//public string relation = "eq";
/// <summary>
/// Initializes a new instance of the TreeNode
/// </summary>
/// <param name="attribute">attribute to which the node is connected </param>
public TreeNode(TreeAttribute attribute)
{
if (attribute != null && attribute.PossibleValues != null)
{
_children = new TreeNodeCollection();
_children2 = new TreeNodeCollection();
_childRelations = new PossibleValueRelationCollection();
_childRelations2 = new PossibleValueRelationCollection();
for (int i = 0; i < attribute.PossibleValues.Count; i++)
{
_children.Add(null);
_children2.Add(null);
_childRelations.Add("eq");
_childRelations2.Add("eq");
}
}
else
{
_children = new TreeNodeCollection();
_children.Add(null);
_children2 = new TreeNodeCollection();
_children2.Add(null);
_childRelations = new PossibleValueRelationCollection();
_childRelations2 = new PossibleValueRelationCollection();
_childRelations.Add("eq");
_childRelations2.Add("eq");
}
_attribute = attribute;
}
private void getValuesToAttributeContinous(DataTable samples, TreeAttribute attribute, string value, out int positives, out int negatives)
{
positives = 0;
negatives = 0;
double refval = double.Parse(value);
string positveRef = getUniformRefValue(samples, mTargetAttribute);
foreach (DataRow aRow in samples.Rows)
{
///To do: Figure out if this is correct - it looks bad
double rowV = double.Parse((string)aRow[attribute.AttributeName]);
if ((rowV <= refval))
{
if (aRow[mTargetAttribute].ToString().Trim().ToLower() == positveRef)
{
positives++;
}
else
{
negatives++;
}
}
}
}
/// <summary>
/// Calculate the gain of an attribute
/// </summary>
/// <param name="attribute">Attribute to be calculated </param>
/// <returns> Gain attribute </returns>
private double gain(DataTable samples, TreeAttribute attribute)
{
PossibleValueCollection values = attribute.PossibleValues;
if (isContinousSet(values))
{
double sum = 0.0;
double bsum = -9999999.0;
// return the value for the best possible split
for (int i = 0; i < values.Count; i++)
{
int positives, negatives;
positives = negatives = 0;
getValuesToAttributeContinous(samples, attribute, values[i], out positives, out negatives);
double entropy = getCalculatedEntropy(positives, negatives);
sum = -(double)(positives + negatives) / mTotal * entropy;
if (sum > bsum)
{
bsum = sum;
}
}
return(mEntropySet + bsum);
}
else
{
double sum = 0.0;
for (int i = 0; i < values.Count; i++)
{
int positives, negatives;
positives = negatives = 0;
getValuesToAttribute(samples, attribute, values[i], out positives, out negatives);
// does it really split?
int remainder = mTotal - (positives + negatives);
if (remainder > 0)
{
double entropy = getCalculatedEntropy(positives, negatives);
sum += -(double)(positives + negatives) / mTotal * entropy;
}
}
return(mEntropySet + sum);
}
}
/// <summary>
///Returns the best attribute.
/// </summary>
/// <param name="attributes"> A vector with attributes </param>
/// <returns>Returns which has higher gain </returns>
private TreeAttribute getBestAttribute(DataTable samples, TreeAttributeCollection attributes)
{
double maxGain = -9999999.0;
TreeAttribute result = null;
foreach (TreeAttribute attribute in attributes)
{
double aux = gain(samples, attribute);
if (aux > maxGain)
{
maxGain = aux;
result = attribute;
}
}
return(result);
}
public TreeAttributeCollection GetValidAttributeCollection()
{
TreeAttributeCollection returnCollection = new TreeAttributeCollection();
foreach (DataColumn column in this.Columns)
{
TreeAttribute currentAttribute = new TreeAttribute(column.ColumnName, GetValuesFromColumn(column.ColumnName));
if (returnCollection.ContainsAttribute(currentAttribute) || currentAttribute.AttributeName.ToUpper().Trim() == "RESULT")
{
continue;
}
returnCollection.Add(currentAttribute);
}
return(returnCollection);
}
/// <summary>
/// Sample table scans checking an attribute and if the result is positive or negative
/// </summary>
/// <param name="samples">DataTable with samples</param>
/// <param name="attribute"> attribute to be searched </param>
/// <param name="value"> value allowed for the attribute </param>
/// <param name="positives"> nro will contain all the attributes with the value determined with positive results </param>
/// <param name="negatives">nro will contain all the attributes with the value determined with negative</param>
private void getValuesToAttribute0(DataTable samples, TreeAttribute attribute, string value, out int positives, out int negatives)
{
positives = 0;
negatives = 0;
foreach (DataRow aRow in samples.Rows)
{
///To do: Figure out if this is correct - it looks bad
if (((string)aRow[attribute.AttributeName] == value))
{
if (aRow[mTargetAttribute].ToString().Trim().ToUpper() == "TRUE")
{
positives++;
}
else
{
negatives++;
}
}
}
}
/// <summary>
/// Sets up a decision tree based on samples submitted
/// </summary>
/// <param name="samples">Table with samples that will be provided for mounting the tree </param>
/// <param name="targetAttribute"> Name column of the table that otherwise has the value true or false to
/// Validate or not a sample</param>
/// <returns>The root of the decision tree mounted </returns></returns?>
private TreeNode internalMountTree(DataTable samples, string targetAttribute, TreeAttributeCollection attributes)
{
if (allSamplesAreUniform(samples, targetAttribute) == true)
{
return(new TreeNode(new OutcomeTreeAttribute(getUniformRefValue(samples, targetAttribute))));
}
//if (allSamplesArePositive(samples, targetAttribute) == true)
// return new TreeNode(new OutcomeTreeAttribute(true));
//if (allSamplesAreNegative(samples, targetAttribute) == true)
// return new TreeNode(new OutcomeTreeAttribute(false));
if (attributes.Count == 0)
{
return(new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(samples, targetAttribute))));
}
mTotal = samples.Rows.Count;
mTargetAttribute = targetAttribute;
mTotalPositives = countTotalPositives(samples);
mEntropySet = getCalculatedEntropy(mTotalPositives, mTotal - mTotalPositives);
TreeAttribute bestAttribute = getBestAttribute(samples, attributes);
TreeNode root = new TreeNode(bestAttribute);
if (bestAttribute == null)
{
return(root);
}
PossibleValueCollection bestAttrValues = bestAttribute.PossibleValues;
bool continousSet = isContinousSet(bestAttrValues);
//DataTable aSample = samples.Clone();
if (continousSet)
{
string value = bestSplitValue(samples, bestAttribute);
{
DataTable aSample = samples.Clone();
//First Below then Above
DataRow[] rows;
string cond = bestAttribute.AttributeName + " <= " + "" + value + "";
rows = samples.Select(cond);
aSample.Rows.Clear();
foreach (DataRow row in rows)
{
aSample.Rows.Add(row.ItemArray);
Console.WriteLine(" SPLIT {0} ROW:", cond);
foreach (DataColumn myCol in samples.Columns)
{
Console.WriteLine(" " + row[myCol]);
}
}
// Create a new attribute list unless the attribute which is the current best attribute
TreeAttributeCollection aAttributes = new TreeAttributeCollection();
//ArrayList aAttributes = new ArrayList(attributes.Count - 1);
for (int i = 0; i < attributes.Count; i++)
{
if (attributes[i].AttributeName != bestAttribute.AttributeName)
{
aAttributes.Add(attributes[i]);
}
}
//Recycle the best continous attribute if there are others
if (aAttributes.Count > 0)
{
aAttributes.Add(bestAttribute);
}
// Create a new attribute list unless the attribute which is the current best attribute
if (rows.Length == 0)
{
//return new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(aSample, targetAttribute)));
return(new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(samples, targetAttribute))));
}
else
{
DecisionTree dc3 = new DecisionTree();
TreeNode ChildNode = dc3.mountTree(aSample, targetAttribute, aAttributes);
root.AddTreeNode(ChildNode, value, "leq");
}
}
{
DataTable aSample = samples.Clone();
DataRow[] rows2;
string cond = bestAttribute.AttributeName + " > " + "" + value + "";
rows2 = samples.Select(cond);
aSample.Rows.Clear();
foreach (DataRow row in rows2)
{
aSample.Rows.Add(row.ItemArray);
Console.WriteLine(" SPLIT {0} ROW:", cond);
foreach (DataColumn myCol in samples.Columns)
{
Console.WriteLine(" " + row[myCol]);
}
}
// Create a new attribute list unless the attribute which is the current best attribute
TreeAttributeCollection aAttributes2 = new TreeAttributeCollection();
//ArrayList aAttributes = new ArrayList(attributes.Count - 1);
for (int i = 0; i < attributes.Count; i++)
{
if (attributes[i].AttributeName != bestAttribute.AttributeName)
{
aAttributes2.Add(attributes[i]);
}
}
//Recycle the best continous attribute if there are others
if (aAttributes2.Count > 0)
{
aAttributes2.Add(bestAttribute);
}
// Create a new attribute list unless the attribute which is the current best attribute
if (rows2.Length == 0)
{
//return new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(aSample, targetAttribute)));
return(new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(samples, targetAttribute))));
}
else
{
DecisionTree dc3 = new DecisionTree();
TreeNode ChildNode = dc3.mountTree(aSample, targetAttribute, aAttributes2);
root.AddTreeNode(ChildNode, value, "gt");
}
}
}
else
{
DataTable aSample = samples.Clone();
foreach (string value in bestAttribute.PossibleValues)
{
// Select all elements with the value of this attribute
aSample.Rows.Clear();
DataRow[] rows;
rows = samples.Select(bestAttribute.AttributeName + " = " + "'" + value + "'");
foreach (DataRow row in rows)
{
aSample.Rows.Add(row.ItemArray);
}
// Select all elements with the value of this attribute
// Create a new attribute list unless the attribute which is the current best attribute
TreeAttributeCollection aAttributes = new TreeAttributeCollection();
//ArrayList aAttributes = new ArrayList(attributes.Count - 1);
for (int i = 0; i < attributes.Count; i++)
{
if (attributes[i].AttributeName != bestAttribute.AttributeName)
{
aAttributes.Add(attributes[i]);
}
}
// Create a new attribute list unless the attribute which is the current best attribute
if (aSample.Rows.Count == 0)
{
//return new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(aSample, targetAttribute)));
return(new TreeNode(new OutcomeTreeAttribute(getMostCommonValue(samples, targetAttribute))));
}
else
{
DecisionTree dc3 = new DecisionTree();
TreeNode ChildNode = dc3.mountTree(aSample, targetAttribute, aAttributes);
root.AddTreeNode(ChildNode, value, "eq");
}
}
}
return(root);
}