private void btNext_Click(object sender, EventArgs e)
{
if (rbGini.Checked == true) {
Def.Tree.Algorithm = Tree.AlgorithmEnum.Gini;
Def.FrmMain.Text = Def.APPLICATION_NAME + " using Gini index " + "on the " + Def.DbTableInUse + " data set";
} else
if (rbEntropy.Checked == true) {
Def.Tree.Algorithm = Tree.AlgorithmEnum.Entropy;
Def.FrmMain.Text = Def.APPLICATION_NAME + " using Information Entropy" + " on the " + Def.DbTableInUse + " data set"; ;
}
else
if (rbMaxDif.Checked == true) {
Def.Tree.Algorithm = Tree.AlgorithmEnum.MaxDif;
Def.FrmMain.Text = Def.APPLICATION_NAME + " using MaxDif" + " on the " + Def.DbTableInUse + " data set"; ;
}//else
//if (rbHybrid.Checked == true) {
// Def.Tree.Algorithm = Tree.AlgorithmEnum.Hybrid;
// Def.FrmMain.Text = Def.APPLICATION_NAME + " using mixed algorithms" + " on the " + Def.DbTableInUse + " data set"; ;
//}
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Categorical) {
NodeTargetCategorical root = new NodeTargetCategorical();
Def.Tree.Root = root;
} else {
NodeTargetContinuous root = new NodeTargetContinuous();
Def.Tree.Root = root;
}
Def.Tree.GrowthState = Tree.GrowthStateEnum.Root;
Def.TreeCanBeDisplayed = true;
Def.FrmMain.TreeBuild();
Def.ToolBar.Items["btNew"].Enabled = true;
Close();
}
private void btNext_Click(object sender, EventArgs e)
{
if (rbNetReductionInVariance.Checked == true) {
Def.Tree.Algorithm = Tree.AlgorithmEnum.NetRiV;
Def.FrmMain.Text = Def.APPLICATION_NAME + " using Net Reduction in Variance" + " on the " + Def.DbTableInUse + " data set"; ;
} else
if (rbGrossReductionInVariance.Checked == true) {
Def.Tree.Algorithm = Tree.AlgorithmEnum.GrossRiV;
Def.FrmMain.Text = Def.APPLICATION_NAME + " using Gross Reduction in Variance" + " on the " + Def.DbTableInUse + " data set"; ;
}
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Categorical) {
NodeTargetCategorical root = new NodeTargetCategorical();
Def.Tree.Root = root;
} else {
NodeTargetContinuous root = new NodeTargetContinuous();
Def.Tree.Root = root;
}
Def.Tree.GrowthState = Tree.GrowthStateEnum.Root;
Def.TreeCanBeDisplayed = true;
Def.FrmMain.TreeBuild();
Def.ToolBar.Items["btNew"].Enabled = true;
Close();
}
public void ManuallySplit(Node node)
{
int nextNodeId=-1;
string sql="";
NodeTargetCategorical lcat, rcat;
NodeTargetContinuous lcon, rcon;
if (node.SplitVariable.Variable.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Continuous) {
nextNodeId = Node.CountId+1;
Def.Db.TableDropIfExists(Def.DbTrTb + nextNodeId);
Def.Db.ReferenceTableIndexDropIfExists(nextNodeId);
sql =
@"CREATE TABLE " +
Def.DbTrTb + nextNodeId +
" AS " +
"SELECT " + Def.DbTrTb + node.Id + "." + Def.DbTableIdName + " FROM " +
Def.DbTrTb + node.Id + ", " + Def.DbBsTb +
" WHERE " +
Def.DbTrTb + node.Id + "." + Def.DbTableIdName + "=" + Def.DbBsTb + "." + Def.DbTableIdName +
" AND " + node.SplitVariable.Variable.Name + " <= " + node.SplitValue;
Def.Db.ExecuteNonQuery(sql);
Def.Db.ReferenceTableIndexCreate(nextNodeId);
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Continuous) {
lcon = new NodeTargetContinuous(node);
lcon.LabelTopText = " <= " + Math.Round(node.SplitValue, 2);
} else {
lcat = new NodeTargetCategorical(node);
lcat.LabelTopText = " <= " + Math.Round(node.SplitValue, 2);
}
//Right
nextNodeId = Node.CountId + 1;
Def.Db.TableDropIfExists(Def.DbTrTb + nextNodeId);
Def.Db.ReferenceTableIndexDropIfExists(nextNodeId);
sql =
@"CREATE TABLE " +
Def.DbTrTb + nextNodeId +
" AS " +
"SELECT " + Def.DbTrTb + node.Id + "." + Def.DbTableIdName + " FROM " +
Def.DbTrTb + node.Id + ", " + Def.DbBsTb +
" WHERE " +
Def.DbTrTb + node.Id + "." + Def.DbTableIdName + "=" + Def.DbBsTb + "." + Def.DbTableIdName +
" AND " + node.SplitVariable.Variable.Name + " > " + node.SplitValue;
Def.Db.ExecuteNonQuery(sql);
Def.Db.ReferenceTableIndexCreate(nextNodeId);
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Continuous) {
rcon = new NodeTargetContinuous(node);
rcon.LabelTopText = " > " + Math.Round(node.SplitValue, 2);
} else {
rcat = new NodeTargetCategorical(node);
rcat.LabelTopText = " > " + Math.Round(node.SplitValue, 2);
}
} else //SchemaVariable.VariableTypeEnum.Continuous
if (node.SplitVariable.Variable.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Categorical) {
string vals = "";
nextNodeId = Node.CountId + 1;
//LEFT NODE
List<string> caseLst = node.SplitVariable.ChildrenGroups.ValueGroupLst[0];
for (int i = 0; i < caseLst.Count; ++i) {
vals += node.SplitVariable.Variable.Name + "='" + caseLst[i] + "' ";
if (i < (caseLst.Count - 1)) {
vals += " or ";
} else
vals += ")";
}
Def.Db.TableDropIfExists(Def.DbTrTb + nextNodeId);
Def.Db.ReferenceTableIndexDropIfExists(nextNodeId);
sql =
@"CREATE TABLE " +
Def.DbTrTb + nextNodeId +
" AS " +
"SELECT " + Def.DbTrTb + node.Id + "." + Def.DbTableIdName + " FROM " +
Def.DbTrTb + node.Id + ", " + Def.DbBsTb +
" WHERE (" +
Def.DbTrTb + node.Id + "." + Def.DbTableIdName + "=" + Def.DbBsTb + "." + Def.DbTableIdName +
") AND (" + vals;
Def.Db.ExecuteNonQuery(sql);
Def.Db.ReferenceTableIndexCreate(nextNodeId);
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Continuous) {
lcon = new NodeTargetContinuous(node);
} else {
lcat = new NodeTargetCategorical(node);
}
//RIGHT NODE
vals = "";
nextNodeId = Node.CountId + 1;
caseLst = node.SplitVariable.ChildrenGroups.ValueGroupLst[1];
for (int i = 0; i < caseLst.Count; ++i) {
vals += node.SplitVariable.Variable.Name + "='" + caseLst[i] + "' ";
if (i < (caseLst.Count - 1)) {
vals += " or ";
} else
vals += ")";
}
Def.Db.TableDropIfExists(Def.DbTrTb + nextNodeId);
Def.Db.ReferenceTableIndexDropIfExists(nextNodeId);
sql =
@"CREATE TABLE " +
Def.DbTrTb + nextNodeId +
" AS " +
"SELECT " + Def.DbTrTb + node.Id + "." + Def.DbTableIdName + " FROM " +
Def.DbTrTb + node.Id + ", " + Def.DbBsTb +
" WHERE (" +
Def.DbTrTb + node.Id + "." + Def.DbTableIdName + "=" + Def.DbBsTb + "." + Def.DbTableIdName +
") AND (" + vals;
Def.Db.ExecuteNonQuery(sql);
Def.Db.ReferenceTableIndexCreate(nextNodeId);
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Continuous) {
rcon = new NodeTargetContinuous(node);
} else {
rcat = new NodeTargetCategorical(node);
}
}
afterSplit(node);
}
public bool AutoSplitMv(Node node)
{
int nextNodeId = -1;
string sql = "";
double varSum=0;
NodeTargetCategorical lcat, rcat;
NodeTargetContinuous lcon, rcon;
//if (node.SplitVariable.SplitStatus != Predictor.SplitStatusEnum.CanBeUsed)
// return false;
if (node.Level >= Def.TreeLevelsMax)
return false;
nextNodeId = Node.CountId + 1;
Def.Db.TableDropIfExists(Def.DbTrTb + nextNodeId);
Def.Db.ReferenceTableIndexDropIfExists(nextNodeId);
//Left node
sql =
@"CREATE TABLE " +
Def.DbTrTb + nextNodeId + "(" + Def.DbTableIdName + " integer NOT NULL)";
Def.Db.ExecuteNonQuery(sql);
//Right node
nextNodeId = nextNodeId + 1;
Def.Db.TableDropIfExists(Def.DbTrTb + nextNodeId);
Def.Db.ReferenceTableIndexDropIfExists(nextNodeId);
sql =
@"CREATE TABLE " +
Def.DbTrTb + nextNodeId + "(" + Def.DbTableIdName + " integer NOT NULL)";
Def.Db.ExecuteNonQuery(sql);
List<string> insert = new List<string>();
for (int y = 0; y < node.Table.RowCount; ++y) {
varSum = 0;
for (int i = 0; i < node.PredMvLst.Count; ++i) {
varSum += node.PredMvLst[i].Coef * node.PredMvLst[i].X(y);
}
if ((varSum + node.C) <= 0) {
insert.Add(@"insert into " + Def.DbTrTb + (nextNodeId - 1) + " values (" + node.MvTb.Data.ID[y] + ")");
} else {
insert.Add(@"insert into " + Def.DbTrTb + (nextNodeId) + " values (" + node.MvTb.Data.ID[y] + ")");
}
}
Def.Db.NonQueryTransaction(insert);
Def.Db.ReferenceTableIndexCreate(nextNodeId-1);
Def.Db.ReferenceTableIndexCreate(nextNodeId);
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Continuous) {
rcon = new NodeTargetContinuous(node);
lcon = new NodeTargetContinuous(node);
// node equation goes here
} else {
rcat = new NodeTargetCategorical(node);
lcat = new NodeTargetCategorical(node);
// node equation goes here
}
afterSplit(node);
return true;
}
public NodeTargetContinuousUI(NodeTargetContinuous node)
{
int x, y;
x = y = 0;
Def.PbBase.Controls.Add(this);
this.components = new System.ComponentModel.Container();
this.Selector = new System.Windows.Forms.ContextMenuStrip(this.components);
this.Selector.GripMargin = new System.Windows.Forms.Padding(2);
this.Selector.SuspendLayout();
this.SuspendLayout();
// this.AutoSizeColumnsMode = System.Windows.Forms.DataGridViewAutoSizeColumnsMode.AllCells;
// this.AutoSizeRowsMode = System.Windows.Forms.DataGridViewAutoSizeRowsMode.AllCells;
this.Node = node;
LabelTop = new Label();
LabelBottom = new Label();
LabelTop.SuspendLayout();
LabelBottom.SuspendLayout();
Def.PbBase.Controls.Add(this.LabelTop);
Def.PbBase.Controls.Add(this.LabelBottom);
LabelTop.AutoSize = true;
LabelBottom.AutoSize = true;
this.viewdataToolStripMenuItem = new System.Windows.Forms.ToolStripMenuItem();
this.manuallysplitToolStripMenuItem = new System.Windows.Forms.ToolStripMenuItem();
this.autosplitToolStripMenuItem = new System.Windows.Forms.ToolStripMenuItem();
this.fullautogrowToolStripMenuItem = new System.Windows.Forms.ToolStripMenuItem();
this.removeSubNodesToolStripMenuItem = new System.Windows.Forms.ToolStripMenuItem();
this.MouseHover += new System.EventHandler(grid_MouseHover);
this.MouseLeave += new System.EventHandler(grid_MouseLeave);
this.ContextMenuStrip = this.Selector;
//
// Selector
//
this.Selector.AllowDrop = true;
this.Selector.Items.AddRange(new System.Windows.Forms.ToolStripItem[] {
this.viewdataToolStripMenuItem,
this.manuallysplitToolStripMenuItem,
this.autosplitToolStripMenuItem,
this.fullautogrowToolStripMenuItem,
this.removeSubNodesToolStripMenuItem});
this.Selector.Location = new System.Drawing.Point(23, 61);
this.Selector.Name = "Menu";
this.Selector.Size = new System.Drawing.Size(133, 148);
this.Selector.Visible = true;
//
// viewdataToolStripMenuItem
//
this.viewdataToolStripMenuItem.Name = "viewdataToolStripMenuItem";
this.viewdataToolStripMenuItem.Text = "View data";
this.viewdataToolStripMenuItem.Click += new System.EventHandler(this.viewdataToolStripMenuItem_Click);
//
// manuallysplitToolStripMenuItem
//
this.manuallysplitToolStripMenuItem.Name = "manuallysplitToolStripMenuItem";
this.manuallysplitToolStripMenuItem.Text = "Manually split";
this.manuallysplitToolStripMenuItem.Click += new System.EventHandler(this.manuallysplitToolStripMenuItem_Click);
//
// autosplitToolStripMenuItem
//
this.autosplitToolStripMenuItem.Name = "autosplitToolStripMenuItem";
this.autosplitToolStripMenuItem.Text = "Auto split";
this.autosplitToolStripMenuItem.Click += new System.EventHandler(this.autosplitToolStripMenuItem_Click);
//
// fullautogrowToolStripMenuItem
//
this.fullautogrowToolStripMenuItem.Name = "fullautogrowToolStripMenuItem";
this.fullautogrowToolStripMenuItem.Text = "Full auto grow";
this.fullautogrowToolStripMenuItem.Click += new System.EventHandler(this.fullautogrowToolStripMenuItem_Click);
//
// removeSubNodesToolStripMenuItem
//
this.removeSubNodesToolStripMenuItem.Name = "removeSubNodesToolStripMenuItem";
this.removeSubNodesToolStripMenuItem.Text = "Remove subnodes";
this.removeSubNodesToolStripMenuItem.Click += new System.EventHandler(this.removeSubNodesToolStripMenuItem_Click);
DataGridViewCellStyle Style1 = new DataGridViewCellStyle();
DataGridViewCellStyle Style2 = new DataGridViewCellStyle();
C1 = new DataGridViewTextBoxColumn();
C2 = new DataGridViewTextBoxColumn();
AllowUserToAddRows = false;
AllowUserToDeleteRows = false;
AllowUserToOrderColumns = false;
AllowUserToResizeColumns = false;
AllowUserToResizeRows = false;
Style1.Alignment = DataGridViewContentAlignment.TopLeft;
Style1.WrapMode = DataGridViewTriState.False;
AlternatingRowsDefaultCellStyle = Style1;
CellBorderStyle = DataGridViewCellBorderStyle.SingleVertical;
ClipboardCopyMode = DataGridViewClipboardCopyMode.Disable;
Style2.Alignment = DataGridViewContentAlignment.TopLeft;
Style2.BackColor = System.Drawing.SystemColors.Control;
Style2.Font = new System.Drawing.Font("Microsoft Sans Serif", 8.25F, System.Drawing.FontStyle.Regular, System.Drawing.GraphicsUnit.Point, ((byte)(0)));
Style2.ForeColor = System.Drawing.SystemColors.WindowText;
Style2.SelectionBackColor = System.Drawing.SystemColors.Highlight;
Style2.SelectionForeColor = System.Drawing.SystemColors.HighlightText;
Style2.WrapMode = DataGridViewTriState.True;
ColumnHeadersDefaultCellStyle = Style2;
ColumnHeadersVisible = false;
Columns.Add(this.C1);
Columns.Add(this.C2);
Cursor = Cursors.Hand;
EditMode = DataGridViewEditMode.EditProgrammatically;
Margin = new Padding(0);
MultiSelect = false;
Name = "nodeContinuous";
ReadOnly = true;
RowHeadersVisible = false;
RowTemplate.Height = 18;
ScrollBars = ScrollBars.None;
// SelectionMode = DataGridViewSelectionMode.ColumnHeaderSelect;
this.C1.DefaultCellStyle = Style1;
this.C1.Frozen = true;
this.C1.HeaderText = "Column1";
this.C1.Name = "c1";
this.C1.ReadOnly = true;
this.C1.Width = 50;
this.C2.DefaultCellStyle = Style2;
this.C2.Frozen = true;
this.C2.HeaderText = "Column2";
this.C2.Name = "c2";
this.C2.ReadOnly = true;
this.C2.Width = 60;
Rows.Add(6);
Rows[y].Cells[x].Value = "Node " + Node.Id; ++y;
Rows[y].Cells[x].Value = "Mean"; ++y;
Rows[y].Cells[x].Value = "Std Dev"; ++y;
Rows[y].Cells[x].Value = "Cf Var"; ++y;
Rows[y].Cells[x].Value = "n"; ++y;
Rows[y].Cells[x].Value = "%"; ++y;
++x; y = 1;
double mean = Node.Mean;
double stdDev = Node.StdDev;
Rows[y].Cells[x].Value = Math.Round(mean, 2); ++y;
Rows[y].Cells[x].Value = Math.Round(stdDev, 2); ++y;
Rows[y].Cells[x].Value = Math.Round(stdDev / mean, 2); ++y;
Rows[y].Cells[x].Value = Node.Table.RowCount; ++y;
Rows[y].Cells[x].Value = Math.Round((double)Node.Table.RowCount / (double)Def.TrainingSetRowCount * (double)100, 2); ++y;
++x; y = 1;
// AutoSizeColumn(DataGridViewAutoSizeColumnCriteria.HeaderAndRows, 0);
// AutoSizeColumn(DataGridViewAutoSizeColumnCriteria.HeaderAndRows, 1);
LabelTop.Text = "";
LabelBottom.Text = "";
Height = (RowTemplate.Height * 6) + 4;
Width = 110;
removeSubNodesToolStripMenuItem.Enabled = false;
this.Selector.ResumeLayout(false);
this.Selector.Hide();
this.ResumeLayout(false);
}
private void treeInitialise()
{
if (Def.Schema.Target.VariableTypeUserSet == SchemaVariable.VariableTypeEnum.Categorical) {
NodeTargetCategorical root = new NodeTargetCategorical();
Def.Tree.Root = root;
} else {
NodeTargetContinuous root = new NodeTargetContinuous();
Def.Tree.Root = root;
}
Def.Tree.GrowthState = Tree.GrowthStateEnum.Root;
Def.TreeCanBeDisplayed = false;
//Def.FrmMain.TreeBuild();
}
//============================================================================================================
//============================================================================================================
//Sets the best p.SplitValue p.Gain
//public static double MinInfoCatHeuristicSLOW(NodeTargetContinuous n, Predictor p) {
// int i, bestPartitionSplitPoint = 0;
// // int dfd; //delete
// double minVar = 0, var, lVar, rVar, instanceCount = n.Table.RowCount;
// lVar = rVar = minVar = var = 0;
// List<NNT> nntLst;
// List<string> left = new List<string>();
// List<string> right = new List<string>();
// List<string> leftBest = new List<string>();
// List<string> rightBest = new List<string>();
// //Tries each partition:
// //AvsLst[i].N0 = y
// //AvsLst[i].N1 = Value of the dependent varible
// //AvsLst[i].N2 = Frequency of y
// //AvsLst[i].N3 = Total of distinct registries until that row
// string sql =
// @"SELECT count(*), " +
// "0, " +
// Def.DbBsTb + "." + p.Variable.Name + " " +
// "FROM "
// + Def.DbBsTb + "," + Def.DbTrTb + n.Id + " " +
// "WHERE "
// + Def.DbBsTb + "." + Def.DbTableIdName + " = " +
// Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " " +
// "GROUP BY " +
// Def.DbBsTb + "." + p.Variable.Name + " " +
// "ORDER BY avg("
// + Def.DbBsTb + "." + Def.Schema.Target.Name + ")";
// nntLst = Def.Db.GetNNTLst(sql);
// //N1 is the number of registries until a given row
// left.Add(nntLst[0].T);
// leftBest.Add(nntLst[0].T);
// if (nntLst.Count > 0) {
// nntLst[0].N1 = nntLst[0].N0;
// for (i = 1; i < nntLst.Count; ++i) {
// nntLst[i].N1 = nntLst[i - 1].N1 + nntLst[i].N0;
// right.Add(nntLst[i].T);
// rightBest.Add(nntLst[i].T);
// }
// }
// n.DescendentImpPreCalculated = new List<double>(2);
// n.DescendentImpPreCalculated.Add(0);
// n.DescendentImpPreCalculated.Add(0);
// p.SplitStatus = Predictor.SplitStatusEnum.CanBeUsed;
// if (nntLst.Count == 0) {
// p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
// p.Gain = 0;
// return 0;
// }
// for (i = 0; i < nntLst.Count - 1; ++i) {
// //if(i==12 && p.Variable.Name=="apache_2")
// // dfd=8;
// lVar = Fcn.InfoTimesCount(left, n, p);
// rVar = Fcn.InfoTimesCount(right, n, p);
// if (i == 0) {
// minVar = (lVar + rVar) / instanceCount;
// n.DescendentImpPreCalculated[0] = lVar;
// n.DescendentImpPreCalculated[1] = rVar;
// } else {
// var = (lVar + rVar) / instanceCount;
// if (var < minVar) {
// minVar = var;
// bestPartitionSplitPoint = i;
// n.DescendentImpPreCalculated[0] = lVar;
// n.DescendentImpPreCalculated[1] = rVar;
// leftBest.Clear();
// rightBest.Clear();
// foreach (string s in left)
// leftBest.Add(s);
// foreach (string s in right)
// rightBest.Add(s);
// }
// }
// left.Add(right[0]);
// right.RemoveAt(0);
// }
// p.ChildrenGroups.ValueLst[0].Clear();
// p.ChildrenGroups.ValueLst[1].Clear();
// //Add left node values
// for (i = 0; i <= bestPartitionSplitPoint; ++i)
// p.ChildrenGroups.ValueLst[0].Add(p.CaseSd.IndexOfKey(nntLst[i].T));
// //Add right node values
// for (i = bestPartitionSplitPoint + 1; i < nntLst.Count; ++i)
// p.ChildrenGroups.ValueLst[1].Add(p.CaseSd.IndexOfKey(nntLst[i].T));
// p.Gain = (n.Imp - minVar) * 100 / n.Imp;
// return p.Gain;
//}
////============================================================================================================
////============================================================================================================
//Sets the best p.SplitValue p.Gain
public static double MinInfoCont(NodeTargetContinuous n, Predictor p)
{
int i, leftRowCount = 0, rightRowCount = 0;
// int dfd; //delete
double minVar, var, lVar, rVar, instanceCount = n.Table.RowCount;
lVar = rVar = minVar = var = double.NaN;
List<N4> AvsLst;
List<int> thresholdIndexLst;
//Tries each partition:
//AvsLst[i].N0 = y
//AvsLst[i].N1 = Value of the dependent varible
//AvsLst[i].N2 = Frequency of y
//AvsLst[i].N3 = Total of distinct registries until that row
string sql =
@"SELECT ALL "
+ Def.DbBsTb + "." + Def.Schema.Target.Name + ", " +
Def.DbBsTb + "." + p.Variable.Name + ", " +
" count(" + Def.DbBsTb + "." + p.Variable.Name + "), 0 " +
"FROM "
+ Def.DbBsTb + "," + Def.DbTrTb + n.Id + " " +
"WHERE "
+ Def.DbBsTb + "." + Def.DbTableIdName + " = " +
Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " " +
" AND " + Def.DbBsTb + "." + p.Variable.Name + " IS NOT NULL " +
"GROUP BY " +
Def.DbBsTb + "." + Def.Schema.Target.Name + ", " +
Def.DbBsTb + "." + p.Variable.Name + " " +
"ORDER BY " +
Def.DbBsTb + "." + p.Variable.Name;
AvsLst = Def.Db.GetN4Lst(sql);
//N3 is the number of registries until a given row
if (AvsLst.Count > 0) {
AvsLst[0].N3 = AvsLst[0].N2;
for (i = 1; i < AvsLst.Count; ++i)
AvsLst[i].N3 = AvsLst[i - 1].N3 + AvsLst[i].N2;
}
thresholdIndexLst = Fcn.SetPossibleThresholdIndexLst(AvsLst);
if (thresholdIndexLst.Count == 0) {
p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
p.Gain = 0;
return 0;
}
n.DescendentImpPreCalculated = new List<double>(2);
n.DescendentImpPreCalculated.Add(0);
n.DescendentImpPreCalculated.Add(0);
p.SplitStatus = Predictor.SplitStatusEnum.CanBeUsed;
if (AvsLst.Count == 0) {
p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
p.Gain = 0;
return 0;
}
for (i = 0; i < thresholdIndexLst.Count; ++i) {
lVar = Info(0, thresholdIndexLst[i], AvsLst, out leftRowCount);
rVar = Info(thresholdIndexLst[i] + 1, AvsLst.Count - 1, AvsLst, out rightRowCount);
if (Double.IsNaN(minVar) && leftRowCount >= Def.TreeMinNumberOfCasesPerNode && rightRowCount >= Def.TreeMinNumberOfCasesPerNode) {
var = (leftRowCount * lVar + rightRowCount * rVar) / (leftRowCount + rightRowCount);
if (Double.IsNaN(var) == false) {
minVar = var;
p.SplitValue = AvsLst[thresholdIndexLst[i]].N1;
n.DescendentImpPreCalculated[0] = lVar;
n.DescendentImpPreCalculated[1] = rVar;
}
} else {
var = (leftRowCount * lVar + rightRowCount * rVar) / (leftRowCount + rightRowCount);
if (var < minVar && !Double.IsNaN(var) && leftRowCount >= Def.TreeMinNumberOfCasesPerNode && rightRowCount >= Def.TreeMinNumberOfCasesPerNode) {
minVar = var;
p.SplitValue = AvsLst[thresholdIndexLst[i]].N1;
n.DescendentImpPreCalculated[0] = lVar;
n.DescendentImpPreCalculated[1] = rVar;
}
}
}
if (Double.IsNaN(minVar)) {
p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
p.Gain = 0;
return 0;
}
p.Gain = (n.Imp - minVar) * 100 / n.Imp;
p.Gain *= (double)(n.Table.RowCount - p.NullCount) / n.Table.RowCount;
return p.Gain;
}
//Sets the best p.SplitValue p.Gain
//public static double MinInfoContSQL(NodeTargetContinuous n, Predictor p) {
// int i;
// double l, r, minVar, var;
// l = r = minVar = var = 0;
// //Tries each partition:
// //p.Gain = (n.Imp - minVar) * 100 / n.Imp;
// List<double> vlLst;
// List<double> NLst;
// OdbcTransaction dbTrans = null;
// dbTrans = Def.Db.Con.BeginTransaction();
// string sql =
// @"SELECT DISTINCT " +
// Def.DbBsTb + "." + p.Variable.Name + " " +
// "FROM " +
// Def.DbBsTb + "," + Def.DbTrTb + n.Id + " " +
// "WHERE " +
// Def.DbBsTb + "." + Def.DbTableIdName + " = " +
// Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " " +
// "ORDER BY " +
// Def.DbBsTb + "." + p.Variable.Name;
// vlLst = Def.Db.GetNumberLst(sql, dbTrans);
// p.SplitStatus = Predictor.SplitStatusEnum.CanBeUsed;
// if (vlLst.Count == 0) {
// p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
// p.Gain = 0;
// return 0;
// }
// for (i = 0; i < vlLst.Count; ++i) {
// sql =
// @"SELECT
// COALESCE(variance(" + n.Tree.Schema.Target.Name + "), 0), count(*) " +
// "FROM "
// + Def.DbBsTb + " , " + Def.DbTrTb + n.Id + " " +
// "WHERE "
// + Def.DbBsTb + "." + Def.DbTableIdName + "=" +
// Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " and " +
// Def.DbBsTb + "." + p.Variable.Name + "<=" + vlLst[i];
// NLst = Def.Db.GetNumberRowLst(sql, dbTrans);
// l = NLst[0] * NLst[1] / n.Table.RowCount;
// sql =
// @"SELECT
// COALESCE(variance(" + n.Tree.Schema.Target.Name + "), 0), count(*) " +
// "FROM "
// + Def.DbBsTb + " , " + Def.DbTrTb + n.Id + " " +
// "WHERE "
// + Def.DbBsTb + "." + Def.DbTableIdName + "=" +
// Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " and " +
// Def.DbBsTb + "." + p.Variable.Name + ">" + vlLst[i];
// NLst = Def.Db.GetNumberRowLst(sql, dbTrans);
// r = NLst[0] * NLst[1] / n.Table.RowCount;
// var = (l + r);
// if (i == 0) {
// minVar = var;
// p.SplitValue = vlLst[i];
// } else {
// if (var < minVar) {
// minVar = var;
// p.SplitValue = vlLst[i];
// }
// }
// }
// dbTrans.Commit();
// p.Gain = (n.Imp - minVar) * 100 / n.Imp; ;
// return p.Gain;
//}
//============================================================================================================
//============================================================================================================
//Sets the best p.SplitValue p.Gain
//Used to be: public static double MinInfoCatHeuristic(NodeTargetContinuous n, Predictor p) {
public static double MinInfoCat(NodeTargetContinuous n, Predictor p)
{
int i, bestPartitionSplitPoint = 0;
double minVar, lVar, rVar, instanceCount = n.Table.RowCount, partitionInfo;
lVar = rVar = minVar = double.NaN;
List<string> valLst;
List<NNT> DepIndepLst;
List<string> left = new List<string>();
string sqlAverage =
@"SELECT " +
Def.DbBsTb + "." + p.Variable.Name + " " +
"FROM "
+ Def.DbBsTb + "," + Def.DbTrTb + n.Id + " " +
"WHERE "
+ Def.DbBsTb + "." + Def.DbTableIdName + " = " +
Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " " +
"AND " + Def.DbBsTb + "." + p.Variable.Name + " IS NOT NULL " +
"GROUP BY " +
Def.DbBsTb + "." + p.Variable.Name + " " +
"ORDER BY avg("
+ Def.DbBsTb + "." + Def.Schema.Target.Name + ")";
valLst = Def.Db.GetTextLst(sqlAverage);
if (valLst.Count == 0) {
p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
p.Gain = 0;
return 0;
}
string sqlDepVar =
@"SELECT " +
Def.DbBsTb + "." + Def.Schema.Target.Name + ", " +
"count(*), " +
Def.DbBsTb + "." + p.Variable.Name + " " +
"FROM "
+ Def.DbBsTb + "," + Def.DbTrTb + n.Id + " " +
"WHERE "
+ Def.DbBsTb + "." + Def.DbTableIdName + " = " +
Def.DbTrTb + n.Id + "." + Def.DbTableIdName + " " +
"AND " + Def.DbBsTb + "." + p.Variable.Name + " IS NOT NULL " +
"GROUP BY " +
Def.DbBsTb + "." + Def.Schema.Target.Name + ", " +
Def.DbBsTb + "." + p.Variable.Name;
DepIndepLst = Def.Db.GetNNTLst(sqlDepVar);
p.SplitStatus = Predictor.SplitStatusEnum.CanBeUsed;
left.Add(valLst[0]);
for (i = 0; i < valLst.Count - 1; ++i) {
if (Double.IsNaN(minVar)) {
partitionInfo = PartitionInfo(left, DepIndepLst, n, p);
if (Double.IsNaN(partitionInfo) == false) {
minVar = partitionInfo;
bestPartitionSplitPoint = i;
}
} else {
partitionInfo = PartitionInfo(left, DepIndepLst, n, p);
if (Double.IsNaN(partitionInfo) == false && partitionInfo < minVar) {
minVar = partitionInfo;
bestPartitionSplitPoint = i;
}
}
left.Add(valLst[i + 1]);
}
if(Double.IsNaN(minVar)){
p.SplitStatus = Predictor.SplitStatusEnum.NotEnoughCases;
p.Gain = 0;
return 0;
}
p.ChildrenGroups.ValueGroupLst[0].Clear();
p.ChildrenGroups.ValueGroupLst[1].Clear();
//Add left node values
// int IndexOfKeyValLsti; // Just to check negative indexes
for (i = 0; i <= bestPartitionSplitPoint; ++i) {
// IndexOfKeyValLsti = p.ValueSd.IndexOfKey(valLst[i]);
// if (IndexOfKeyValLsti < 0)
// MessageBox.Show("Negative index in MinInfoCatHeuristic", "Error");
p.ChildrenGroups.ValueGroupLst[0].Add(valLst[i]);
}
//Add right node values
for (i = bestPartitionSplitPoint + 1; i < valLst.Count; ++i) {
// IndexOfKeyValLsti = p.ValueSd.IndexOfKey(valLst[i]);
// if (IndexOfKeyValLsti < 0)
// MessageBox.Show("Negative index in MinInfoCatHeuristic", "Error");
p.ChildrenGroups.ValueGroupLst[1].Add(valLst[i]);
}
p.Gain = (n.Imp - minVar) * 100 / n.Imp;
p.Gain *= (double)(n.Table.RowCount - p.NullCount) / n.Table.RowCount;
return p.Gain;
}
