樹狀結構

TREE 可在 預測及決策樹狀結構中使用。

TREE 程序會建立樹狀結構型模型。 它會根據預測變數的值，將觀察值分類為群組或預測應變數的值。 這個程序會提供用於解釋與確認分類分析的驗證工具。

附註: TREE 語法圖中使用的方括弧是語法的必要部分，不會用來指出選用元素。 語法圖中使用的等號 (=) 是必要元素。 所有次指令都是選用的。

TREE dependent_variable [level]
     BY variable [level] variable [level]...
     FORCE = variable [level]

  /TREE DISPLAY ={TOPDOWN**  } NODES = {STATISTICS**}
                 {LEFTTORIGHT}         {CHART       }
                 {RIGHTTOLEFT}         {BOTH        }
                 {NONE       }
        BRANCHSTATISTICS ={YES**} NODEDEFS = {YES**}
                          {NO   }            {NO   }
        SCALE={AUTO**       }
              {percent value}
                       
  /DEPCATEGORIES USEVALUES=[VALID** value, value...MISSING]
                 TARGET=[value value...]

  /PRINT MODELSUMMARY** CLASSIFICATION** RISK** CPS** IMPORTANCE SURROGATES
         TREETABLE CATEGORYSPECS NONE

  /GAIN SUMMARYTABLE = {YES**} CATEGORYTABLE = {YES**}
                       {NO   }                 {NO   }  
        TYPE = [NODE** PTILE] SORT = {DESCENDING**}
                                     {ASCENDING   }
        INCREMENT = {10** } CUMULATIVE = {YES**}
                    {value}              {NO   }
               
  /PLOT GAIN RESPONSE INDEX MEAN PROFIT ROI IMPORTANCE INCREMENT = {10   }
                                                                   {value}

  /RULES NODES = {TERMINAL**     } SYNTAX = {INTERNAL** }
                 {ALL            }          {SQL        }
                 {TOPN(value)    }          {GENERIC    }
                 {TOPPCT(value)  }
                 {MININDEX(value)}
         TYPE = {SCORING**} SURROGATES = {EXCLUDE**} LABELS = {YES**}
                {SELECTION}              {INCLUDE  }          {NO   }
         OUTFILE = ’filespec’

  /SAVE NODEID(varname) PREDVAL(varname) PREDPROB(rootname) ASSIGNMENT(varname)  
  
 /METHOD TYPE = {CHAID**        }
                {EXHAUSTIVECHAID}
                {CRT            }
                {QUEST          }
         MAXSURROGATES = {AUTO**} PRUNE = {NONE**     } 
                         {value }         {SE({1    })}
                                              {value}

 /GROWTHLIMIT MAXDEPTH = {AUTO**} MINPARENTSIZE = {100**}
                         {value }                 {value}
                             
              MINCHILDSIZE = {50** } 
                             {value}                  
                             
 /VALIDATION  TYPE = {NONE**                  } OUTPUT = {BOTHSAMPLES}
                     {SPLITSAMPLE({50       })}          {TESTSAMPLE }
                                  {percent  }
                                  {varname}
                     {CROSSVALIDATION({10   })}
                                      {value}   
             
 /CHAID  ALPHASPLIT = {.05**}  ALPHAMERGE = {.05**}   
                      {value}               {value}
         SPLITMERGED = {NO**} CHISQUARE = {PEARSON**}     
                       {YES }             {LR       }
         CONVERGE = {.001**}  MAXITERATIONS = {100**}
                    {value }                  {value}
         ADJUST = {BONFERRONI**}
                  {NONE        }
         INTERVALS = {10**                           }
                     {value                          }
                     {varlist(value) varlist(value) …}

  /CRT IMPURITY = {GINI**       } MINIMPROVEMENT = {.0001**}
                  {TWOING       }                  {value  }
                  {ORDEREDTWOING}

  /QUEST ALPHASPLIT = {.05**} 
                      {value} 

 /COSTS {EQUAL**                                                     }
        {CUSTOM = actcat, predcat [value] actcat, predcat [value] ...}

 /PRIORS {FROMDATA**                          } ADJUST = {NO**}
         {EQUAL                               }          {YES }
         {CUSTOM = cat [value] cat [value] ...}
               
 /SCORES {EQUALINCREMENTS**                   }
         {CUSTOM = cat [value] cat [value] ...}

 /PROFITS CUSTOM = cat [revenue, expense] cat [revenue, expense] ...

 /INFLUENCE varname

 /OUTFILE TRAININGMODEL = ’filespec’ TESTMODEL = ’filespec’      

 /MISSING  NOMINALMISSING = {MISSING**}
                            {VALID    }
 
 /TARGETRESPONSE RATE = {NONE   }
                        {value  }

** 如果省略次指令，則為預設值。

此指令會讀取作用中資料集，並導致執行任何擱置指令。 如需相關資訊，請參閱主題 指令順序 。

可以從「 決策樹狀結構 」對話框產生 TREE 指令的語法。

發行歷程

版本 13.0

• 已建立指令。

版本 18.0

• 已建立 TARGETRESPONSE 次指令。

範例

TREE risk BY income age creditscore employment.