ForestSize: Integer;
The ForestSize property determines the number of trees in random forest.
Available values within the [1; ∞) range.
The default value is 100.
Executing the example requires that the repository contains:
A table with the DM_TABLE identifier containing data for analysis
A regular report with the DM_REPORT_RES identifier, to which analysis results will be loaded.
Add links to the Metabase, Ms, Report, Stat, Tab, Ui system assemblies.
Sub UserRandFor;
Var
mb: IMetabase;
TableDS: IDmTableDataSource;
ReportDS: IDmReportDataSource;
Method: IDmMethod;
Report: IPrxReport;
Shs: IPrxSheets;
Sheet: ITabSheet;
RandFor: IDmRandomForest;
Binning: IDmField;
i: Integer;
Attrs: Array Of Integer;
Target: IUiCommandTarget;
Reports: IDmReports;
DmReport: IDmReport;
Begin
mb := MetabaseClass.Active;
// Create table data source
TableDS := (New TableDataSource.Create) As IDmTableDataSource;
// Determine source table
TableDS.Table := mb.ItemByID("DM_TABLE").Bind;
// Determine that data is located in table columns
TableDS.DataInColumns := True;
// Create data source which is a regular report
ReportDS := (New ReportDataSource.Create) As IDmReportDataSource;
// Get regular report
Report := mb.ItemByID("DM_REPORT_RES").Edit As IPrxReport;
Shs := Report.Sheets;
Shs.Clear;
// Create page to load results
Sheet := (Shs.Add("", PrxSheetType.Table) As IPrxTable).TabSheet;
// Determine page to which data will be loaded
ReportDS.TabSheet := Sheet;
// Determine data range
ReportDS.Range := Sheet.Cell(0, 0);
ReportDS.AddResultColumn("Category");
// Create calculation method
Method := (New DataMiningMethod.Create) As IDmMethod;
// Determine method type
Method.Kind := DmMethodKind.RandomForest;
// Set entry data source
Method.InputDataSource := TableDS;
// Set data consumer
Method.OutputDataSource := ReportDS;
// Set up calculation method parameters
RandFor := Method.Details As IDmRandomForest;
// Forest size
RandFor.ForestSize := 20;
RandFor.LearningSamplePortion := 0.6;
// Number of attributes
RandFor.NumberOfPredictors := 2;
//Forest size
RandFor.TreeSizeSpecification.MaximumNumberOfLevels := 10;
RandFor.TreeSizeSpecification.MinimumNumberOfCases := 2;
// Consider each category as an attribute
RandFor.TreeSizeSpecification.ReduceCategories := True;
// Determine column for analysis
RandFor.Target := TableDS.FieldCount - 1;
Debug.WriteLine("Column for key impact factors analysis:");
Debug.WriteLine(" - " + TableDS.Field(RandFor.Target).Name);
// Set factors impacting analysis
Attrs := New Integer[TableDS.FieldCount - 2];
Debug.WriteLine("Factors that impact analysis:");
For i := 0 To Attrs.Length - 1 Do
Attrs[i] := i + 1;
Binning := TableDS.Field(i + 1);
Debug.WriteLine(Binning.Index.ToString + ". " + Binning.Name);
Debug.Indent;
// Set parameters of the Binning procedure
If Binning.IsNumerical Then
Binning.BinningType := BinningMethod.EqualDepth;
Binning.CategoriesCount := 4;
Binning.TreatNanAsCategory := False;
Debug.WriteLine("number of non empty values: " + Binning.NonEmptyCount.ToString);
End If;
Select Case Binning.FieldType
Case DmFieldType.Date: Debug.WriteLine("data type: date");
Case DmFieldType.Integer: Debug.WriteLine("data type: integer");
Case DmFieldType.Numeric: Debug.WriteLine("data type: numeric");
Case DmFieldType.String: Debug.WriteLine("data type: string");
End Select;
Debug.WriteLine("data source: " + Binning.DataSource.Caption);
Select Case Binning.ExplanatoryType
Case DmExplanatoryType.Continuous: Debug.WriteLine("factor type: quantitative");
Case DmExplanatoryType.Ordered: Debug.WriteLine("factor type: quantitative");
Case DmExplanatoryType.Categorical: Debug.WriteLine("factor type: quantitative");
End Select;
Debug.Unindent;
End For;
RandFor.Attributes := Attrs;
// Analyze and load results
Reports := Method.Execute;
DmReport := reports.FindByType(DmReportType.Forest);
ReportDS := DmReport.Generate;
Debug.WriteLine("Report title: " + DmReport.Caption);
Debug.WriteLine("Analysis type: " + RandFor.DisplayName);
ReportDS.TabSheet.View.Selection.SelectAll;
ReportDS.TabSheet.View.Selection.Copy;
Sheet.Table.Paste;
Sheet.Columns(0, 1).AdjustWidth;
Sheet.Rows(0, 1).AdjustHeight;
Report.Sheets.Item(0).Name := ReportDS.Caption;
(Report As IMetabaseObject).Save;
// Open regular report containing analysis result
Target := WinApplication.Instance.GetObjectTarget(Report As IMetabaseObject);
Target.Execute("Object.Open", Null);
End Sub UserRandFor;
After executing the procedure, the Random Forest algorithm of the Decision Tree Ensembles data mining method will be executed for data from the DM_TABLE table using the Binning procedure for numeric data. Analysis parameters will be displayed in the console window. Analysis results will be loaded to the DM_REPORT_RES report.
See also: