/// <summary>
/// Calculates a sorted list of information gains by splitting
/// on each axes and value and measuring the resulting disorder
/// </summary>
/// <param name="disorderType">The disorder calculation algorithm to use</param>
/// <returns></returns>
public IDictionary <int, Number> GetInformationGain(DisorderType disorderType)
{
var grid = new Dictionary <int, Number>(_matrix.Rows);
var baseline = GetDisorder(disorderType, _matrix);
var axes = _matrix.Columns;
for (var i = 0; i < axes; i++)
{
var row = _matrix[i];
var values = row.Distinct();
var disorder = 0.0;
foreach (var value in values)
{
var split = Split(i, value);
var probability = split.Rows / (float)_matrix.Rows;
disorder += probability * GetDisorder(disorderType, split);
}
var gain = baseline - disorder;
grid.Add(i, gain);
}
var sorted = grid.OrderBy(kv => kv.Value).ToDictionary(x => x.Key, x => x.Value);;
return(sorted);
}
/// <summary>
/// Calculates a sorted list of information gains by splitting
/// on each axes and value and measuring the resulting disorder
/// </summary>
/// <param name="disorderType">The disorder calculation algorithm to use</param>
/// <returns></returns>
public IDictionary<int, Number> GetInformationGain(DisorderType disorderType)
{
var grid = new Dictionary<int, Number>(_matrix.Rows);
var baseline = GetDisorder(disorderType, _matrix);
var axes = _matrix.Columns;
for (var i = 0; i < axes; i++)
{
var row = _matrix[i];
var values = row.Distinct();
var disorder = 0.0;
foreach(var value in values)
{
var split = Split(i, value);
var probability = split.Rows / (float)_matrix.Rows;
disorder += probability * GetDisorder(disorderType, split);
}
var gain = baseline - disorder;
grid.Add(i, gain);
}
var sorted = grid.OrderBy(kv => kv.Value).ToDictionary(x => x.Key, x => x.Value); ;
return sorted;
}
private static ITreeNode CreateTree(Matrix dataSet, IList<string> labels, DisorderType disorderType)
{
var outcomes = dataSet.GetOutcomes();
if(outcomes.Distinct().Count().Equals(1))
{
return new DecisionTreeLeafNode(outcomes[0]);
}
var outcome = dataSet.GetMajorityOutcome();
if(dataSet.Columns == 1)
{
return new DecisionTreeLeafNode(outcome);
}
ITreeNode node = new DecisionTreeNode(dataSet, outcome);
var casted = (DecisionTreeNode) node;
var best = casted.GetInformationGain(disorderType).First();
var label = labels[best.Key];
casted.Label = label;
labels = labels.Where(l => !l.Equals(label)).ToList();
var subset = dataSet.Select(row => row[best.Key]).ToList();
var values = subset.Distinct().ToList();
foreach(var value in values)
{
var copy = new List<string>(labels);
var split = casted.Split(best.Key, value);
var child = CreateTree(split, copy, disorderType);
casted.AddChild(child);
}
ReverseNodeOrder(casted);
return node;
}
private static ITreeNode CreateTree(Matrix dataSet, IList <string> labels, DisorderType disorderType)
{
var outcomes = dataSet.GetOutcomes();
if (outcomes.Distinct().Count().Equals(1))
{
return(new DecisionTreeLeafNode(outcomes[0]));
}
var outcome = dataSet.GetMajorityOutcome();
if (dataSet.Columns == 1)
{
return(new DecisionTreeLeafNode(outcome));
}
ITreeNode node = new DecisionTreeNode(dataSet, outcome);
var casted = (DecisionTreeNode)node;
var best = casted.GetInformationGain(disorderType).First();
var label = labels[best.Key];
casted.Label = label;
labels = labels.Where(l => !l.Equals(label)).ToList();
var subset = dataSet.Select(row => row[best.Key]).ToList();
var values = subset.Distinct().ToList();
foreach (var value in values)
{
var copy = new List <string>(labels);
var split = casted.Split(best.Key, value);
var child = CreateTree(split, copy, disorderType);
casted.AddChild(child);
}
ReverseNodeOrder(casted);
return(node);
}
protected override void PrepareMenuCategories()
{
int categoryTypes = disordersSceneManager.DisordersCategories.DisordersCategoriesList.Count;
for (int i = 0; i < categoryTypes; i++)
{
GameObject category = (GameObject)Instantiate(categoryWholePrefab, transform.position, Quaternion.identity);
category.SetActive(true);
category.name = disordersSceneManager.DisordersCategories.DisordersCategoriesList [i].Type.ToString();
category.transform.SetParent(menuContent.transform, true);
category.transform.localScale = Vector3.one;
category.transform.Find("ToggleCategory").GetComponentInChildren <TextMeshProUGUI> ().text = disordersSceneManager.DisordersCategories.DisordersCategoriesList [i].Name;
categoryList.Add(category);
DisorderType statement = disordersSceneManager.DisordersCategories.DisordersCategoriesList [i].Type;
int elements = 0;
float height = 0;
float spacing = category.GetComponent <VerticalLayoutGroup> ().spacing;
switch (statement)
{
case DisorderType.REFRACTIVE_ERRORS:
elements = disordersSceneManager.DisordersElements.RefractiveErrorsList.Count;
height = elements * (toggleElementPrefab.GetComponent <RectTransform> ().sizeDelta.y + spacing) + category.GetComponent <RectTransform> ().sizeDelta.y + spacingCategory;
category.GetComponent <RectTransform> ().sizeDelta = new Vector2(category.GetComponent <RectTransform> ().sizeDelta.x, height);
break;
case DisorderType.DISEASES:
elements = disordersSceneManager.DisordersElements.DiseasesList.Count;
height = elements * (toggleElementPrefab.GetComponent <RectTransform> ().sizeDelta.y + spacing) + category.GetComponent <RectTransform> ().sizeDelta.y + spacingCategory;
category.GetComponent <RectTransform> ().sizeDelta = new Vector2(category.GetComponent <RectTransform> ().sizeDelta.x, height);
break;
default:
break;
}
}
}
/// <summary>
/// Calculates the disorder of given matrix of values
/// </summary>
/// <param name="disorderType">The disorder calculation algorithm to use</param>
/// <param name="matrix">The data set</param>
/// <returns></returns>
public static Number GetDisorder(DisorderType disorderType, Matrix matrix)
{
Number baseline;
switch (disorderType)
{
case DisorderType.ShannonEntropy:
baseline = Disorder.ShannonEntropy(matrix);
break;
case DisorderType.GiniImpurity:
baseline = Disorder.GiniImpurity(matrix);
break;
case DisorderType.ClassificationError:
baseline = Disorder.ClassificationError(matrix);
break;
default:
throw new ArgumentOutOfRangeException("disorderType");
}
return(baseline);
}
/// <summary>
/// Calculates the disorder of given matrix of values
/// </summary>
/// <param name="disorderType">The disorder calculation algorithm to use</param>
/// <param name="matrix">The data set</param>
/// <returns></returns>
public static Number GetDisorder(DisorderType disorderType, Matrix matrix)
{
Number baseline;
switch (disorderType)
{
case DisorderType.ShannonEntropy:
baseline = Disorder.ShannonEntropy(matrix);
break;
case DisorderType.GiniImpurity:
baseline = Disorder.GiniImpurity(matrix);
break;
case DisorderType.ClassificationError:
baseline = Disorder.ClassificationError(matrix);
break;
default:
throw new ArgumentOutOfRangeException("disorderType");
}
return baseline;
}
/// <summary>
/// Builds a tree using the Iterative Dichotimiser 3 (ID3) algorithm
/// <see href="http://en.wikipedia.org/wiki/ID3_algorithm" />
/// </summary>
/// <returns></returns>
public static ITreeNode ID3(Matrix dataSet, IList<string> labels, DisorderType disorderType)
{
return CreateTree(dataSet, labels, disorderType);
}
/// <summary>
/// Builds a tree using the Iterative Dichotimiser 3 (ID3) algorithm
/// <see href="http://en.wikipedia.org/wiki/ID3_algorithm" />
/// </summary>
/// <returns></returns>
public static ITreeNode ID3(Matrix dataSet, IList <string> labels, DisorderType disorderType)
{
return(CreateTree(dataSet, labels, disorderType));
}