{1}\",\"{2}\"", j+1, result, i.Class);
if (result == i.Class)
successes++;
else
failures++;
}
results.WriteLine();
double successRate = successes / testingInstances.Count;
successRate = Math.Round(successRate, 6);
successRates.Add(successRate * 100);
Math.Round(successes, 0);
correctlyClassifiedInstances.Add(successes.ToString()+"/"+(testingInstances.Count.ToString()));
}
//outputs the success rates and the decision trees nodes to the view
ViewBag.SuccessRates = successRates;
ViewBag.CorrectlyClassifiedInstances = correctlyClassifiedInstances;
ViewBag.TestInstances = (instances.Count / 3).ToString();
ViewBag.TrainingInstances = (instances.Count - (instances.Count / 3)).ToString();
ViewBag.TotalInstances = instances.Count.ToString();
ViewBag.AverageSuccessRate = Math.Round(successRates.Average(),2);
int decisionTreeId = decisionTrees.ElementAt(successRates.IndexOf(successRates.Max())).ID;
List<Node> decisionTreeNodes = db.Nodes.Where(i => i.TreeID == decisionTreeId).OrderBy(i=>i.ID).ThenBy(i=>i.ParentID).ToList();
return View(decisionTreeNodes);
}
//classifies an instance - returns "" if the instance cannot be classified
private string classifyInstance(Instance instance, Tree tree)
{
Node selectedNode = db.Nodes.Single(i => i.ParentID < 1 && i.TreeID == tree.ID);
while (db.Nodes.Where(i => i.ParentID == selectedNode.ID).Count() > 0)
{
Boolean result = false;
List<Node> childNodes = db.Nodes.Where(i => i.ParentID == selectedNode.ID).ToList();
int j = 0;
while (result != true)
{
if (j >= childNodes.Count)
return ""; // this is the case where the instance cannot be classified, this can happen occasionally with nominal data
Node childNode = childNodes.ElementAt(j);
result = checkBranch(instance, selectedNode, childNode);
if (result == true)
{
selectedNode = childNodes.ElementAt(j);
}
j++;
}
}
return selectedNode.Attribute;
}
