1.60.2.794

[gikonavigoeson/gikonavi.git] / GikoBayesian.pas

diff --git a/GikoBayesian.pas b/GikoBayesian.pas
index ae587a0..55791a3 100644 (file)
--- a/GikoBayesian.pas
+++ b/GikoBayesian.pas
@@ -4,7 +4,8 @@ unit GikoBayesian;
 \file          GikoBayesian.pas
 \brief \83x\83C\83W\83A\83\93\83t\83B\83\8b\83^
 
-$Id: GikoBayesian.pas,v 1.14 2004/11/01 09:51:57 yoffy Exp $
+
+$Id: GikoBayesian.pas,v 1.22 2009/01/31 15:47:15 h677 Exp $
 }
 
 //! \95½\89¼\96¼\82ð\8e«\8f\91\82É\8aÜ\82ß\82È\82¢
@@ -309,12 +310,12 @@ begin
 
                for i := 1 to sl.Count - 1 do begin
                        s := sl[ i ];
-                       name := RemoveToken( s, #1 );
+                       name := GikoBayesian.RemoveToken( s, #1 );
                        info := TWordInfo.Create;
-                       info.NormalWord                 := StrToIntDef( '$' + RemoveToken( s, #1 ), 0 );
-                       info.ImportantWord      := StrToIntDef( '$' + RemoveToken( s, #1 ), 0 );
-                       info.NormalText                 := StrToIntDef( '$' + RemoveToken( s, #1 ), 0 );
-                       info.ImportantText      := StrToIntDef( '$' + RemoveToken( s, #1 ), 0 );
+                       info.NormalWord                 := StrToIntDef( '$' + GikoBayesian.RemoveToken( s, #1 ), 0 );
+                       info.ImportantWord      := StrToIntDef( '$' + GikoBayesian.RemoveToken( s, #1 ), 0 );
+                       info.NormalText                 := StrToIntDef( '$' + GikoBayesian.RemoveToken( s, #1 ), 0 );
+                       info.ImportantText      := StrToIntDef( '$' + GikoBayesian.RemoveToken( s, #1 ), 0 );
 
                        AddObject( name, info );
                end;
@@ -416,7 +417,7 @@ var
        wKanjiDelimiter         : TStringList;
        words                                                   : TStringList;
        aWord                                                   : string;
-       countInfo                                       : TWordCountInfo;
+//     countInfo                                       : TWordCountInfo;
 
        function cutBoth( _aWord : string; _delim : TStringList ) : string;
        var
@@ -699,10 +700,15 @@ function TGikoBayesian.CalcPaulGraham( wordCount : TWordCount ) : Extended;
                        Result := 0.01
                else if info.ImportantWord + info.NormalWord * 2 < 5 then
                        Result := 0.5
-               else
-                       Result := ( info.ImportantWord / info.ImportantText ) /
-                               ((info.NormalWord * 2 / info.NormalText ) +
-                                (info.ImportantWord / info.ImportantText));
+               else begin
+                       try
+                               Result := ( info.ImportantWord / info.ImportantText ) /
+                                       ((info.NormalWord * 2 / info.NormalText ) +
+                                        (info.ImportantWord / info.ImportantText));
+                       except
+               on EZeroDivide do Result := 0.99;
+                       end;
+               end;
        end;
 
 var
@@ -734,8 +740,11 @@ begin
                        s := s * Single( narray[ i ] );
                        q := q * (1 - Single( narray[ i ] ));
                end;
-
-               Result := s / (s + q);
+               try
+                       Result := s / (s + q);
+               except
+            Result := 0.5;
+               end;
        finally
                narray.Free;
        end;
@@ -759,14 +768,23 @@ function TGikoBayesian.CalcGaryRobinson( wordCount : TWordCount ) : Extended;
                else if info.NormalWord = 0 then
                        Result := 0.99
                else
+               {
                        Result := ( info.ImportantWord / info.ImportantText ) /
                                ((info.NormalWord / info.NormalText ) +
                                 (info.ImportantWord / info.ImportantText));
+               }
+                       try
+                               Result := (info.ImportantWord * info.NormalText) /
+                                       (info.NormalWord * info.ImportantText +
+                                       info.ImportantWord * info.NormalText);
+                       except
+                               Result := 0.5;
+                       end;
        end;
 
        function f( cnt : Integer; n, mean : Single ) : Extended;
        const
-               k = 0.00001;
+               k = 0.001;
        begin
                Result := ( (k * mean) + (cnt * n) ) / (k + cnt);
        end;
@@ -777,7 +795,7 @@ var
        mean                            : Extended;
        countInfo               : TWordCountInfo;
        i                                               : Integer;
-       P1, Q1, R1      : Extended;
+       P1, Q1{, R1}    : Extended;
        cnt                                     : Extended;
 begin
 
@@ -805,10 +823,15 @@ begin
        end;
        cnt := wordCount.Count;
        if cnt = 0 then
-               cnt := 1
-       else
-       P1 := 1 - Power( P1, 1 / cnt );
-       Q1 := 1 - Power( Q1, 1 / cnt );
+               cnt := 1;
+       try
+               P1 := 1 - Power( P1, 1 / cnt );
+       except
+       end;
+       try
+               Q1 := 1 - Power( Q1, 1 / cnt );
+       except
+       end;
 
        if P1 + Q1 = 0 then begin
                Result := 0.5
@@ -838,41 +861,27 @@ function TGikoBayesian.CalcGaryRobinsonFisher(
                else if info.NormalWord = 0 then
                        Result := 0.99
                else
-                       Result := info.ImportantWord /
-                               (info.ImportantWord + info.NormalWord *
-                                info.ImportantText / info.NormalText);
+               {
+                       Result := ( info.ImportantWord / info.ImportantText ) /
+                               ((info.NormalWord / info.NormalText ) +
+                                (info.ImportantWord / info.ImportantText));
+               }
+                       Result := (info.ImportantWord * info.NormalText) /
+                               (info.NormalWord * info.ImportantText +
+                               info.ImportantWord * info.NormalText);
        end;
 
        function f( cnt : Integer; n, mean : Single ) : Extended;
        const
-               k = 0.00001;
+               k = 0.001;
        begin
                Result := ( (k * mean) + (cnt * n) ) / (k + cnt);
        end;
 
        function prbx( x2, degree : Extended ) : Extended;
-       var
-               m : Extended;
-               sum : Extended;
-               term : Extended;
-               i : extended;
        begin
 
-               m := x2 / 2;
-               sum := exp( -m );
-               term := -m;
-
-               i := 1;
-               while i < (degree / 2 - 1) do begin
-                       term := term + ln( m / i );
-                       sum := sum + exp( term );
-                       i := i + 1;
-               end;
-
-               if sum < 1 then
-                       Result := sum
-               else
-                       Result := 1.0;
+               Result := 0.5;
 
        end;
 
@@ -882,8 +891,8 @@ var
        mean                            : Extended;
        countInfo               : TWordCountInfo;
        i                                               : Integer;
-       normal                  : Extended;
-       important               : Extended;
+//     normal                  : Extended;
+//     important               : Extended;
        P1, Q1                  : Extended;
        cnt                                     : Extended;
 begin
@@ -902,41 +911,30 @@ begin
        end;
        mean := mean / wordCount.Count;
 
-       cnt := 0;
-(*
        P1 := 1;
        Q1 := 1;
-(*)
-       P1 := 0;
-       Q1 := 0;
-//*
        for i := 0 to wordCount.Count - 1 do begin
                countInfo       := TWordCountInfo( wordCount.Objects[ i ] );
                n                                               := f( countInfo.WordCount, narray[ i ], mean );
-               if countInfo <> nil then
-                       cnt := cnt + countInfo.WordCount;
-(*
-               P1 := P1 + Ln( 1 - n ) * countInfo.WordCount;
-               Q1 := Q1 + Ln( n ) * countInfo.WordCount;
-(*)
-               P1 := P1 + Ln( 1 - n );
-               Q1 := Q1 + Ln( n );
-//*)
+               P1 := P1 * ( 1 - n );
+               Q1 := Q1 * n;
        end;
+       cnt := wordCount.Count;
        if cnt = 0 then
                cnt := 1;
-//(*
-       P1 := prbx( -2 * P1, 2 * cnt );
-       Q1 := prbx( -2 * Q1, 2 * cnt );
-(*)
-       P1 := prbx( -2 * Ln( P1 ), 2 * cnt );
-       Q1 := prbx( -2 * Ln( Q1 ), 2 * cnt );
-//*)
-       if P1 + Q1 = 0 then begin
-               Result := 0.5
-       end else begin
-               Result := (1 + Q1 + P1) / 2;
+       try
+               P1 := Power( P1, 1 / cnt );
+       except
        end;
+       try
+               Q1 := Power( Q1, 1 / cnt );
+       except
+       end;
+
+       P1 := 1 - prbx( -2 * Ln( P1 ), 2 * cnt );
+       Q1 := 1 - prbx( -2 * Ln( Q1 ), 2 * cnt );
+
+       Result := (1 + P1 - Q1) / 2;
 
 end;