CBModules customizations diff

README.txt

The CBModules.diff contains the output of `diff -r` performed between the original CBModules directory,
as downloaded from https://www.uef.fi/web/machine-learning/software (CBModules 2.0 package, dated
Jun 4, 2019), and my modified version. Since the original source code does not include licensing
information, I'm only uploading the diff, for documentation and reproducibility purposes.

The modifications consist in:

1. Disabling the data preprocessing step that scaled all dimensions individually, only allowing an
   overall uniform scaling factor. Note that all of my tests the data was shifted to start at the
   origin and to contain only positive values. See also the additional option -Z.
   
2. Changing the reporting to output the scaled sum of squared errors rather than the rescaled mean
   squared error.
   
3. Increasing the maximum population size. This was actually only applied as needed, but it does
   not seem to make any significant difference in performance anyway.

CBModules.diff

diff -r CBModules/ga/cbga.c CBModules.orig/ga/cbga.c
42c42
< #define  MaxGenerations  1000
---
> #define  MaxGenerations  100
154,155c154
<                           long        watch,
<                           double      errorscale)
---
>                           long        watch)
161,162c160,161
<   PrevError = PrintableError(PrevError * errorscale, CB);
<   NewError  = PrintableError(Error[0]  * errorscale,  CB);
---
>   PrevError = PrintableError(PrevError, CB);
>   NewError  = PrintableError(Error[0],  CB);
169,170c168,169
<     case 2:  printf("%3i: %9.9f \n", gen, NewError); break;
<     case 3:  printf("%3i: %9.9f  ", gen, NewError); PrintTime(watch); break;
---
>     case 2:  printf("%3i: %9.4f \n", gen, NewError); break;
>     case 3:  printf("%3i: %9.4f  ", gen, NewError); PrintTime(watch); break;
173c172
<                  printf("%.9f ", PrintableError(Error[i],CB));
---
>                  printf("%.4f ", PrintableError(Error[i],CB));
176c175
<     default: printf("Iter=%3i   Error=%9.9f  Change=%-9.9f ",
---
>     default: printf("Iter=%3i   Error=%9.4f  Change=%-9.4f ",
180c179
< 			   printf(" Temp=%8.9f ", SAS->CurrentTemperature);
---
> 			   printf(" Temp=%8.4f ", SAS->CurrentTemperature);
198c197
<      case 2:  printf("%9.9f  (%i)", error, iterations); break;
---
>      case 2:  printf("%9.4f  (%i)", error, iterations); break;
200c199
<                 printf("Distortion: %-9.9f\n", error);
---
>                 printf("Distortion: %-9.4f\n", error);
1821d1819
<   double        errorscale;
1827d1824
<   errorscale = Value(ErrorScale) / 1e3;
1837c1834
<   PrintProgress(&Snew[0]->CB, &SAS, 0, PrevError, Error, watch, errorscale);
---
>   PrintProgress(&Snew[0]->CB, &SAS, 0, PrevError, Error, watch);
1853c1850
<      PrintProgress(&Snew[0]->CB, &SAS, g, PrevError, Error, watch, errorscale);
---
>      PrintProgress(&Snew[0]->CB, &SAS, g, PrevError, Error, watch);
1861c1858
<   PrintResult(watch, PrintableError(Error[0] * errorscale, &Snew[0]->CB), g-1 );
---
>   PrintResult(watch, PrintableError(Error[0], &Snew[0]->CB), g-1 );
diff -r CBModules/ga/cbga.fac CBModules.orig/ga/cbga.fac
54c54
<       'Z',    1,  INT,    0,    1,    1000,      8,   YES,
---
>       'Z',    1,  INT,    0,    1,    100,      8,   YES,
170,177d169
< 
< Fact( ErrorScale,
<        "Apply this scale, divided by 1e3, to error reporting ",
<        'A',    1,  INT,    1,    1,100000000,     1000,      NO,
<        es1, es2, es3, es4, es5, es6, es7, es8, es9, es10,
<        0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
<        NO )
< 
diff -r CBModules/modules/cb.c CBModules.orig/modules/cb.c
2875c2875,2876
<   return (double) totalerror / 1e12;
---
>   return (double) totalerror /
>          (double) ( (usefreqs? TotalFreq(TS) : BookSize(TS)) * VectorSize(TS));
diff -r CBModules/modules/textfile.c CBModules.orig/modules/textfile.c
322,331d321
<     float overallscale;
<     float maxwidth = 0.0F;
<     for (j = 0; j < dim; j++) {
<         float w = MinMax[j][1];
<         if (w > maxwidth) {
<             maxwidth = w;
<         }
<     }
<     overallscale = maxval / maxwidth;
< 
336,338c326,327
<             else {
<                 scale = overallscale;
<             }
---
>             else
>                 scale = maxval / (MinMax[j][1] - MinMax[j][0]);
342c331
<                 VectorScalar(CB, i, j) = ROUND((Data[i][j]) * scale);
---
>                 VectorScalar(CB, i, j) = ROUND((Data[i][j] - MinMax[j][0]) * scale);
525,536d513
<     float overallscale;
<     float maxwidth = 0.0F;
<     int dim = VectorSize(CB);
<     int j1;
<     for (j1 = 0; j1 < dim; j1++) {
<         float w = MinMax[j1][1];
<         if (w > maxwidth) {
<             maxwidth = w;
<         }
<     }
<     overallscale = maxwidth / maxval;
< 
538,539c515,516
<         scale = overallscale;
<         val   = VectorScalar(CB, i, j)*scale;
---
>         scale = (MinMax[j][1] - MinMax[j][0]) / maxval;
>         val   = MinMax[j][0] + VectorScalar(CB, i, j)*scale;
diff -r CBModules/rs/rs.c CBModules.orig/rs/rs.c
63c63
< #define CALC_MSE(val) (double) (val) / 1e12
---
> #define CALC_MSE(val) (double) (val) / (TotalFreq(pTS) * VectorSize(pTS))